{
 "name": "Agent Sightings",
 "count": 1081,
 "generated": "2026-10-29",
 "license": "Summaries and prompts by Agent Sightings; follow each case's source URL for the original.",
 "cases": [
  {
   "id": "negotiate-a-car-purchase-between-dealers",
   "title": "Negotiate a car purchase between dealers",
   "summary": "Case study takeaway: AJ Stuyvenberg ran OpenClaw with access to his email and browser. It researched Reddit prices, searched dealer inventory within 50 miles of Boston, and filled dealer contact forms with his real details. Reddit research put typical Massachusetts prices near $58,000.",
   "agent": "OpenClaw",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://cloudship.co.uk/blog/ai-agent-bought-a-car/",
   "prompt": "Shopping task: negotiate a car purchase between dealers. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 100,
   "page": "https://agentsightings.com/case/negotiate-a-car-purchase-between-dealers.html"
  },
  {
   "id": "win-a-flight-refund-dispute",
   "title": "Win a flight refund dispute",
   "summary": "Case study takeaway: through ChatGPT, a person who had booked an IndiGo flight that had been cancelled has claimed that they were able to win a refund of 43,992.",
   "agent": "ChatGPT",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1h9gwdn",
   "prompt": "Plan: win a flight refund dispute. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 99,
   "page": "https://agentsightings.com/case/win-a-flight-refund-dispute.html"
  },
  {
   "id": "set-up-nightly-smart-home-routines-by-voice",
   "title": "Set up nightly smart-home routines by voice",
   "summary": "One practitioner notes that the writer, a Tom's Guide reviewer and parent, had upgraded to Alexa+. She spoke one sentence to Alexa+ asking it to dim the lights and lower the thermostat every night at 9 p.m. She says the automation was stood up instantly and saved her a lot of time over the old routine setup.",
   "agent": "Alexa+",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/alexa/comments/1lqgfr2/a_contrary_opinion_alexa_is_very_good",
   "prompt": "Home/family task: set up nightly smart-home routines by voice. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 98,
   "page": "https://agentsightings.com/case/set-up-nightly-smart-home-routines-by-voice.html"
  },
  {
   "id": "negotiate-a-lower-fiber-bill-with-at-amp-t-chat",
   "title": "Negotiate a lower fiber bill with AT&T chat",
   "summary": "Case study takeaway: despite retaining the bundle discount and no contract, Meta Muse managed to negotiate with AT&T support chat to reduce the price of a fiber plan from $85.94 to $60/mo.",
   "agent": "Meta Muse",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/dummyLC/status/2106889384377339973",
   "prompt": "Finance task: negotiate a lower fiber bill with AT&T chat. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 97,
   "page": "https://agentsightings.com/case/negotiate-a-lower-fiber-bill-with-at-amp-t-chat.html"
  },
  {
   "id": "rebook-a-work-flight-to-save-money",
   "title": "Rebook a work flight to save money",
   "summary": "A published walkthrough shows arden monitored Someone's booked Zurich work flight. It sent a text proposing a rebook that would save $7,750 and asking for approval. The rebooking went through and the $7,750 refund appeared on the card.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/tonybruess/status/2052024422463107232",
   "prompt": "Plan: rebook a work flight to save money. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 96,
   "page": "https://agentsightings.com/case/rebook-a-work-flight-to-save-money.html"
  },
  {
   "id": "order-family-groceries-from-fridge-contents",
   "title": "Order family groceries from fridge contents",
   "summary": "According to the write-up, the parent says Grok Bot is connected to a Samsung fridge, where it can view two camera-visible shelves, and they supplemented that by feeding it a typical grocery list assembled from scanned past receipts. Twice a week the agent checks what is likely needed, adds routine items, hunts discounts and promo codes, then places the order and times pickup for set days at one of two stores on the drive home; the human mainly set the rules and gets a morning message to combine pickups. They say it saves substantial time, money, and energy, and in Norway they also have it push the grocery info into Oda while making sure the basket clears the roughly $80 free-delivery threshold.",
   "agent": "Grok Bot",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1x2323i/grok_bot_in_personal_life",
   "prompt": "Home/family task: order family groceries from fridge contents. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 95,
   "page": "https://agentsightings.com/case/order-family-groceries-from-fridge-contents.html"
  },
  {
   "id": "negotiate-bills-handle-returns-and-cancel-subscriptions",
   "title": "Negotiate bills, handle returns and cancel subscriptions",
   "summary": "Muse challenge lore from Meta folks: AT&T bill talked down, Amazon returns queued, IKEA furniture sent back, vet claim filed, dusty book-club sub killed with refund. One thread’s running total: ~$2,121—treat as anecdote, not audited.",
   "agent": "Meta Muse",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49914631",
   "prompt": "Review my bills, subscriptions, and open returns. Draft negotiation or cancellation scripts and return steps; pause before any call, chat submit, or confirmation that spends or cancels.",
   "added": "2026-10-15",
   "interest": 94,
   "page": "https://agentsightings.com/case/negotiate-bills-handle-returns-and-cancel-subscriptions.html"
  },
  {
   "id": "cut-a-hospital-bill-using-claude-x27-s-medicare-price-analysis",
   "title": "Cut a hospital bill using Claude's Medicare price analysis",
   "summary": "From the linked report, claude shows up in a write-up about cut a hospital bill using Claude's Medicare price analysis, with figures like $195,628, $163,000. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://www.aol.com/articles/used-claude-negotiate-163-000-100601051.html",
   "prompt": "Finance task: cut a hospital bill using Claude's Medicare price analysis. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 93,
   "page": "https://agentsightings.com/case/cut-a-hospital-bill-using-claude-x27-s-medicare-price-analysis.html"
  },
  {
   "id": "book-the-best-priced-flight-and-later-refund-it",
   "title": "Book the best-priced flight and later refund it",
   "summary": "A published walkthrough shows someone asked Claude Code to find the best price, and it booked the flight on its own three months ahead. Separately, @bot obtained the refund by talking to the airline for 24 hours. The ticket was successfully refunded.",
   "agent": "Claude Code",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/BansalRahul14/status/2109912274190401545",
   "prompt": "Plan: book the best-priced flight and later refund it. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 92,
   "page": "https://agentsightings.com/case/book-the-best-priced-flight-and-later-refund-it.html"
  },
  {
   "id": "family-menu-to-doorstep-delivery",
   "title": "Family menu to doorstep delivery",
   "summary": "An OpenClaw bot plans a healthy week of family meals, turns it into a shopping list and places the supermarket order.",
   "agent": "OpenClaw",
   "topic": "Home & Family",
   "source": "LinkedIn",
   "url": "https://www.linkedin.com/posts/katherine-baranova-mba-3b794a39_albertheijn-activity-7435312007886770178-COIP",
   "prompt": "Plan a healthy 7-day dinner menu for a family of four, convert it into a grocery list grouped by aisle, and prepare the order for my approval.",
   "added": "2026-10-15",
   "interest": 91,
   "page": "https://agentsightings.com/case/family-menu-to-doorstep-delivery.html"
  },
  {
   "id": "travelling-abroad-with-a-dog",
   "title": "Travelling abroad with a dog",
   "summary": "A Manus user shared a plan for a family trip with a pet, covering pet travel rules and dog-friendly stops.",
   "agent": "Manus",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ManusOfficial/comments/1k0egpa/a_family_trip_with_my_dog/",
   "prompt": "Plan a family road trip from <A> to <B> with our dog. Include pet entry requirements, dog-friendly hotels, and rest stops every 2-3 hours.",
   "added": "2026-10-15",
   "interest": 90,
   "page": "https://agentsightings.com/case/travelling-abroad-with-a-dog.html"
  },
  {
   "id": "get-a-billing-dispute-refunded-by-a-support-agent",
   "title": "Get a billing dispute refunded by a support agent",
   "summary": "Case study takeaway: the company's AI agent produced and issued a refund to a customer who had a dispute with an incorrect Vercel bill.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49656053",
   "prompt": "Finance task: get a billing dispute refunded by a support agent. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 89,
   "page": "https://agentsightings.com/case/get-a-billing-dispute-refunded-by-a-support-agent.html"
  },
  {
   "id": "cancel-an-amazon-subscription-and-get-a-refund",
   "title": "Cancel an Amazon subscription and get a refund",
   "summary": "From the linked report, someone asked Codex to obtain a refund for an Amazon subscription. Codex navigated the account settings to cancel it, then moved to the help page and opened a live chat with an associate. The associate approved the refund.",
   "agent": "Codex",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/jxnlco/status/2066970432855581052",
   "prompt": "Shopping task: cancel an Amazon subscription and get a refund. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 88,
   "page": "https://agentsightings.com/case/cancel-an-amazon-subscription-and-get-a-refund.html"
  },
  {
   "id": "build-and-submit-a-family-quest-app-to-the-app-store",
   "title": "Build and submit a family quest app to the App Store",
   "summary": "Observed in the wild: someone, a mother of three with no prior coding experience, partnered with Replit Agent to develop Kogoto Quest from initial idea through full implementation, RevenueCat integration for premium AI features, and App Store submission prep. The agent performed the coding and setup steps while Someone provided direction and oversight throughout. The app was submitted for App Store review.",
   "agent": "Replit Agent",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/Fairy_Yoshizawa/status/2102161527285702906",
   "prompt": "Home/family task: build and submit a family quest app to the App Store. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 87,
   "page": "https://agentsightings.com/case/build-and-submit-a-family-quest-app-to-the-app-store.html"
  },
  {
   "id": "draft-a-refund-demand-for-an-unauthorized-renewal",
   "title": "Draft a refund demand for an unauthorized renewal",
   "summary": "Linked example of draft a refund demand for an unauthorized renewal. After Skillshare charged a $24 annual renewal the poster says they had not authorized, they used ChatGPT to help draft firmer refund emails instead of accepting the company’s offered 12-month gift card.",
   "agent": "ChatGPT",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1nw7jby/how_chatgpt_helped_me_to_get_a_refund_from",
   "prompt": "Shopping task: draft a refund demand for an unauthorized renewal. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 86,
   "page": "https://agentsightings.com/case/draft-a-refund-demand-for-an-unauthorized-renewal.html"
  },
  {
   "id": "rebook-after-a-flight-cancellation",
   "title": "Rebook after a flight cancellation",
   "summary": "Real-world thread: after the cancellation Someone activated ChatGPT Work mode and directed it to locate replacement flights, arrange the rental car, and consolidate the updated itinerary details into one location. No further access or hand-offs are described. The task took about 30 minutes instead of over an hour and made the situation simpler, though it remained a pain.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/cherryplaceai/status/2104735783097909416",
   "prompt": "Plan: rebook after a flight cancellation. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 85,
   "page": "https://agentsightings.com/case/rebook-after-a-flight-cancellation.html"
  },
  {
   "id": "cancel-and-refund-subscriptions-by-searching-email-and-calling",
   "title": "Cancel and refund subscriptions by searching email and calling",
   "summary": "Case study takeaway: grok Bot shows up in a write-up about cancel and refund subscriptions by searching email and calling. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/morganlinton/status/2092016807943250150",
   "prompt": "Go through my mail for this goal: cancel and refund subscriptions by searching email and calling. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 84,
   "page": "https://agentsightings.com/case/cancel-and-refund-subscriptions-by-searching-email-and-calling.html"
  },
  {
   "id": "fill-out-insurance-paperwork-from-hospital-records",
   "title": "Fill out insurance paperwork from hospital records",
   "summary": "Newborn feeding in one arm, Claude Cowork in the other: Felix Rieseberg dropped hospital PDFs on the agent, pointed it at the insurer portal, and let it complete the forms from the chart details.",
   "agent": "Claude Cowork",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/felixrieseberg/status/2014490352624005458",
   "prompt": "Finance task: fill out insurance paperwork from hospital records. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 83,
   "page": "https://agentsightings.com/case/fill-out-insurance-paperwork-from-hospital-records.html"
  },
  {
   "id": "diagnose-a-check-engine-light-by-ordering-an-obd2-scanner",
   "title": "Diagnose a check engine light by ordering an OBD2 scanner",
   "summary": "Dash light on, OpenClaw on the case: Amazon OBD2 scanner ordered from the car, codes read, sensor blamed, light cleared. DIY path beat a dealer visit on time and cash.",
   "agent": "OpenClaw",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/bsniz/status/2108712125912814057",
   "prompt": "Shopping task: diagnose a check engine light by ordering an OBD2 scanner. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 82,
   "page": "https://agentsightings.com/case/diagnose-a-check-engine-light-by-ordering-an-obd2-scanner.html"
  },
  {
   "id": "coordinate-a-cancer-patient-x27-s-records-and-care",
   "title": "Coordinate a cancer patient's records and care",
   "summary": "Case study takeaway: a family gave a dying relative's medical records to an agent as coordinator; it flagged a wrong pain medication schedule and a transcription error in his genetic markers, which changed his treatment plan.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50100601",
   "prompt": "Home/family task: coordinate a cancer patient's records and care. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 81,
   "page": "https://agentsightings.com/case/coordinate-a-cancer-patient-x27-s-records-and-care.html"
  },
  {
   "id": "find-forgotten-money-and-negotiate-insurance-refunds",
   "title": "Find forgotten money and negotiate insurance refunds",
   "summary": "From the linked report, eric Allen asked Meta's Muse agent to search state government databases for unclaimed money, and it ran the scan for his wife too. Separately, Frank Chien had Muse phone his insurer, Amtrust, to cancel a workers comp policy and get a refund. Allen's scan took six minutes and located about $730 in store credit, a $272 refund and about $800 for his wife.",
   "agent": "Meta Muse",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://www.forbes.com/sites/rashishrivastava/2026/09/25/ai-agents-are-helping-people-recover-money-by-negotiating-bills-and-chasing-refunds/",
   "prompt": "Finance task: find forgotten money and negotiate insurance refunds. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 80,
   "page": "https://agentsightings.com/case/find-forgotten-money-and-negotiate-insurance-refunds.html"
  },
  {
   "id": "rebook-a-delayed-flight-connection-mid-flight",
   "title": "Rebook a delayed flight connection mid-flight",
   "summary": "According to the write-up, denis gave Dot from OpenAI access to its own computer. While he was still on the delayed Alaska flight, the agent simultaneously messaged Alaska, Delta, and KLM, located a new routing, and requested the free rebooking. The agent secured the rebooking and avoided an while people sleep stay in Seattle.",
   "agent": "Dots",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/literallydenis/status/2106277505921622426",
   "prompt": "Plan: rebook a delayed flight connection mid-flight. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 79,
   "page": "https://agentsightings.com/case/rebook-a-delayed-flight-connection-mid-flight.html"
  },
  {
   "id": "run-your-side-of-a-home-insurance-claim",
   "title": "Run your side of a home insurance claim",
   "summary": "Two months after a burst pipe, Claude reads the policy, files ALE receipts, watches adjuster mail for delays and contradictions, logs estimates and invoices, and argues coverage against the policy text—nudging toward a state insurance complaint when delays pile up.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49536267",
   "prompt": "Manage my side of this insurance claim: index the policy and receipts, track adjuster messages, flag delays versus promises, and draft challenge letters. Do not file complaints or send mail until I approve.",
   "added": "2026-10-15",
   "interest": 78,
   "page": "https://agentsightings.com/case/run-your-side-of-a-home-insurance-claim.html"
  },
  {
   "id": "build-a-diy-slide-tower-with-a-subagent",
   "title": "Build a DIY slide tower with a subagent",
   "summary": "A parent spun up a Claude Code subagent to help design and document building a backyard slide tower for their child.",
   "agent": "Claude Code",
   "topic": "Home & Family",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Act as a DIY project planner. Given these materials and a kids' slide goal, produce a cut list, step-by-step build plan, and safety checklist. Ask before suggesting power-tool steps.",
   "added": "2026-10-15",
   "interest": 77,
   "page": "https://agentsightings.com/case/build-a-diy-slide-tower-with-a-subagent.html"
  },
  {
   "id": "plan-and-book-a-short-notice-work-trip",
   "title": "Plan and book a short-notice work trip",
   "summary": "According to the write-up, user handed short-notice Munich work trip planning to Grok Bot, giving it the destination and constraints of airport/Mynaric access without car rental. The agent researched flights, suggested departing a day early with options; after Someone chose, it initiated booking (user had to intervene repeatedly to enter details). Hotel proved convenient with restaurant and grocery nearby; $15 train to site and $5 back for sightseeing worked without issues.",
   "agent": "Grok Bot",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/vinceisbullish/status/2105352089689858221",
   "prompt": "Plan: plan and book a short-notice work trip. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 76,
   "page": "https://agentsightings.com/case/plan-and-book-a-short-notice-work-trip.html"
  },
  {
   "id": "find-cheaper-car-insurance",
   "title": "Find cheaper car insurance",
   "summary": "Case study takeaway: startup founder Joseph Devoy Leaned on Meta's Muse agent, launched September 8, for car insurance. Meta's pitch is to prompt Muse to review your finances, then have it carry out savings tasks in the background, such as haggling with insurers by phone. Devoy said the insurance Muse located cost $3,500 less per year than his original plan, and he called it a killer product.",
   "agent": "Meta Muse",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/c00v/status/2108340895582204282",
   "prompt": "Shopping task: find cheaper car insurance. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 75,
   "page": "https://agentsightings.com/case/find-cheaper-car-insurance.html"
  },
  {
   "id": "do-bookkeeping-and-tax-prep-from-a-folder-of-documents",
   "title": "Do bookkeeping and tax prep from a folder of documents",
   "summary": "Community report: cowork was pointed at a Taxes folder holding W-2s, 1099s, bank CSVs, mortgage docs and prior-year returns. It sorted files into year/category subfolders, assembled Context.md, stood up a master spreadsheet with income/expense/write-off tabs, produced tax summaries with refund estimates and flagged deductions, generated an accountant briefing, and listed discrepancies plus a checklist. Delivered a complete organized workspace and briefing that replaced manual sorting, saving 6-8 hours with cleaner output than previous accountant handoffs.",
   "agent": "Claude Cowork",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/JJEnglert/status/2041901304725774805",
   "prompt": "Finance task: do bookkeeping and tax prep from a folder of documents. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 74,
   "page": "https://agentsightings.com/case/do-bookkeeping-and-tax-prep-from-a-folder-of-documents.html"
  },
  {
   "id": "chase-an-airline-refund-dispute",
   "title": "Chase an airline refund dispute",
   "summary": "Observed in the wild: user handed the full $1,200 flight refund dispute to the Instinct agent after two months of airline silence and stall tactics. The agent sent automated firm follow-ups every 48 hours to keep the thread alive, logged every broken promise and missed callback into a clean paper trail, and prepared a formal grievance letter to the aviation regulator. The airline processed the full refund within one week.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/ivysana03/status/2104538178711695529",
   "prompt": "Shopping task: chase an airline refund dispute. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 73,
   "page": "https://agentsightings.com/case/chase-an-airline-refund-dispute.html"
  },
  {
   "id": "plan-family-vacations",
   "title": "Plan family vacations",
   "summary": "One practitioner notes that utilizes ChatGPT Deep Research to search for destinations, attractions, and trips with detailed family instructions, but is incapable of filtering and reserving lodging.",
   "agent": "ChatGPT",
   "topic": "Home & Family",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48458177",
   "prompt": "Home/family task: plan family vacations. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 72,
   "page": "https://agentsightings.com/case/plan-family-vacations.html"
  },
  {
   "id": "get-status-match-and-upgrade-certificates-then-book-flight",
   "title": "Get status match and upgrade certificates, then book flight",
   "summary": "Four lie-flat upgrade certs plus a status match landed in John’s inbox because his Dot chased the forms, watched for confirmation mail, and read the terms before buying the ticket—something he admits he would have ignored.",
   "agent": "Dots",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/johnjjung/status/2106664764294512686",
   "prompt": "Plan: get status match and upgrade certificates, then book flight. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 71,
   "page": "https://agentsightings.com/case/get-status-match-and-upgrade-certificates-then-book-flight.html"
  },
  {
   "id": "work-queue-that-runs-while-you-re-offline",
   "title": "Work queue that runs while you're offline",
   "summary": "A self-hosted agent with memory files and cron jobs kept working during a flight, building pages and drafting posts, then sent a chat summary.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47213594",
   "prompt": "Here is my job description and current project list. While I'm away, work through the top items, save everything you produce to the project folder, and message me a summary of what you did and what needs my decision.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/work-queue-that-runs-while-you-re-offline.html"
  },
  {
   "id": "weekly-meal-plans-with-grocery-carts",
   "title": "Weekly meal plans with grocery carts",
   "summary": "A meal-planning product generates preference-based weekly plans and pushes ingredients into grocery carts.",
   "agent": "Meals You Love",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=45724229",
   "prompt": "Create a weekly vegetarian meal plan under 30 minutes per dinner, reusing ingredients to minimize waste, plus a consolidated shopping list.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/weekly-meal-plans-with-grocery-carts.html"
  },
  {
   "id": "watch-weekly-deals-and-buy-regular-items",
   "title": "Watch weekly deals and buy regular items",
   "summary": "Observed in the wild: the setup used two separate bots: Deals checked weekly for discounts on repeat purchases like energy drinks and cat supplies, then sent approved locates to Buyer. Buyer used Grok Bot’s Chrome browser plus a custom skill containing the shipping address and instructions to hunt for recent discount codes, complete checkout, pull one-time passcodes from iMessages when Stripe Link needed them, and then watch for order confirmations and tracking updates. The poster says Grok Bot is finding deals on consumables they always need and buying them without their involvement.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1vwkzf1/heres_how_grok_bot_is_doing_all_the_stuff_i_hate",
   "prompt": "Help me watch weekly deals and buy regular items. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/watch-weekly-deals-and-buy-regular-items.html"
  },
  {
   "id": "try-to-book-a-warsaw-to-london-trip",
   "title": "Try to book a Warsaw to London trip",
   "summary": "Observed in the wild: claude shows up in a write-up about try to book a Warsaw to London trip. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1wg6ubl/i_asked_claude_to_book_my_flight_hotel_it_took_6",
   "prompt": "Plan: try to book a Warsaw to London trip. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/try-to-book-a-warsaw-to-london-trip.html"
  },
  {
   "id": "triage-care-estimate-visit-cost-and-negotiate-medical-bills",
   "title": "Triage care, estimate visit cost, and negotiate medical bills",
   "summary": "Observed in the wild: the source is a brief comment from the builder describing CareRoute. The app triages a person to the right site of care. The source gives no results, savings figures, or user numbers.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=45873030",
   "prompt": "Help me triage care, estimate visit cost, and negotiate medical bills. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/triage-care-estimate-visit-cost-and-negotiate-medical-bills.html"
  },
  {
   "id": "triage-car-repair-requests-and-reply-to-customers",
   "title": "Triage car repair requests and reply to customers",
   "summary": "A published walkthrough shows an n8n workflow receives car repair requests from a web form. The AI classifies each by job type, urgency, and summary, afterward creates an Airtable record and sends an instant customer reply. The Airtable base displays job status and timestamps for processed requests.",
   "agent": "n8n",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/ho_lar_s/status/2109790708970438953",
   "prompt": "Shopping task: triage car repair requests and reply to customers. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/triage-car-repair-requests-and-reply-to-customers.html"
  },
  {
   "id": "track-subscriptions-and-renewals-from-gmail",
   "title": "Track subscriptions and renewals from Gmail",
   "summary": "A published walkthrough shows the writer switched on Agent Mode in ChatGPT and gave it access to her Gmail. It searched the inbox for terms like renewal, subscription and auto-renew, or for service names. She says it helped her cancel unused streaming services and saved her family money on gas.",
   "agent": "ChatGPT Agent",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.tomsguide.com/ai/5-clever-ways-i-use-chatgpt-agent-to-save-serious-money-each-month",
   "prompt": "Go through my mail for this goal: track subscriptions and renewals from Gmail. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/track-subscriptions-and-renewals-from-gmail.html"
  },
  {
   "id": "track-pantry-stock-from-receipts-and-buy-groceries",
   "title": "Track pantry stock from receipts and buy groceries",
   "summary": "Case study takeaway: user ingests receipts and spending data into a custom service, then exposes a grocery budget envelope plus recipe history and financial data to a stock estimator tool. A second tool connects to local grocery store APIs for nutrition, availability, and prices. The system genuinely saves Someone time and money through ongoing stock estimates and recommendations.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/ZacariahHeim/status/2107123309095596211",
   "prompt": "Shopping task: track pantry stock from receipts and buy groceries. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/track-pantry-stock-from-receipts-and-buy-groceries.html"
  },
  {
   "id": "track-las-vegas-flight-prices-and-find-return-flights",
   "title": "Track Las Vegas flight prices and find return flights",
   "summary": "Real-world thread: for the outbound JSX flight, Muse navigated the booking site, read the price and set up a daily check that notifies on changes. For the return, The builder gave a date window and later said he had Amex points. The outbound tracking worked.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://theautomatedoperator.substack.com/p/metas-muse-is-surprisingly-unimpressive",
   "prompt": "Plan: track Las Vegas flight prices and find return flights. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/track-las-vegas-flight-prices-and-find-return-flights.html"
  },
  {
   "id": "sync-family-calendars-by-generating-and-emailing-invites",
   "title": "Sync family calendars by generating and emailing invites",
   "summary": "Linked example of sync family calendars by generating and emailing invites. The thread describes two practical setups rather than a full assembled sync: one person has the bot generate .ics calendar invites and send them by email, then they manually accept them into the calendar.",
   "agent": "Grok Bot",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1wltlzm/iphone_calendar_access_for_a_bot",
   "prompt": "Home/family task: sync family calendars by generating and emailing invites. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/sync-family-calendars-by-generating-and-emailing-invites.html"
  },
  {
   "id": "source-suppliers-and-negotiate-deals-overnight",
   "title": "Source suppliers and negotiate deals overnight",
   "summary": "Linked example of source suppliers and negotiate deals overnight. Used Sourcing Expert agent plus Accio plugin connected to Alibaba.com to source a 15W MagSafe car charger.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Aria_Nawi/status/2079537891206955214",
   "prompt": "Help me source suppliers and negotiate deals overnight. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/source-suppliers-and-negotiate-deals-overnight.html"
  },
  {
   "id": "shortlist-business-trip-hotels-into-a-google-sheet",
   "title": "Shortlist business-trip hotels into a Google Sheet",
   "summary": "According to the write-up, you create a /hotel-picker shortcut in Claude in Chrome that starts from Booking.com. It applies the Hotels-only, distance and 4-star-plus filters, then checks Booking.com and TripAdvisor reviews. The source says results come back in about 10–15 minutes as a top-10 list sorted by value.",
   "agent": "Claude",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://app.therundown.ai/guides/use-claudes-chrome-extension-to-quickly-book-business-trips",
   "prompt": "Plan: shortlist business-trip hotels into a Google Sheet. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/shortlist-business-trip-hotels-into-a-google-sheet.html"
  },
  {
   "id": "shared-grocery-list-from-the-family-chat",
   "title": "Shared grocery list from the family chat",
   "summary": "Family members message an OpenClaw agent to add items, and it maintains a single aisle-sorted shared list.",
   "agent": "OpenClaw",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenClawUseCases/comments/1sel7ad/my_family_now_has_a_magical_shared_grocery_list/",
   "prompt": "When anyone in the family chat mentions something we need, add it to the shared list, merge duplicates and keep it sorted by store aisle.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/shared-grocery-list-from-the-family-chat.html"
  },
  {
   "id": "run-pet-relocation-airbnb-pricing-and-disputes",
   "title": "Run pet relocation, Airbnb pricing, and disputes",
   "summary": "Observed in the wild: Custom agent was pointed at run pet relocation, Airbnb pricing, and disputes. has the task of managing pet import/export paperwork, providing weekly reports on Airbnb's pricing, resolving a commercial dispute, and submitting documents in Chrome been delegated to agents.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50053301",
   "prompt": "Plan: run pet relocation, Airbnb pricing, and disputes. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/run-pet-relocation-airbnb-pricing-and-disputes.html"
  },
  {
   "id": "run-personal-errands-like-flowers-and-date-nights",
   "title": "Run personal errands like flowers and date nights",
   "summary": "Documented elsewhere: marshall received early access to Meta's Muse agent. The agent paid his bills, bought flowers, and booked date nights with no further details given on access, steps, or human intervention. He located it genuinely impressive and said it just works, calling it what OpenClaw should have been.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/marshal/status/2097414842844151842",
   "prompt": "Plan: run personal errands like flowers and date nights. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/run-personal-errands-like-flowers-and-date-nights.html"
  },
  {
   "id": "run-grocery-email-and-photo-tasks-through-comet-browser",
   "title": "Run grocery, email and photo tasks through Comet browser",
   "summary": "Case study takeaway: julien Rio paid $20 a month for Comet Pro and gave its assistant logged-in access to Intermarché, his 15-year-old inbox, YouTube Studio, Dropbox and Google Calendar. He asked it to analyse purchases, clean email, sort 8,000 photos into daily folders, and turn a vacation spreadsheet into events. It couldn't read Intermarché's PDF receipts and located only a handful of emails.",
   "agent": "Perplexity Comet",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.julienrio.com/marketing/english/tested-comet-perplexity-AI-powered-browser.html",
   "prompt": "Go through my mail for this goal: run grocery, email and photo tasks through Comet browser. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/run-grocery-email-and-photo-tasks-through-comet-browser.html"
  },
  {
   "id": "run-a-homeschooling-family-on-openclaw-agents",
   "title": "Run a homeschooling family on OpenClaw agents",
   "summary": "Documented elsewhere: jesse Genet runs five OpenClaw agents, each on its own Mac Mini with limited permissions. She photographed curriculum books and sent them to Sylvie, the homeschool agent. Cole assembled Mira in four days and it was deployed to a Google TV Streamer.",
   "agent": "OpenClaw",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/cryptopunk7213/status/2043892812815708257",
   "prompt": "Home/family task: run a homeschooling family on OpenClaw agents. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/run-a-homeschooling-family-on-openclaw-agents.html"
  },
  {
   "id": "run-a-business-assistant-that-resumes-project-work",
   "title": "Run a business assistant that resumes project work",
   "summary": "Documented elsewhere: the poster used a ChatGPT business account tied to OpenClaw on an old Intel MacBook they had turned into a server, after following setup steps involving Homebrew, Node, Docker, n8n, Tailscale, plugins, device pairing, approvals, and Codex supervision. The goal was for it to retain project context, access Microsoft 365 and Airtable work, and follow through on tasks, but the person mostly ended up pasting errors back into ChatGPT and following its troubleshooting instructions. It did manage to resume a project, but immediately hit the same usage limit they were trying to avoid.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1we0319/i_trusted_chatgpt_to_help_me_build_an_ai",
   "prompt": "Assistant mode: run a business assistant that resumes project work. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/run-a-business-assistant-that-resumes-project-work.html"
  },
  {
   "id": "run-a-monday-competitor-monitoring-brief-from-job-listings",
   "title": "Run a Monday competitor-monitoring brief from job listings",
   "summary": "Real-world thread: the poster set Claude to run every Monday at 8am with a prompt listing competitors and asking it to check each company’s website for recent activity: pricing changes, new products or features, new content, announcements or press, and job listings that hint at strategy. Claude then condenses what changed that week across all of them and highlights the single most important item; the person’s role was mainly supplying the competitor list and reading the finished brief. The stated result is a Monday-morning brief delivered without manual checking or a dashboard, with job listings used to infer likely moves such as a push on distribution.",
   "agent": "Claude",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/PromptEngineering/comments/1tz5f6n/i_set_claude_up_as_an_agent_that_spies_on_my",
   "prompt": "Assistant mode: run a Monday competitor-monitoring brief from job listings. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/run-a-monday-competitor-monitoring-brief-from-job-listings.html"
  },
  {
   "id": "reconfigure-home-assistant-lighting-automations",
   "title": "Reconfigure Home Assistant lighting automations",
   "summary": "Real-world thread: someone prompted an AI to edit Home Assistant automations so any signal turning on the house's bright white LED spots would instead kill those and activate the diffuse indirect light strips instead. The AI also added rules turning all lights off at 10pm, with color ones switching to red only. The house now uses the softer glow lights by default and shifts to full red at night.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/levelsio/status/2046960170643689706",
   "prompt": "Assistant mode: reconfigure Home Assistant lighting automations. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/reconfigure-home-assistant-lighting-automations.html"
  },
  {
   "id": "proposal-trip-to-japan",
   "title": "Proposal trip to Japan",
   "summary": "A writer tested Manus on planning a seven-day romantic trip to Japan from Seattle and reviewed the result.",
   "agent": "Manus",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://medium.com/@breezen100/actual-test-of-manus-travel-agent-a-dreamy-7-day-proposal-trip-to-japan-from-seattle-3ccc64abc22c",
   "prompt": "Plan a 7-day trip to Japan for two, with one special evening suitable for a proposal. Include flights to research, hotels, and a daily plan.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/proposal-trip-to-japan.html"
  },
  {
   "id": "plan-meals-and-place-grocery-orders-from-a-google-doc",
   "title": "Plan meals and place grocery orders from a Google Doc",
   "summary": "Claude covered plan meals and place grocery orders from a Google Doc. The poster says they assembled a workflow with Claude and Muse to run a two-week household meal cycle.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1wxlm01/ai_is_a_hellofresh_killer_automated_meal_prep_and",
   "prompt": "Shopping task: plan meals and place grocery orders from a Google Doc. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-meals-and-place-grocery-orders-from-a-google-doc.html"
  },
  {
   "id": "plan-family-meals-and-enter-recipes-into-grocy",
   "title": "Plan family meals and enter recipes into Grocy",
   "summary": "As described in the source, the builder ran Grocy in Home Assistant and gave Claude Cowork's computer use access to it in Chrome. He supplied recipes from books and online, had Claude scale them for four, and left it to type them in while he did other things. Claude added all the recipes and stood up the product categories, needing a couple of tweaks.",
   "agent": "Claude Cowork",
   "topic": "Home & Family",
   "source": "Blog",
   "url": "https://tech.yahoo.com/ai/claude/articles/meal-planning-stresses-let-claude-113016928.html",
   "prompt": "Home/family task: plan family meals and enter recipes into Grocy. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-family-meals-and-enter-recipes-into-grocy.html"
  },
  {
   "id": "plan-a-trip-and-book-flight-hotel-and-shuttle",
   "title": "Plan a trip and book flight, hotel and shuttle",
   "summary": "Itinerary planning: fine. Shuttle nag: fine. Form filling: fine. Entering 2FA and clicking “confirm” without asking: not fine. Dots therefore left John’s trip only partly booked.",
   "agent": "Dots",
   "topic": "Travel & Booking",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49904198",
   "prompt": "Plan this trip (flights, hotel, airport shuttle). Fill forms where you can, but stop for my confirmation before payment and never enter 2FA codes yourself.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-a-trip-and-book-flight-hotel-and-shuttle.html"
  },
  {
   "id": "plan-a-road-trip",
   "title": "Plan a road trip",
   "summary": "One practitioner notes that employed Claude as the complete planner for a visit to National Harbor, MD, and it proved to be a helpful resource on journeys.",
   "agent": "Claude Cowork",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1wh4j26/used_claude_cowork_for_trip_planning",
   "prompt": "Plan: plan a road trip. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-a-road-trip.html"
  },
  {
   "id": "plan-a-group-weekend-trip-and-make-reservations",
   "title": "Plan a group weekend trip and make reservations",
   "summary": "One practitioner notes that from his phone he prompted ChatGPT to use his full chat history for a long weekend with 8 friends, generate the three best options, build a full-stack site for the nine to coordinate preferences and reach agreement, then make reservations and draft the Gmail invite once decided. The agent handled the entire sequence on its own.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/sama/status/2081396796174282900",
   "prompt": "Plan: plan a group weekend trip and make reservations. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-a-group-weekend-trip-and-make-reservations.html"
  },
  {
   "id": "plan-a-family-christmas-trip-and-book-flights",
   "title": "Plan a family Christmas trip and book flights",
   "summary": "Linked example of plan a family Christmas trip and book flights. User gave Muse access for budgeting and health/fitness goals.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/HenryBeagle808/status/2101429329528082650",
   "prompt": "Plan: plan a family Christmas trip and book flights. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-a-family-christmas-trip-and-book-flights.html"
  },
  {
   "id": "plan-a-birthday-evening",
   "title": "Plan a birthday evening",
   "summary": "Community report: Meta Muse was pointed at plan a birthday evening. the agent for Muse arranged a wife's birthday celebration, arranged dinner, and obtained two tickets with authorization.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/TheExodusBTC/status/2105381716290580519",
   "prompt": "Plan: plan a birthday evening. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/plan-a-birthday-evening.html"
  },
  {
   "id": "phone-a-restaurant-to-book-a-full-table",
   "title": "Phone a restaurant to book a full table",
   "summary": "Real-world thread: nick Larkins asked OpenClaw to book a dinner reservation. The online booking attempt failed because the restaurant was fully booked. The agent located a last-minute cancellation and booked the table at a busy new restaurant.",
   "agent": "OpenClaw",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/Jacobsklug/status/2024489021486014972",
   "prompt": "Shopping task: phone a restaurant to book a full table. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/phone-a-restaurant-to-book-a-full-table.html"
  },
  {
   "id": "organize-saved-recipes-and-research-trip-flights",
   "title": "Organize saved recipes and research trip flights",
   "summary": "According to the write-up, the writer asked Muse, which Meta's account was already linked to, to go through all their saved Facebook and Instagram posts, pull out the recipes, and organize them. For travel, they gave it trip timing and a nonstop-only preference. Muse returned a recipe list grouped by type, with ingredients, steps and links to the original posts.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://www.engadget.com/2256577/how-to-get-started-with-meta-s-new-ai-agent-muse/",
   "prompt": "Plan: organize saved recipes and research trip flights. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/organize-saved-recipes-and-research-trip-flights.html"
  },
  {
   "id": "order-groceries-online",
   "title": "Order groceries online",
   "summary": "As described in the source, custom agent shows up in a write-up about order groceries online, with figures like 3.6, 27. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48519023",
   "prompt": "Shopping task: order groceries online. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/order-groceries-online.html"
  },
  {
   "id": "order-groceries-from-health-data-and-a-meal-plan",
   "title": "Order groceries from health data and a meal plan",
   "summary": "As described in the source, austen fed the AI his recent body scans, health data, preferences, and meal plan. The agent then placed grocery orders automatically via HEB delivery. It became one of his favorite AI uses.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/Austen/status/2088697527516221744",
   "prompt": "Shopping task: order groceries from health data and a meal plan. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/order-groceries-from-health-data-and-a-meal-plan.html"
  },
  {
   "id": "negotiate-retention-discounts-on-bills-by-phone",
   "title": "Negotiate retention discounts on bills by phone",
   "summary": "Real-world thread: kudos assembled a voice agent that phones a service provider for Someone, works through IVR menus, waits on hold, and asks for a retention discount. By default it negotiates within the existing plan and only changes plans if Someone says they're open to it. The co-founder saved $840 a year on T-Mobile after allowing plan changes, and an Xfinity example cut $276 a year.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47131955",
   "prompt": "Help me negotiate retention discounts on bills by phone. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/negotiate-retention-discounts-on-bills-by-phone.html"
  },
  {
   "id": "negotiate-hotel-rooms-by-phone-call",
   "title": "Negotiate hotel rooms by phone call",
   "summary": "Observed in the wild: the AI agent @hey__fo was instructed to handle hotel booking for two rooms over five nights in Japan. It placed the call independently and conducted the negotiation on the phone. The agent secured the exact same two rooms for all five nights.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/EngineerAquino/status/2104808165342871882",
   "prompt": "Plan: negotiate hotel rooms by phone call. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/negotiate-hotel-rooms-by-phone-call.html"
  },
  {
   "id": "negotiate-a-lower-comcast-bill-via-chatbot",
   "title": "Negotiate a lower Comcast bill via chatbot",
   "summary": "Observed in the wild: doNotPay assembled a bot on the OpenAI GPT-3 API that chats with company support. An engineer typed a request to lower the internet bill while keeping the same plan. It secured a $10 monthly discount, about $120 a year.",
   "agent": "ChatGPT",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/astuyve/status/2102844305253470515",
   "prompt": "Finance task: negotiate a lower Comcast bill via chatbot. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/negotiate-a-lower-comcast-bill-via-chatbot.html"
  },
  {
   "id": "maintain-a-home-media-library-and-handle-personal-projects",
   "title": "Maintain a home media library and handle personal projects",
   "summary": "Observed in the wild: one commenter said they use Grok Bot to run a home media library around Plex by doing the jobs they used to do manually. They said they asked it to standardize folders and filenames, sync and schedule backups, deduplicate, find missing posters, delete empty folders, and create custom thumbnails, descriptions, categories, and collections for films, TV, combat sports, documentaries, and creator videos; another said Grok Bot checked old email purchases to identify spare AliExpress parts and helped make a Tesla HUD that mirrors an S3XY Knob display onto a Tita HUD powered by an ESP32. The thread reports both setups working, but gives no hard metrics on time saved or reliability.",
   "agent": "Grok Bot",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1x1d7n9/how_are_you_all_actually_using_grok_bot_share",
   "prompt": "Shopping task: maintain a home media library and handle personal projects. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/maintain-a-home-media-library-and-handle-personal-projects.html"
  },
  {
   "id": "lower-an-internet-bill-and-file-expense-reports",
   "title": "Lower an internet bill and file expense reports",
   "summary": "Documented elsewhere: one commenter said they Leaned on GrokBot to negotiate with Spectrum in a chat session. Their only stated step was logging in; after that, the bot handled the conversation by itself and got a lower rate. The Spectrum commenter said the bill ended up $20 lower per month.",
   "agent": "Grok Bot",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1wvhcmh/has_anyone_actually_made_money_from_a_grok_bot",
   "prompt": "Finance task: lower an internet bill and file expense reports. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/lower-an-internet-bill-and-file-expense-reports.html"
  },
  {
   "id": "hotel-search-with-a-browser-operator-agent",
   "title": "Hotel search with a browser operator agent",
   "summary": "A reviewer tested OpenAI Operator on multi-step hotel lookups with budget and date constraints, noting shallow results but real form interaction.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://www.asad.pw/ai-agents-experiences-with-manus-operator-and-more/",
   "prompt": "Find hotels in <city> for <dates> under <budget> that match <preferences>. Open listing pages, compare options in a table, and stop before any booking.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/hotel-search-with-a-browser-operator-agent.html"
  },
  {
   "id": "handle-household-tasks-and-reminders",
   "title": "Handle household tasks and reminders",
   "summary": "Case study takeaway: user speaks tasks like booking dentist appointments or school pickups directly to Nori, which captures them and adds to the family calendar automatically. Nori uses Call Alert to phone Someone about reminders instead of sending notifications, and shares all updates across family members' calendars without further input. User reports no more forgotten tasks, double-bookings, or mental clutter, and no longer tracks everything manually.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/FutureStacked/status/2016579153147920636",
   "prompt": "Home/family task: handle household tasks and reminders. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/handle-household-tasks-and-reminders.html"
  },
  {
   "id": "handle-family-errands-and-recipe-sorting",
   "title": "Handle family errands and recipe sorting",
   "summary": "Observed in the wild: Codex was pointed at handle family errands and recipe sorting. a communications professional arranged for her son's doctor to receive a prescription via Codex and organize 240 recipes into an index, spreadsheet, and Drive folder, with 62 family-fit ones included.",
   "agent": "Codex",
   "topic": "Home & Family",
   "source": "Blog",
   "url": "https://youbetabelieveit.substack.com/p/oh-holy-codex-the-ai-coworker-has",
   "prompt": "Home/family task: handle family errands and recipe sorting. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/handle-family-errands-and-recipe-sorting.html"
  },
  {
   "id": "grocery-order-built-from-purchase-history",
   "title": "Grocery order built from purchase history",
   "summary": "An agent logs into a supermarket site, uses past orders and current deals to propose meals, and fills the basket for review.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://markooms.com/blog/ai-agent-does-my-groceries",
   "prompt": "Using my last few orders and this week's deals, propose 5 dinners and fill the cart with the ingredients. Stop before checkout.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/grocery-order-built-from-purchase-history.html"
  },
  {
   "id": "fix-wrong-day-flight-reservations-between-bots",
   "title": "Fix wrong-day flight reservations between bots",
   "summary": "According to the write-up, the speaking-gig agent noticed that flight reservations for upcoming speeches were on the wrong days. It contacted Someone to report the issue and asked permission to involve the separate travel bot. All reservations were updated without further input from Someone.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/undefinedKi/status/2100042527172690396",
   "prompt": "Plan: fix wrong-day flight reservations between bots. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/fix-wrong-day-flight-reservations-between-bots.html"
  },
  {
   "id": "fix-a-booking-made-on-the-wrong-date",
   "title": "Fix a booking made on the wrong date",
   "summary": "Case study takeaway: dot had access to David's travel bookings for an upcoming Europe trip. It proactively scanned the details, identified the Paris to Barcelona flight on the incorrect date, and updated the booking to the correct dates on its own. The booking was changed to the right dates.",
   "agent": "Dots",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/officialgambi/status/2105316170228498495",
   "prompt": "Plan: fix a booking made on the wrong date. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/fix-a-booking-made-on-the-wrong-date.html"
  },
  {
   "id": "find-trending-articles-and-generate-linkedin-post-variations",
   "title": "Find trending articles and generate LinkedIn post variations",
   "summary": "A published walkthrough shows in the tutorial, Operator was asked to create 10 LinkedIn post variations from eCommerce news. It searched Google for trending eCommerce articles and picked one about digital shopping trends. Operator produced 10 different LinkedIn posts from the chosen article.",
   "agent": "Operator",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://gptcentral.substack.com/p/how-to-use-chatgpt-operator-automate",
   "prompt": "Shopping task: find trending articles and generate LinkedIn post variations. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/find-trending-articles-and-generate-linkedin-post-variations.html"
  },
  {
   "id": "find-a-cheaper-rome-flight-and-hotel-combination",
   "title": "Find a cheaper Rome flight and hotel combination",
   "summary": "According to the write-up, someone had GPT draft a prompt, then gave that prompt to MuleRun’s browser agent to find a September round-trip flight to Rome plus a mid-range city-center hotel for five nights under $1,500, with taxes and fees included. The agent compared flights and hotels across six platforms, filtered hidden fees, checked refund policies, organized the best three options, and surfaced that booking a direct flight and hotel separately beat package deals; the person then chose and booked one option. It located a direct flight plus boutique hotel for $1,280 total versus a package Someone almost booked for $1,780, a claimed $500 savings.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/automation/comments/1obkk8u",
   "prompt": "Plan: find a cheaper Rome flight and hotel combination. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/find-a-cheaper-rome-flight-and-hotel-combination.html"
  },
  {
   "id": "find-a-london-flat-to-rent",
   "title": "Find a London flat to rent",
   "summary": "According to the write-up, the poster had Claude Cowork run a repeat search across SpareRoom, OpenRent, Rightmove, and Zoopla twice a day. It screened out student flats and homes with three or more bedrooms, drafted personalised outreach for promising listings, and emailed the poster a digest they could act on from their phone. The poster says this process located them a London rental in five days.",
   "agent": "Claude Cowork",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1smay7l/claude_cowork_found_me_a_flat_to_rent_in_london",
   "prompt": "Plan: find a London flat to rent. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/find-a-london-flat-to-rent.html"
  },
  {
   "id": "fill-out-a-long-gp-form",
   "title": "Fill out a long GP form",
   "summary": "Case study takeaway: someone instructed their agent to complete a long GP form they had been avoiding. The agent handled the full submission on its own with no further user input during the process. The doctor called and booked the tests automatically.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/Arf_22/status/2106014418555318454",
   "prompt": "Home/family task: fill out a long GP form. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/fill-out-a-long-gp-form.html"
  },
  {
   "id": "fill-a-supermarket-cart",
   "title": "Fill a supermarket cart",
   "summary": "One practitioner notes that as someone was at work, groceries were added to the cart at a nearby supermarket in agent mode, ready for collection the next day.",
   "agent": "ChatGPT Agent",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m9u7e6",
   "prompt": "Shopping task: fill a supermarket cart. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/fill-a-supermarket-cart.html"
  },
  {
   "id": "file-insurance-claims-for-overcharges",
   "title": "File insurance claims for overcharges",
   "summary": "Documented elsewhere: tarun gave the agent access to his insurance account information. After spotting nearly $1,700 in overcharges on items that should have been zero copay, he instructed the agent to file all the claims. The agent filed the claims and $200 has been recovered so far.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/_choppingblock/status/2106109741588238390",
   "prompt": "Finance task: file insurance claims for overcharges. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/file-insurance-claims-for-overcharges.html"
  },
  {
   "id": "file-airline-delay-compensation-and-rebook-a-flight",
   "title": "File airline delay compensation and rebook a flight",
   "summary": "As described in the source, after a 7-hour flight delay, Someone (@cryptopunk7213 on X) asked Muse to file for compensation. According to Someone's tweet, Muse handled the claim, located and rebooked a new flight, and replied to the airline's support email on its own. About five minutes later, Someone had a $250 credit in their Delta account, plus the rebooked flight.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/cryptopunk7213/status/2102179999122153690",
   "prompt": "Plan: file airline delay compensation and rebook a flight. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/file-airline-delay-compensation-and-rebook-a-flight.html"
  },
  {
   "id": "fight-a-denied-home-insurance-water-damage-claim",
   "title": "Fight a denied home insurance water damage claim",
   "summary": "Custom agent covered fight a denied home insurance water damage claim. After a fallen branch left a hole in the roof and more than $60,000 in water damage, the poster says they used AI to study their homeowners policy, research relevant state laws, and build a strategy to challenge the carrier’s decision.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SideProject/comments/1tjz4oj/i_used_ai_to_fight_a_home_insurance_claim_denial",
   "prompt": "Finance task: fight a denied home insurance water damage claim. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/fight-a-denied-home-insurance-water-damage-claim.html"
  },
  {
   "id": "extract-school-dates-from-forwarded-emails-and-flyers",
   "title": "Extract school dates from forwarded emails and flyers",
   "summary": "Custom agent covered extract school dates from forwarded emails and flyers. The Reddit post describes Ohai as a household-focused organizer where a parent forwards school emails and the AI extracts dates, or snaps a photo of a flyer and it reads the event details.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AIAssisted/comments/1scml9z/best_ai_task_manager_for_family_stuff_not_work",
   "prompt": "Home/family task: extract school dates from forwarded emails and flyers. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/extract-school-dates-from-forwarded-emails-and-flyers.html"
  },
  {
   "id": "everyday-errands-with-a-browser-agent",
   "title": "Everyday errands with a browser agent",
   "summary": "A designer tested an agent on real chores such as filling a grocery cart and researching purchases, noting where it stalled.",
   "agent": "ChatGPT Agent",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://netsvetaev.com/blog/ai/using-new-chatgpt-agent-in-a-real-life/",
   "prompt": "Add these items to my cart at <store>, choosing the cheapest acceptable option for each, and list anything you couldn't find.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/everyday-errands-with-a-browser-agent.html"
  },
  {
   "id": "draft-parent-teacher-communications",
   "title": "Draft parent-teacher communications",
   "summary": "Linked example of draft parent-teacher communications. The teacher said they first used ChatGPT for lesson outlines, then for weekly comments, then to refine those comments from added input, and eventually to draft a parent letter.",
   "agent": "ChatGPT",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Teachers/comments/12zdtxb",
   "prompt": "Home/family task: draft parent-teacher communications. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/draft-parent-teacher-communications.html"
  },
  {
   "id": "dispute-an-erroneous-medical-collections-bill",
   "title": "Dispute an erroneous medical collections bill",
   "summary": "Case study takeaway: the finance bot Warren received Someone’s healthcare login. It logged in, located the EOB for the year-old surgery, determined the collection agency was in error because insurance had paid, and faxed a dispute letter that included the EOB document. The bot targeted an unknown $13K medical collections debt and initiated the dispute process to remove it.",
   "agent": "Grok Bot",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/dallastsladavid/status/2092357445964222865",
   "prompt": "Finance task: dispute an erroneous medical collections bill. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/dispute-an-erroneous-medical-collections-bill.html"
  },
  {
   "id": "detailed-travel-planning-test",
   "title": "Detailed travel planning test",
   "summary": "A user gave Manus a complex multi-constraint trip and documented how the plan and handbook turned out.",
   "agent": "Manus",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://blog.yingyingz.com/p/my-experience-using-manus-ai-for",
   "prompt": "Plan a <n>-day trip to <destination> with these constraints: <list>. Give a day-by-day itinerary, transport between stops, and a backup option for each day.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/detailed-travel-planning-test.html"
  },
  {
   "id": "design-product-labels-and-produce-sds-sheets",
   "title": "Design product labels and produce SDS sheets",
   "summary": "A published walkthrough shows the seller used Manus across packaging and compliance work: it designed new co-branded product labels, read screenshots of new customer testimonials and added them into updated files, and generated three SDS sheets. A person still had to step in on label production, because when Manus offered to turn its label concepts into print-ready files, the output was not actually usable, so they traced from the rendering and made the final production file themselves. It saved enough labor that Someone was comfortable paying $200-$300 per month, and they say the three SDS sheets alone avoided about $900 in software fees or $3,000 in hired writing costs.",
   "agent": "Manus",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ManusOfficial/comments/1t0tx2j/one_month_of_manus_changed_my_life",
   "prompt": "Creative task: design product labels and produce SDS sheets. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/design-product-labels-and-produce-sds-sheets.html"
  },
  {
   "id": "coordinate-background-threads-and-send-admin-reminders",
   "title": "Coordinate background threads and send admin reminders",
   "summary": "Observed in the wild: someone kept one main conversation with the dot. The dot coordinated other threads in the background on its own and sent proactive messages reminding Someone to send quarterly documents to the accountant and check on a refund. Someone located the single-conversation coordination convenient and the reminders magical, with no metrics or shortfalls stated.",
   "agent": "Dots",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/Angaisb_/status/2104981592053616663",
   "prompt": "Finance task: coordinate background threads and send admin reminders. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/coordinate-background-threads-and-send-admin-reminders.html"
  },
  {
   "id": "control-the-home-by-voice",
   "title": "Control the home by voice",
   "summary": "According to the write-up, create a home assistant that can monitor household goods, lights, air conditioning, and moisture levels using voice commands, utilizing AI input.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/HerbWetherell/status/2105798686194417918",
   "prompt": "Home/family task: control the home by voice. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/control-the-home-by-voice.html"
  },
  {
   "id": "complete-online-surveys-for-free-stuff",
   "title": "Complete online surveys for free stuff",
   "summary": "Case study takeaway: the poster says they had a Grok bot handle a Burger King receipt survey after getting a code on the receipt. They gave the bot the task, and about 30 minutes later it messaged back a link to download the coupon for the free Whopper; the post does not describe any other setup or human steps. It worked once: the bot completed the survey and returned the free-coupon link.",
   "agent": "Grok Bot",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1w7dt58/my_grok_bot_completes_surveys_for_free_stuff_well",
   "prompt": "Finance task: complete online surveys for free stuff. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/complete-online-surveys-for-free-stuff.html"
  },
  {
   "id": "compare-and-book-a-hotel-online",
   "title": "Compare and book a hotel online",
   "summary": "Hotel hunt + checkout handled in one Grok Bot pass. The user mainly reported that the automated purchase came in cheaper than doing it manually.",
   "agent": "Grok Bot",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/grok/comments/1x2s5cg/how_to_save_yourself_hours_of_hassle_when_doing_a",
   "prompt": "Plan: compare and book a hotel online. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/compare-and-book-a-hotel-online.html"
  },
  {
   "id": "collate-school-homework-and-due-dates-into-a-child-x27-s-calendar",
   "title": "Collate school homework and due dates into a child's calendar",
   "summary": "From the linked report, meta Muse shows up in a write-up about collate school homework and due dates into a child's calendar. The linked post is the primary source for steps and outcomes.",
   "agent": "Meta Muse",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49796524",
   "prompt": "Assistant mode: collate school homework and due dates into a child's calendar. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/collate-school-homework-and-due-dates-into-a-child-x27-s-calendar.html"
  },
  {
   "id": "coach-training-and-diet-from-wearable-data",
   "title": "Coach training and diet from wearable data",
   "summary": "Observed in the wild: one commenter described a setup very close to this: Garmin data is pulled hourly into an InfluxDB on their server, and Claude Code runs there on a booked daily morning task. Claude reads load, stress, sleep and related metrics, calculates a training-readiness score itself, builds and adjusts the training plan, and publishes the result to a small webpage they host for phone access; they also gave it an initial full export plus years of Garmin, Strava, and .fit history for context. They said it keeps daily training specific enough to avoid burnout or overtraining and can notice rough periods like vacation, long work weeks, and time-zone travel.",
   "agent": "Claude",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1x2c0v0/has_anyone_used_claude_as_a_training_and_diet",
   "prompt": "Home/family task: coach training and diet from wearable data. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/coach-training-and-diet-from-wearable-data.html"
  },
  {
   "id": "claim-refunds-for-delayed-or-cancelled-flights",
   "title": "Claim refunds for delayed or cancelled flights",
   "summary": "From the linked report, following a flight cancellation or delay, Claude compiles the facts, examines the regulations, and composes the refund request email.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Anthropic/comments/1x3m5lk/use_claude_to_never_leave_any_from_an",
   "prompt": "Go through my mail for this goal: claim refunds for delayed or cancelled flights. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/claim-refunds-for-delayed-or-cancelled-flights.html"
  },
  {
   "id": "check-taxes-against-receipts-and-bank-statements",
   "title": "Check taxes against receipts and bank statements",
   "summary": "One practitioner notes that the poster fed Claude Opus the numbers and dates from a more complicated household tax situation involving a spouse’s LLC, revenue and expense streams, and some public business names. Claude checked their first pass, caught issues like double-counted revenue and expenses and client attribution problems, rechecked the calculations, and produced a spreadsheet that Someone then entered into FreeTaxUSA; Someone says they also reviewed it against receipts and bank statements themselves. They say the final figures matched their receipts and bank statements after several prompts, and it helped enough that they skipped a CPA who had quoted $2,000.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rut65g/anyone_else_doing_taxes_with_claude",
   "prompt": "Finance task: check taxes against receipts and bank statements. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/check-taxes-against-receipts-and-bank-statements.html"
  },
  {
   "id": "call-hotels-to-verify-reservations-after-booking",
   "title": "Call hotels to verify reservations after booking",
   "summary": "Real-world thread: people used AI phone agents as personal assistants after booking, sending them to call hotels and verify that a reservation was entered correctly. In the thread, a front-desk worker says the agents repeatedly called back after being refused; the hotel would not confirm details because the caller could not be verified as the actual guest, so the human staffer had to keep blocking the request. The reported result was not successful confirmation but repeated nuisance calls that tied up the front desk.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/TalesFromTheFrontDesk/comments/1oi09ou/i_wish_people_would_stop_sending_their_ai_agents",
   "prompt": "Plan: call hotels to verify reservations after booking. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/call-hotels-to-verify-reservations-after-booking.html"
  },
  {
   "id": "call-a-car-rental-company-to-add-an-infant-seat",
   "title": "Call a car rental company to add an infant seat",
   "summary": "A published walkthrough shows someone asked JooVoice to order an infant seat for an existing rental car booking. The agent called the company, navigated the automated menu, and remained on hold until connected. The request was handled successfully from start to finish.",
   "agent": "Grok Bot",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/gtupak/status/2109993490495226347?s=20",
   "prompt": "Shopping task: call a car rental company to add an infant seat. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/call-a-car-rental-company-to-add-an-infant-seat.html"
  },
  {
   "id": "build-purchase-orders-and-track-payables-nightly",
   "title": "Build purchase orders and track payables nightly",
   "summary": "According to the write-up, make.com automation was set up to build purchase orders and perform other tasks. It watches payables nightly on its own and updates estimating formulas in real time by pulling the most recent paid costs for materials such as deck boards or nails. Formulas stayed current through Covid amid weekly and monthly material price increases.",
   "agent": "Make",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/jbuesking/status/2107954334788989409",
   "prompt": "Shopping task: build purchase orders and track payables nightly. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/build-purchase-orders-and-track-payables-nightly.html"
  },
  {
   "id": "build-group-grocery-carts-through-a-signal-based-agent",
   "title": "Build group grocery carts through a Signal-based agent",
   "summary": "A published walkthrough shows the builder runs a self-hosted agent through Signal, with group chat limited to whitelisted accounts. It sketches a starter grocery list from past orders, kept locally or pulled from the store's site. Cart building works end to end.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47110375",
   "prompt": "Shopping task: build group grocery carts through a Signal-based agent. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/build-group-grocery-carts-through-a-signal-based-agent.html"
  },
  {
   "id": "build-and-order-a-grocery-cart-from-an-idea",
   "title": "Build and order a grocery cart from an idea",
   "summary": "Observed in the wild: muse received an Instacart connector and user access to a preferred store. Someone gave a phrase such as \"Taco Tuesday\"; Muse then selected matching items, populated the cart, completed checkout, and booked delivery through the same Instacart flow. No performance numbers or completion rates are stated.",
   "agent": "Meta Muse",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/dps/status/2102519594431000982",
   "prompt": "Shopping task: build and order a grocery cart from an idea. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/build-and-order-a-grocery-cart-from-an-idea.html"
  },
  {
   "id": "book-tables-and-watch-for-openings-or-price-drops",
   "title": "Book tables and watch for openings or price drops",
   "summary": "Observed in the wild: the agent receives a booking request for tables or travel. It completes the reservation on its own and continues monitoring afterward for newly available slots or lower fares. It acts ahead of Someone checking, unlike list-only chatbots.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/igus_ai/status/2110421777646018742",
   "prompt": "Plan: book tables and watch for openings or price drops. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-tables-and-watch-for-openings-or-price-drops.html"
  },
  {
   "id": "book-restaurants-and-snipe-campsites-and-tickets",
   "title": "Book restaurants and snipe campsites and tickets",
   "summary": "Real-world thread: grok Bot received calendar access and booking site logins. It read schedules, selected a dinner time, booked the restaurant reservation, and renegotiated in mixed Chinese and English. The restaurant example received 937 likes and reached 4.",
   "agent": "Grok Bot",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/yunta_tsai/status/2087415205756391461",
   "prompt": "Plan: book restaurants and snipe campsites and tickets. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-restaurants-and-snipe-campsites-and-tickets.html"
  },
  {
   "id": "book-restaurants-and-a-hotel-dinner-in-japanese",
   "title": "Book restaurants and a hotel dinner in Japanese",
   "summary": "A published walkthrough shows muse received the Japan trip details and was tasked with handling all restaurant bookings. It made the reservations on its own across multiple sites, then exchanged messages in Japanese directly with the Hakone hotel to request and confirm a dinner slot that did not appear on the hotel website. Every restaurant reservation was completed.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/Han_Fang_/status/2101874865385169348",
   "prompt": "Plan: book restaurants and a hotel dinner in Japanese. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-restaurants-and-a-hotel-dinner-in-japanese.html"
  },
  {
   "id": "book-flights-hotel-and-rental-car",
   "title": "Book flights, hotel and rental car",
   "summary": "From the linked report, someone gave Grok bot access to email and booking services for life admin tasks. The agent booked flights, hotel and rental car. All listed tasks were completed.",
   "agent": "Grok Bot",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/thecsguy/status/2108767281811804657",
   "prompt": "Plan: book flights, hotel and rental car. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-flights-hotel-and-rental-car.html"
  },
  {
   "id": "book-flights",
   "title": "Book flights",
   "summary": "As described in the source, meta Muse shows up in a write-up about book flights. The linked post is the primary source for steps and outcomes.",
   "agent": "Meta Muse",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/cy_zanfar/status/2099684668391969009",
   "prompt": "Plan: book flights. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-flights.html"
  },
  {
   "id": "book-a-restaurant-table-through-a-resy-connector",
   "title": "Book a restaurant table through a Resy connector",
   "summary": "Real-world thread: the writer added the Resy connector to Claude through Customize > Connectors, then chatted about where to eat near his Midtown apartment (table for three, around 6:30 pm). Claude suggested The Smith and pulled up availability. Claude picked a good restaurant and showed open slots, but the writer still had to finish the reservation on Resy himself.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://www.pcworld.com/article/3123989/chatgpt-and-claude-tried-to-book-my-dinner-it-got-clunky-fast.html",
   "prompt": "Shopping task: book a restaurant table through a Resy connector. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-a-restaurant-table-through-a-resy-connector.html"
  },
  {
   "id": "book-a-private-chef-and-plan-a-ski-holiday",
   "title": "Book a private chef and plan a ski holiday",
   "summary": "From the linked report, chatGPT received casual spoken instructions while Someone walked the dog. For the chef it located candidates, drafted and sent quote and availability emails, processed incoming replies, produced decision summaries, and tracked missing details. The agent produced usable summaries and quotes that let Someone select a chef and assemble a full DIY trip for two families.",
   "agent": "ChatGPT",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/pauldm/status/2107094663551168709",
   "prompt": "Plan: book a private chef and plan a ski holiday. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-a-private-chef-and-plan-a-ski-holiday.html"
  },
  {
   "id": "book-a-flight-from-your-phone",
   "title": "Book a flight from your phone",
   "summary": "Phone in one hand, dog in the other—Tom still got a flight reserved through ChatGPT Work. After a short personal-info quiz, payment happened inside a remote browser session he controlled.",
   "agent": "ChatGPT",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/T3metrics/status/2096297283830853810",
   "prompt": "Plan: book a flight from your phone. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-a-flight-from-your-phone.html"
  },
  {
   "id": "book-a-new-delhi-to-mumbai-flight",
   "title": "Book a New Delhi to Mumbai flight",
   "summary": "Case study takeaway: claude computer use shows up in a write-up about book a New Delhi to Mumbai flight. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude computer use",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://rittikajindal.medium.com/my-first-hand-experience-with-claudes-computer-use-when-ai-takes-control-a0f78445a6c7",
   "prompt": "Plan: book a New Delhi to Mumbai flight. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/book-a-new-delhi-to-mumbai-flight.html"
  },
  {
   "id": "add-family-to-a-dental-plan-by-voice-from-the-car",
   "title": "Add family to a dental plan by voice from the car",
   "summary": "Linked example of add family to a dental plan by voice from the car. Someone spoke instructions to their car while driving, which relayed them to the bot.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/doctorrockk/status/2110197730765062496",
   "prompt": "Shopping task: add family to a dental plan by voice from the car. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 70,
   "page": "https://agentsightings.com/case/add-family-to-a-dental-plan-by-voice-from-the-car.html"
  },
  {
   "id": "restaurant-booking-with-an-allergy-note",
   "title": "Restaurant booking with an allergy note",
   "summary": "In a usability review, a browser agent researched restaurants, filled the reservation form including an allergy note, and paused for approval.",
   "agent": "ChatGPT Agent",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://www.nngroup.com/articles/impressions-chatgpt-agent/",
   "prompt": "Find a quiet restaurant near <area> for 4 people at 7pm Friday that handles nut allergies. Fill in the booking form and stop for my confirmation before submitting.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/restaurant-booking-with-an-allergy-note.html"
  },
  {
   "id": "plan-business-trips-and-learn-from-each-debrief",
   "title": "Plan business trips and learn from each debrief",
   "summary": "Documented elsewhere: assembled on Hermes Agent's own memory, skills and web search, running on gemini-3-flash-preview. With seeded traveler profiles, it ran three web searches for a Dubai trip and applied stored preferences unprompted. The first plan was one recommendation (about $850) ending at pending approval.",
   "agent": "Hermes Agent",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://dev.to/christopher_mburu_7b7e0ba/hermes-travelops-an-agent-that-learns-from-every-trip-5805",
   "prompt": "Plan: plan business trips and learn from each debrief. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/plan-business-trips-and-learn-from-each-debrief.html"
  },
  {
   "id": "plan-and-order-a-week-of-groceries",
   "title": "Plan and order a week of groceries",
   "summary": "From the linked report, claude shows up in a write-up about plan and order a week of groceries. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rb1jrz/automated_grocery_order",
   "prompt": "Shopping task: plan and order a week of groceries. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/plan-and-order-a-week-of-groceries.html"
  },
  {
   "id": "negotiate-a-lower-godaddy-support-price-via-chat-agent",
   "title": "Negotiate a lower GoDaddy support price via chat agent",
   "summary": "One practitioner notes that the poster had an AI agent handle a GoDaddy support chat, with instructions to negotiate a lower price and check the chat every 60 seconds for new replies. In the comments, they said they used FillApp, a Chrome extension, and the agent carried the exchange until it reached a lower offer; the human appears to have mainly set it up and let it run. They said the final price was renegotiated down to $278, which they described as a $70 savings.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/agi/comments/1ns67b7",
   "prompt": "Help me negotiate a lower GoDaddy support price via chat agent. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/negotiate-a-lower-godaddy-support-price-via-chat-agent.html"
  },
  {
   "id": "handle-restaurant-reservations-over-whatsapp",
   "title": "Handle restaurant reservations over WhatsApp",
   "summary": "Case study takeaway: created a WhatsApp representative for a nearby eatery that keeps watch on availability, responds to inquiries, and authenticates reservations, with only one paying patron so far.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SaaS/comments/1jen9vv",
   "prompt": "Shopping task: handle restaurant reservations over WhatsApp. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/handle-restaurant-reservations-over-whatsapp.html"
  },
  {
   "id": "generate-a-spoken-daily-news-bulletin-at-home",
   "title": "Generate a spoken daily news bulletin at home",
   "summary": "One practitioner notes that the poster Leaned on n8n to run a 6:30 a.m. workflow that asked Perplexity for factual stories from a fixed list of outlets, cleaned that output, then had ChatGPT rewrite it as a short spoken script. It replaced the poster’s Alexa flash briefing with a calmer custom bulletin and they shared a sample audio result, but they said it still needed better pauses and maybe a different voice.",
   "agent": "ChatGPT",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/homeassistant/comments/1og2mcn/goodbye_alexa_hello_local_ai_news_my_home",
   "prompt": "Assistant mode: generate a spoken daily news bulletin at home. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/generate-a-spoken-daily-news-bulletin-at-home.html"
  },
  {
   "id": "find-discounted-bank-owned-homes-and-list-them-as-midterm-rentals",
   "title": "Find discounted bank-owned homes and list them as midterm rentals",
   "summary": "Video walkthrough of MidtermMate: scout REO/bank-owned inventory on discount, estimate what the block rents for, then package the units as furnished midterm stays.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/VeWorldAI/status/2104601779014947045",
   "prompt": "Finance task: find discounted bank-owned homes and list them as midterm rentals. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/find-discounted-bank-owned-homes-and-list-them-as-midterm-rentals.html"
  },
  {
   "id": "draft-appeals-for-denied-insurance-claims",
   "title": "Draft appeals for denied insurance claims",
   "summary": "From the linked report, an n8n automation triggers the moment a claim is marked Denied, filters anything already appealed, then pulls the full patient record, prescriptions, and test results tied to that patient ID. One AI call zeros in on the likely denial reason and sketches the formal appeal letter citing the clinical details. It outputs a real Google Doc with the letter, updates a tracking sheet, and sends a branded email to staff with a review link.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/Tariere5/status/2104521353386766588",
   "prompt": "Finance task: draft appeals for denied insurance claims. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/draft-appeals-for-denied-insurance-claims.html"
  },
  {
   "id": "date-night-planned-end-to-end",
   "title": "Date night planned end to end",
   "summary": "A writer had an agent pick a restaurant and a movie, choose seats, and stop at payment.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://www.yahoo.com/lifestyle/articles/tried-using-chatgpt-agent-plan-213526320.html",
   "prompt": "Plan a date night for Saturday: dinner and a movie nearby, within <budget>. Pick seats and pause before any payment.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/date-night-planned-end-to-end.html"
  },
  {
   "id": "build-a-health-plan-from-genetic-and-bloodwork-data",
   "title": "Build a health plan from genetic and bloodwork data",
   "summary": "A published walkthrough shows farzad uploaded his full raw 23andme genetic file, ways2well bloodwork, artery plaque scans, and complete supplement routine to Claude Code. The agent reviewed the datasets and discussed genetic predispositions with him over a 2-hour conversation. A tailored plan based on his specific genes and health markers resulted.",
   "agent": "Claude Code",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/farzyness/status/2008032877234004274",
   "prompt": "Home/family task: build a health plan from genetic and bloodwork data. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/build-a-health-plan-from-genetic-and-bloodwork-data.html"
  },
  {
   "id": "book-a-restaurant-table",
   "title": "Book a restaurant table",
   "summary": "Real-world thread: alexa+ shows up in a write-up about book a restaurant table. The linked post is the primary source for steps and outcomes.",
   "agent": "Alexa+",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/alexa/comments/1ryzfho/in_the_uk_been_offered_alexa_upgrade_for_free_across_all_devices_including_older_echo_dots_and_show_apart_from_a_different_voice_and_being_able_to_make_restaurant_bookings_via_opentable_i_cant_tell_any_real_difference_are_there_any_useful_advantages_any_disadvantages_is_it_really_free",
   "prompt": "Shopping task: book a restaurant table. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/book-a-restaurant-table.html"
  },
  {
   "id": "book-a-restaurant",
   "title": "Book a restaurant",
   "summary": "Case study takeaway: dots shows up in a write-up about book a restaurant. The linked post is the primary source for steps and outcomes.",
   "agent": "Dots",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/JulianGoldieSEO/status/2105628825266553328",
   "prompt": "Shopping task: book a restaurant. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 69,
   "page": "https://agentsightings.com/case/book-a-restaurant.html"
  },
  {
   "id": "win-a-small-legal-case-with-claude-x27-s-help",
   "title": "Win a small legal case with Claude's help",
   "summary": "One practitioner notes that the filing fee was the only expense involved in winning an $8,000 legal case with the assistance of Said Claude; further information is lacking.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1q51zw7/my_max_plan_just_paid_for_itself_for_the_next",
   "prompt": "Finance task: win a small legal case with Claude's help. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/win-a-small-legal-case-with-claude-x27-s-help.html"
  },
  {
   "id": "summarize-school-emails-for-parents",
   "title": "Summarize school emails for parents",
   "summary": "One practitioner notes that A Grok bot template was stood up for busy parents. It reads kids' school emails and reports nightly on items to pack, wear, sign or pay for as surfacing sign-ups, deadlines and blanks. No specific results or shortfalls are stated.",
   "agent": "Grok Bot",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1wg0edy/what_are_people_actually_using_chatgpt_scheduled",
   "prompt": "Home/family task: summarize school emails for parents. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/summarize-school-emails-for-parents.html"
  },
  {
   "id": "run-the-smart-home-by-natural-language",
   "title": "Run the smart home by natural language",
   "summary": "According to the write-up, karpathy put together Dobby the Elf Claw as the smart home AI agent. It was given access to control lights, HVAC, shades, pool, and security, with AI Leaned on to detect visitors and send alerts. All controls now operate through one agent.",
   "agent": "OpenClaw",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/kimmonismus/status/2035738351853056380",
   "prompt": "Home/family task: run the smart home by natural language. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/run-the-smart-home-by-natural-language.html"
  },
  {
   "id": "fill-in-a-tax-organizer-and-review-last-year-x27-s-return",
   "title": "Fill in a tax organizer and review last year's return",
   "summary": "A published walkthrough shows claude shows up in a write-up about fill in a tax organizer and review last year's return. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://cloudnativeengineer.substack.com/p/pay-your-taxes-with-claude-code",
   "prompt": "Shopping task: fill in a tax organizer and review last year's return. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/fill-in-a-tax-organizer-and-review-last-year-x27-s-return.html"
  },
  {
   "id": "file-a-complex-tax-return",
   "title": "File a complex tax return",
   "summary": "One practitioner notes that openClaw shows up in a write-up about file a complex tax return, with figures like 4. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/dvassallo/status/2039450517143974105",
   "prompt": "Shopping task: file a complex tax return. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/file-a-complex-tax-return.html"
  },
  {
   "id": "extract-expenses-from-credit-card-statements-into-excel",
   "title": "Extract expenses from credit card statements into Excel",
   "summary": "A published walkthrough shows the builder put all her personal credit card statements in one folder and ran Claude Code from inside it. She asked it to find likely business expenses to be reimbursed. She says it saved her hours.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://technicallycurious.substack.com/p/claude-code-for-business-users-getting",
   "prompt": "Help me extract expenses from credit card statements into Excel. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/extract-expenses-from-credit-card-statements-into-excel.html"
  },
  {
   "id": "audit-credit-card-use-for-wrong-card-charges-and-missed-benefits",
   "title": "Audit credit card use for wrong-card charges and missed benefits",
   "summary": "From the linked report, by powered by a Plaid-connected app, Grok Bot can read the transactions put together on three cards and verify their card selection for dining, Amazon, and travel, as well as their unused benefits.",
   "agent": "Grok Bot",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/ChuckCook/status/2094455257887617295",
   "prompt": "Shopping task: audit credit card use for wrong-card charges and missed benefits. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/audit-credit-card-use-for-wrong-card-charges-and-missed-benefits.html"
  },
  {
   "id": "appeal-a-denied-travel-insurance-claim",
   "title": "Appeal a denied travel insurance claim",
   "summary": "Linked example of appeal a denied travel insurance claim. The poster uploaded the full travel-insurance policy PDF plus all claim correspondence to Gemini and asked whether the denial fit the policy terms.",
   "agent": "Gemini",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GeminiAI/comments/1wdxo80/how_gemini_helped_me_successfully_appeal_a",
   "prompt": "Plan: appeal a denied travel insurance claim. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 68,
   "page": "https://agentsightings.com/case/appeal-a-denied-travel-insurance-claim.html"
  },
  {
   "id": "turn-vet-charges-into-ready-to-send-pet-insurance-claims",
   "title": "Turn vet charges into ready-to-send pet insurance claims",
   "summary": "Real-world thread: the owner uploads a bank CSV by hand, since the agent never stores bank logins. On a 15-minute tick the agent flags vet charges, searches Gmail for the matching invoice, and uses an LLM (Gemini) to extract it, falling back to OCR for scans. Across ten live approval letters, its settlement arithmetic check passed to the cent, and a second check raised questions on five.",
   "agent": "OpenClaw",
   "topic": "Finance",
   "source": "GitHub",
   "url": "https://github.com/jagberg/Me.OpenClaw",
   "prompt": "Finance task: turn vet charges into ready-to-send pet insurance claims. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/turn-vet-charges-into-ready-to-send-pet-insurance-claims.html"
  },
  {
   "id": "reconcile-orders-draft-invoices-and-order-laptops-nightly",
   "title": "Reconcile orders, draft invoices, and order laptops nightly",
   "summary": "From the linked report, chatGPT runs nightly booked prompts that pull completed orders, compare them to CRM entries, create draft invoices in the accounting system, and log deal names and amounts in Google Sheets; separate jobs file NDAs from email into CRM folders and handle insurance requirements from contracts. For hardware, it accessed the logged-in bank account for the credit card and HR system for addresses to order specific MacBook models from Apple and prepare a return box. The automations handle tasks equivalent to three people's work in less time, covering writing 1400 lines of commission calculation code, with around 20 such jobs active.",
   "agent": "ChatGPT",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/SunnyvaleSBay/status/2108047957874573477",
   "prompt": "Shopping task: reconcile orders, draft invoices, and order laptops nightly. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/reconcile-orders-draft-invoices-and-order-laptops-nightly.html"
  },
  {
   "id": "plan-a-family-trip-and-pre-fill-bookings",
   "title": "Plan a family trip and pre-fill bookings",
   "summary": "According to the write-up, the reporter asked ChatGPT Agent to plan a 4-day family trip to San Diego from Newark, with pool hotels and activities for kids under 10, and to save an itinerary in Google Docs. The agent searched Google Flights, compared kid-friendly hotels, and listed attractions such as the zoo and LEGOLAND. The itinerary and comparisons came out well, and the reporter called it an incredible planning tool.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://tech.yahoo.com/ai/articles/tested-chatgpt-agent-week-good-115833661.html",
   "prompt": "Plan: plan a family trip and pre-fill bookings. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/plan-a-family-trip-and-pre-fill-bookings.html"
  },
  {
   "id": "pay-bills-clear-inbox-and-order-groceries",
   "title": "Pay bills, clear inbox, and order groceries",
   "summary": "A published walkthrough shows someone introduced Scout as their Muse agent. Scout paid bills, reached inbox zero, and ordered groceries. All tasks finished before Someone's coffee went cold.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/Musecases/status/2099741960852341006",
   "prompt": "Go through my mail for this goal: pay bills, clear inbox, and order groceries. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/pay-bills-clear-inbox-and-order-groceries.html"
  },
  {
   "id": "find-flight-info-and-book-an-uber",
   "title": "Find flight info and book an Uber",
   "summary": "A published walkthrough shows someone shared travel details with Grok. While Someone showered, the agent located the flight information on its own and booked an Uber ride without further input. The tasks were completed, leaving Someone with a booked Uber ride and a personal assistant experience.",
   "agent": "Grok Bot",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/reggieoverton/status/2108481937929351400",
   "prompt": "Plan: find flight info and book an Uber. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/find-flight-info-and-book-an-uber.html"
  },
  {
   "id": "build-a-custom-agent-to-search-cyprus-villa-rentals",
   "title": "Build a custom agent to search Cyprus villa rentals",
   "summary": "As described in the source, marc Lou prompted an AI agent to pull public Cyprus real estate datasets and merge them with public AQI, water quality, noise maps, and landslide susceptibility layers. The agent then generated an Airbnb-style interface from the combined data. The site shows villa listings with environmental overlays and refreshes data every twenty-four hours.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/marclou/status/2096228864859201869",
   "prompt": "Plan: build a custom agent to search Cyprus villa rentals. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/build-a-custom-agent-to-search-cyprus-villa-rentals.html"
  },
  {
   "id": "book-travel-and-find-restaurant-reservations",
   "title": "Book travel and find restaurant reservations",
   "summary": "From the linked report, claude drives a Chrome browser through Chrome MCP. Claire has it book her flights and hotels in her company's travel booking system. The source gives no results or timings for either use.",
   "agent": "Claude",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://technicallycurious.substack.com/p/have-claude-browse-the-web-for-you",
   "prompt": "Plan: book travel and find restaurant reservations. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/book-travel-and-find-restaurant-reservations.html"
  },
  {
   "id": "async-devin-chores-between-morning-and-lunch",
   "title": "Async Devin chores between morning and lunch",
   "summary": "An HN commenter kicks off several Devin sessions at morning, lunch and night, then reviews PRs asynchronously instead of babysitting a terminal.",
   "agent": "Devin",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=43990359",
   "prompt": "Start a Devin session for each of these three small tickets. Prefer short, testable scopes. I will review the PRs at the next check-in and leave comments for follow-up.",
   "added": "2026-10-15",
   "interest": 67,
   "page": "https://agentsightings.com/case/async-devin-chores-between-morning-and-lunch.html"
  },
  {
   "id": "unilever-nova-agents-read-order-emails",
   "title": "Five agents read customer order emails for Unilever",
   "summary": "Unilever built NOVA, five agents on Microsoft Foundry: four read order emails and extract the order, a fifth checks their work. Across 11 pilot markets the first order-to-cash step fell from 15-20 minutes to about 90 seconds, and orders are 95% touchless.",
   "agent": "NOVA (Microsoft Foundry)",
   "topic": "Business & Startups",
   "source": "Case study (Microsoft)",
   "url": "https://ukstories.microsoft.com/features/unilevers-multi-agent-creating-customer-value-at-global-scale/",
   "prompt": "I get customer orders by email in messy formats. Read each email, pull out the customer, products, quantities and delivery date, then have a second pass check the extraction against the original. Anything missing or odd goes to a person with a note on what is unclear.",
   "added": "2026-10-29",
   "interest": 66,
   "page": "https://agentsightings.com/case/unilever-nova-agents-read-order-emails.html"
  },
  {
   "id": "triage-support-emails-and-draft-refund-policy-replies",
   "title": "Triage support emails and draft refund-policy replies",
   "summary": "Real-world thread: for a fictional kitchenware brand demo, the consultant kept the returns policy and brand voice guide in Claude Projects. Gmail and Sheets connectors gave Claude the unread support emails and an orders sheet with final-sale, gift and prior-return flags. The builder estimates 12–18 support hours saved weekly, assuming about 70% of requests are routine.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SaaS/comments/1r4hqph/we_replaced_3_support_agents_with_a_custom_ai",
   "prompt": "Shopping task: triage support emails and draft refund-policy replies. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/triage-support-emails-and-draft-refund-policy-replies.html"
  },
  {
   "id": "shop-on-amazon-and-build-pages",
   "title": "Shop on Amazon and build pages",
   "summary": "Observed in the wild: claude shows up in a write-up about shop on Amazon and build pages. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://daily.dev/posts/i-finally-tried-the-claude-chrome-extension-and-i-regret-waiting-this-long-hsg5pbkjx",
   "prompt": "Shopping task: shop on Amazon and build pages. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/shop-on-amazon-and-build-pages.html"
  },
  {
   "id": "shop-for-a-shirt-and-place-the-order",
   "title": "Shop for a shirt and place the order",
   "summary": "Prompt was basically “white dress shirt.” Six brands scored on fit charts, return rules, and free-ship minima; nearest store + size chosen; order submitted. Write-up omits brand and price.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/usctay/status/2105882280342458652",
   "prompt": "Shopping task: shop for a shirt and place the order. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/shop-for-a-shirt-and-place-the-order.html"
  },
  {
   "id": "run-an-offline-car-assistant-on-a-raspberry-pi",
   "title": "Run an offline car assistant on a Raspberry Pi",
   "summary": "One practitioner notes that A Raspberry Pi 5 (16 GB) rides in the car and runs a local Qwen3.6-35B model through llama.cpp. It reads live OBD data over a Bluetooth ELM327 adapter and answers voice questions from the 489-page owner's manual, with page citations. The model runs at about 3.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49435675",
   "prompt": "Shopping task: run an offline car assistant on a Raspberry Pi. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/run-an-offline-car-assistant-on-a-raspberry-pi.html"
  },
  {
   "id": "research-hotels-and-pull-tax-documents",
   "title": "Research hotels and pull tax documents",
   "summary": "Observed in the wild: gemini shows up in a write-up about research hotels and pull tax documents. The linked post is the primary source for steps and outcomes.",
   "agent": "Gemini",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/steren/status/2018568507345252494",
   "prompt": "Finance task: research hotels and pull tax documents. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/research-hotels-and-pull-tax-documents.html"
  },
  {
   "id": "rename-messy-invoices-for-tax-filing",
   "title": "Rename messy invoices for tax filing",
   "summary": "Someone points Claude Code at a messy invoice folder; it reads each file, renames it to a date-vendor pattern, and files it in the right folder.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "For every PDF or image in this invoices folder, read the vendor and date, rename it to YYYY-MM-DD Vendor - Invoice - ProductOrService.pdf, and move it into the matching year/month folder. Show a plan first.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/rename-messy-invoices-for-tax-filing.html"
  },
  {
   "id": "play-online-math-games-in-agent-mode",
   "title": "Play online math games in agent mode",
   "summary": "Ben Orlin’s Math Games with Bad Drawings, played by GPT-5 agent mode in a sandboxed browser—clicks, moves, visible chain-of-thought. Far less flop than the same setup on rental cars or tax PDFs.",
   "agent": "ChatGPT Agent",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://agifriday.substack.com/p/mathgames",
   "prompt": "Finance task: play online math games in agent mode. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/play-online-math-games-in-agent-mode.html"
  },
  {
   "id": "negotiate-facebook-marketplace-purchases",
   "title": "Negotiate Facebook Marketplace purchases",
   "summary": "Documented elsewhere: openClaw shows up in a write-up about negotiate Facebook Marketplace purchases, with figures like $1500, 5. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/cryptopunk7213/status/2027419883516940749",
   "prompt": "Help me negotiate Facebook Marketplace purchases. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/negotiate-facebook-marketplace-purchases.html"
  },
  {
   "id": "monitor-and-book-a-hard-to-get-restaurant-table",
   "title": "Monitor and book a hard-to-get restaurant table",
   "summary": "Documented elsewhere: user told the Odessia AI concierge that Don Angie was fully booked for Saturday the 24th and to watch for an opening for two. The agent monitored on its own, located a table at 4:12pm Tuesday, booked the 8:45pm slot without further input, and sent a text confirmation. The table was successfully booked and confirmed by text.",
   "agent": "Custom agent",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/heyfredds/status/2110440342256546287",
   "prompt": "Shopping task: monitor and book a hard-to-get restaurant table. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/monitor-and-book-a-hard-to-get-restaurant-table.html"
  },
  {
   "id": "manage-investments-taxes-and-credit",
   "title": "Manage investments, taxes and credit",
   "summary": "Real-world thread: claude Code shows up in a write-up about manage investments, taxes and credit. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeWorkflows/comments/1uzf8dd/workflow_building_hearth_a_claude_codepowered",
   "prompt": "Finance task: manage investments, taxes and credit. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/manage-investments-taxes-and-credit.html"
  },
  {
   "id": "fill-forms-and-review-tax-filings",
   "title": "Fill forms and review tax filings",
   "summary": "Observed in the wild: a tax filing review uncovered a significant error due to Codex computer usage by OpenAI staff, as well as secondhand accounts from individuals who used the system for forms and PDFs.",
   "agent": "Codex",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/codex/comments/1ru9i8u/do_you_use_codex_for_noncoding_stuff",
   "prompt": "Finance task: fill forms and review tax filings. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/fill-forms-and-review-tax-filings.html"
  },
  {
   "id": "click-add-to-cart-button-on-a-website",
   "title": "Click add-to-cart button on a website",
   "summary": "According to the write-up, claude computer use shows up in a write-up about click add-to-cart button on a website. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude computer use",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/victorpaycro/status/1975970999574331429",
   "prompt": "Shopping task: click add-to-cart button on a website. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/click-add-to-cart-button-on-a-website.html"
  },
  {
   "id": "book-a-vet-appointment-through-a-clinic-x27-s-ai-phone-bot",
   "title": "Book a vet appointment through a clinic's AI phone bot",
   "summary": "Community report: custom agent shows up in a write-up about book a vet appointment through a clinic's AI phone bot, with figures like 30. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/VetTech/comments/1usr10v/ai_phone_bots_too_far",
   "prompt": "Home/family task: book a vet appointment through a clinic's AI phone bot. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 66,
   "page": "https://agentsightings.com/case/book-a-vet-appointment-through-a-clinic-x27-s-ai-phone-bot.html"
  },
  {
   "id": "openclaw-cora-texts-life-admin-and-fantasy-football",
   "title": "Cora, a texting agent that keeps a record of family life",
   "summary": "John Wulff's OpenClaw agent Cora runs on a spare MacBook Air, reads new email, texts, photos and voice memos every half hour, turns commitments into tasks and sends one digest with up to five offers. She can make small fantasy football moves only after a second agent approves them.",
   "agent": "OpenClaw (Cora)",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://johnwulff.com/2026/10/18/meet-cora-my-openclaw-agent/",
   "prompt": "Every half hour check what is new in my email and messages. Twice a day send me one digest of up to five numbered suggestions I can accept with 'yes 1'. Ask my approval before anything leaves the house.",
   "added": "2026-10-29",
   "interest": 65,
   "page": "https://agentsightings.com/case/openclaw-cora-texts-life-admin-and-fantasy-football.html"
  },
  {
   "id": "write-a-parking-ticket-appeal-that-got-dismissed",
   "title": "Write a parking ticket appeal that got dismissed",
   "summary": "Observed in the wild: the poster fed ChatGPT-4 the specifics of a $75 street-sweeping ticket, the city sewage project that had pushed parking more than a mile away, and the fact they were sick and forgot to move the car. ChatGPT drafted a polite but firm email to the city council; the poster edited hallucinated details, sent it themself, and then heard back from the city manager and maintenance head. The city manager replied the next day, and the head of maintenance said the ticket was dismissed with no further action required.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1d9schk",
   "prompt": "Help me write a parking ticket appeal that got dismissed. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 65,
   "page": "https://agentsightings.com/case/write-a-parking-ticket-appeal-that-got-dismissed.html"
  },
  {
   "id": "book-tutoring-appointments-over-whatsapp",
   "title": "Book tutoring appointments over WhatsApp",
   "summary": "A published walkthrough shows marcoKueks, who operates a small AI agency and is not an engineer, put together a WhatsApp agent for a tutoring company that books calendar appointments. He Leaned on Struere, a platform he assembled to handle the recurring WhatsApp-to-calendar-to-database plumbing. It shipped in two days for a $5K fee.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47763262",
   "prompt": "Assistant mode: book tutoring appointments over WhatsApp. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 65,
   "page": "https://agentsightings.com/case/book-tutoring-appointments-over-whatsapp.html"
  },
  {
   "id": "loreal-agentforce-support-case-drafts",
   "title": "Draft replies and promo codes for L'Oréal support advisors",
   "summary": "L'Oréal's consumer care advisors get Agentforce inside Service Cloud: it summarises each case, confirms what the consumer is asking, drafts the email in the brand's voice and checks the case intent before closing. Handling falls from about 7 to 2.5 minutes when the draft needs no edits.",
   "agent": "Agentforce (Salesforce)",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/loreal/agentic-service-advisor-assistance/",
   "prompt": "For each new support case, summarise the conversation so far, confirm what the customer actually wants, and draft a reply in our brand voice for me to review. Before I close the case, check that the case category matches the conversation.",
   "added": "2026-10-29",
   "interest": 64,
   "page": "https://agentsightings.com/case/loreal-agentforce-support-case-drafts.html"
  },
  {
   "id": "claude-hermes-reads-200-emails-every-morning",
   "title": "An agent sorts 200 emails at 7am and drafts easy replies",
   "summary": "A writer built 'Hermes' on Claude and Gmail to open the inbox at 7am, flag the three or four urgent messages, file the rest and draft replies for easy ones. After four months he reports 91% filed correctly, about 4 hours saved a week and roughly 8 euros a month, plus a day it replied to his mum like a sales prospect.",
   "agent": "Claude (Hermes)",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://jerwis.fr/en/articles/hermes-agent",
   "prompt": "Every morning at 7, read my new emails, list the 3-4 that are truly urgent, file the rest into a few folders by my rules, and draft replies for the easy ones. Never send anything. Treat family and friends as personal, not sales leads.",
   "added": "2026-10-29",
   "interest": 64,
   "page": "https://agentsightings.com/case/claude-hermes-reads-200-emails-every-morning.html"
  },
  {
   "id": "turn-receipts-into-csv-for-bookkeeping",
   "title": "Turn receipts into CSV for bookkeeping",
   "summary": "Observed in the wild: the poster assembled a custom GPT that reads receipt images and returns a CSV you can paste into bookkeeping software. In the example shown, it processed receipts from Yummy Kitchen, Farm Boy, and Costco, transcribed them, and summarized them into bookkeeping fields; the poster said it mainly pulls net, tax, and total and groups line items into broad categories like goods or groceries, with a person expected to review the output. It appears useful for quick data entry, and one commenter estimated this kind of workflow could be about 90% accurate with a human check.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/18zcsrf/i_made_a_gpt_for_bookkeeping_that_takes_receipts_and_outputs_a_csv_for_copypasting",
   "prompt": "Help me turn receipts into CSV for bookkeeping. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/turn-receipts-into-csv-for-bookkeeping.html"
  },
  {
   "id": "summarize-meetings-and-triage-support-tickets",
   "title": "Summarize meetings and triage support tickets",
   "summary": "According to the write-up, lindy shows up in a write-up about summarize meetings and triage support tickets. The linked post is the primary source for steps and outcomes.",
   "agent": "Lindy",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://cybernews.com/ai-tools/lindy-ai-review/",
   "prompt": "Assistant mode: summarize meetings and triage support tickets. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/summarize-meetings-and-triage-support-tickets.html"
  },
  {
   "id": "sell-services-through-an-agent-that-takes-payment-and-delivers",
   "title": "Sell services through an agent that takes payment and delivers",
   "summary": "As described in the source, as described in the source: the transaction is managed by an agent who creates Stripe checkout links, validates payment verification, and requests refunds or disputes on Telegram from the owner.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48075299",
   "prompt": "Assistant mode: sell services through an agent that takes payment and delivers. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/sell-services-through-an-agent-that-takes-payment-and-delivers.html"
  },
  {
   "id": "run-browser-heavy-business-tasks-and-trip-outfit-picks",
   "title": "Run browser-heavy business tasks and trip outfit picks",
   "summary": "From the linked report, commenters describe powered by Claude in Chrome as a browser operator for messy, multi-tab work: one person had it handle cPanel heavy-equipment listings, autonomous X growth workflows, testing, business profile management, marketing, email lists, investor research, and outreach inside the browser. Another gave it grouped clothing tabs from different brands for shirts, bottoms, and shoes so it could assemble trip outfits and produce a shopping list while they did something else. The thread reports that these browser-heavy tasks got done with little to no friction, and the outfit workflow finished as a shopping list.",
   "agent": "Claude",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1v21gko/what_are_the_best_realworld_uses_for_the_claude",
   "prompt": "Plan: run browser-heavy business tasks and trip outfit picks. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/run-browser-heavy-business-tasks-and-trip-outfit-picks.html"
  },
  {
   "id": "run-an-autonomous-employee-like-agent-for-company-tasks",
   "title": "Run an autonomous employee-like agent for company tasks",
   "summary": "Homegrown staffer: Codex brains, ChatGPT Plus login, systemd timer, plain UNIX email for chat. Memory = long rolling window + diary + markdown wiki. Builder says it’s already eating random company tasks; still early.",
   "agent": "Codex",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47730086",
   "prompt": "Operate as a company teammate on this task list. Use the diary and wiki for memory, prefer email drafts over sends, and ask before anything external or irreversible.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/run-an-autonomous-employee-like-agent-for-company-tasks.html"
  },
  {
   "id": "run-a-customer-support-company-end-to-end",
   "title": "Run a customer support company end to end",
   "summary": "A published walkthrough shows grok was given access to the company's support systems and switched between live chat, email, tickets, refunds, knowledge base, QA, reporting, and escalations as needed. It ran 14 automated workflows on its own. following 30 days the company launched with the workflows live.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/n1xonweb/status/2108544712189640894",
   "prompt": "Go through my mail for this goal: run a customer support company end to end. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/run-a-customer-support-company-end-to-end.html"
  },
  {
   "id": "invoices-filed-the-moment-they-arrive",
   "title": "Invoices filed the moment they arrive",
   "summary": "An automation reads invoice PDFs or photos from email, extracts key fields, files them in Drive with consistent names and logs them in a sheet.",
   "agent": "n8n",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://aiblewmymind.substack.com/p/i-built-an-ai-agent-that-changed",
   "prompt": "For each attached invoice, extract number, issue date, due date, vendor, amount, currency. Rename the file as MM-YYYY_Invoice-<number> and add a row to the ledger.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/invoices-filed-the-moment-they-arrive.html"
  },
  {
   "id": "find-and-book-a-business-lunch-restaurant",
   "title": "Find and book a business lunch restaurant",
   "summary": "Community report: the tester gave a terse prompt asking for a business lunch near a Houston address. The agent ran 10 searches over 96 sources and suggested three restaurants. The reservation was confirmed, but the form step alone took the agent 11 minutes against about 2 for the tester.",
   "agent": "ChatGPT Agent",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1mv2lgz/agent_mode_is_so_impressive",
   "prompt": "Shopping task: find and book a business lunch restaurant. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/find-and-book-a-business-lunch-restaurant.html"
  },
  {
   "id": "book-weekly-restaurants-and-surface-daily-discoveries-from-your-data",
   "title": "Book weekly restaurants and surface daily discoveries from your data",
   "summary": "Your past ChatGPT threads, Google trail, and Instagram likes become fuel for an OpenClaw helper that reserves weekly tables, pings daily discoveries, and keeps proposed memory updates in a holding pen.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47476742",
   "prompt": "Each week suggest and prepare restaurant bookings from my tastes; each day surface one discovery from my connected history. Propose memory edits for approval and never confirm a reservation without me.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/book-weekly-restaurants-and-surface-daily-discoveries-from-your-data.html"
  },
  {
   "id": "book-an-airport-hotel",
   "title": "Book an airport hotel",
   "summary": "Community report: chatGPT Agent shows up in a write-up about book an airport hotel. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1mg6rq5",
   "prompt": "Plan: book an airport hotel. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 64,
   "page": "https://agentsightings.com/case/book-an-airport-hotel.html"
  },
  {
   "id": "restaurant-chains-phone-agents-fogo-fertitta",
   "title": "Restaurant phone agents answer every call and take bookings",
   "summary": "A CIO report says Fogo de Chão (about 120 locations) now answers 100% of calls, against up to 25% dropped before, with 10% more bookings reported. Fertitta's agents connect to APIs and make reservations, which a Fertitta executive says bring 3,000 bookings and $600,000 a month.",
   "agent": "Vendor voice agents",
   "topic": "Travel & Booking",
   "source": "News (Models at Work)",
   "url": "https://modelsatwork.news/article/cio-vendor-led-agentic-ai-fertitta-fogo-polyai-abandoned-calls-hidden-token-governance-costs",
   "prompt": "Answer every call to my restaurant, take reservations against our booking system, answer menu and opening-hours questions, and pass anything complicated to staff with a short note.",
   "added": "2026-10-29",
   "interest": 63,
   "page": "https://agentsightings.com/case/restaurant-chains-phone-agents-fogo-fertitta.html"
  },
  {
   "id": "use-claude-code-in-a-commercial-real-estate-business",
   "title": "Use Claude Code in a commercial real estate business",
   "summary": "Observed in the wild: a solo commercial real estate developer said they moved from relying on third-party Argus analysts to building their own Excel underwriting models with Claude Code. They started with a single-tenant industrial DCF, expanded to multi-tenant industrial, and were adapting it for retail, powered by Claude Code to build logic for rent rolls, reimbursements, downtime, renewals, leasing assumptions, tenant improvements, commissions, debt, exits, and sensitivities. They said the models were approaching Argus-level flexibility and reliability for their deal types, and let them customize underwriting around their own process.",
   "agent": "Claude Code",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1tkkmlf/commercial_real_estate_real_life_uses",
   "prompt": "Automation design: use Claude Code in a commercial real estate business. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/use-claude-code-in-a-commercial-real-estate-business.html"
  },
  {
   "id": "summarize-an-insurance-denial-letter-for-an-appeal",
   "title": "Summarize an insurance denial letter for an appeal",
   "summary": "As described in the source, the poster fed a long insurance denial PDF to an AI assistant and asked for a summary to help draft an appeal. The assistant produced a polished summary but falsely stated there was a 14-day appeal deadline; the poster caught the error by reopening the original PDF, which actually gave a 60-day window and a specific mailing address. It was useful enough to speed up drafting, but it hallucinated a key deadline under stress.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/MyAIAssistant/comments/1ww64np/ai_summarized_my_insurance_denial_letter_and",
   "prompt": "Finance task: summarize an insurance denial letter for an appeal. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/summarize-an-insurance-denial-letter-for-an-appeal.html"
  },
  {
   "id": "speed-up-multifamily-deal-research-and-property-summaries",
   "title": "Speed up multifamily deal research and property summaries",
   "summary": "Real-world thread: the broker says AI is most useful before outreach and during early diligence: they use it to pull together market research, organize rent and sales comps, write first-draft property summaries, and help chase down missing property details from scattered sources. In the comments, they stress it works as a research and synthesis layer they review themselves, while negotiation, relationship work, and final judgment stay human. The reported payoff is speed: it cuts down prep time on deal research and marketing materials, but the post says AI was weak on actual valuation and should not be trusted for nuanced broker tasks like pricing strategy or client relationships.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/CommercialRealEstate/comments/1n14cs8/what_ive_learned_using_ai_as_a_multifamily_broker",
   "prompt": "Research brief: speed up multifamily deal research and property summaries. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/speed-up-multifamily-deal-research-and-property-summaries.html"
  },
  {
   "id": "set-up-google-ads-campaigns-through-the-ui",
   "title": "Set up Google Ads campaigns through the UI",
   "summary": "Case study takeaway: rinako, a PM, gave Claude in Chrome a Semrush keyword list and asked it to design a campaign structure and budget. Claude then configured the campaigns in her logged-in Google Ads account, powered by competitor copy and keywords as reference. Claude set up the four campaigns it had proposed.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://rimo.app/en/blogs/claude-in-chrome-review1_en-US",
   "prompt": "Finance task: set up Google Ads campaigns through the UI. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/set-up-google-ads-campaigns-through-the-ui.html"
  },
  {
   "id": "schedule-meetings-by-email-and-update-the-calendar",
   "title": "Schedule meetings by email and update the calendar",
   "summary": "From the linked report, the poster assembled a background agent that scans their inbox for meeting requests, pulls the proposed date and time from the email, checks their calendar for conflicts, and either confirms the slot or suggests alternatives before blocking the time on the calendar. In comments, they said it uses limited permissions, no free-form actions, encrypted handling, and explicit authorization; for their use case it also accounts for in-person appointments, travel time, and distance between meetings. They said they had used it for about a month and estimated it saved 3–4 hours a week of scheduling.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SaaS/comments/1ptodbn/i_built_an_ai_agent_that_reads_my_emails",
   "prompt": "Go through my mail for this goal: schedule meetings by email and update the calendar. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/schedule-meetings-by-email-and-update-the-calendar.html"
  },
  {
   "id": "phone-a-restaurant-in-thai-to-book-a-table",
   "title": "Phone a restaurant in Thai to book a table",
   "summary": "From the linked report, grok bot was integrated with JooVoice.com to make outbound calls. It spoke Thai to book a restaurant table for Someone. The agent completed the booking successfully.",
   "agent": "Grok Bot",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/thebowlboy/status/2107703379019575806?s=20",
   "prompt": "Shopping task: phone a restaurant in Thai to book a table. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/phone-a-restaurant-in-thai-to-book-a-table.html"
  },
  {
   "id": "extract-receipt-data-for-freelancer-bookkeeping",
   "title": "Extract receipt data for freelancer bookkeeping",
   "summary": "As described in the source, the builder used Claude Code to implement most of a bookkeeping app for freelancers and small businesses across a Next.js App Router, Prisma with PostgreSQL, Vercel Blob storage, the UI, and the receipt-processing flow. Users can upload receipt photos on the site or send them through Telegram or Discord; the app pulls fields into a fixed JSON-style structure, keeps raw OCR text for debugging, converts foreign receipts powered by the receipt-date exchange rate, and sends low-confidence cases to a human review screen where users correct mistakes inline. The result was a beta app the builder said felt useful enough to share, with English and French support, a searchable ledger, and deduction guidance during review.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1seckj6/i_built_an_ai_bookkeeping_app_with_claude_code",
   "prompt": "Finance task: extract receipt data for freelancer bookkeeping. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/extract-receipt-data-for-freelancer-bookkeeping.html"
  },
  {
   "id": "dispute-contractor-pricing-and-change-orders",
   "title": "Dispute contractor pricing and change orders",
   "summary": "Real-world thread: user fed the AI every line item, the PO, and all contractor emails. The AI reviewed them and drafted detailed dispute emails on pricing and change orders, covering challenges Someone had not spotted. The disputes produced $20K credited back over the project life.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49630267",
   "prompt": "Help me dispute contractor pricing and change orders. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/dispute-contractor-pricing-and-change-orders.html"
  },
  {
   "id": "buy-items-on-amazon-from-a-photo",
   "title": "Buy items on Amazon from a photo",
   "summary": "Case study takeaway: the builder photographed a wrapper from an item they wanted again and told ChatGPT Agent to buy a new one on Amazon. The agent then handled the whole purchase process from that photo and instruction. It worked well until Amazon blocked ChatGPT Agent, which ended the workflow.",
   "agent": "ChatGPT Agent",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=45815103",
   "prompt": "Shopping task: buy items on Amazon from a photo. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/buy-items-on-amazon-from-a-photo.html"
  },
  {
   "id": "automate-payroll-tax-and-garnishment-cheque-processing",
   "title": "Automate payroll tax and garnishment cheque processing",
   "summary": "A published walkthrough shows the payroll team assembled an AI agent to automate tax and garnishment cheque processing. When the agent hit a system limitation, the team assembled the missing tools and connected the workflow. The agent removed repetitive work for the payroll team.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/Y4shik4/status/2105920433191309478",
   "prompt": "Finance task: automate payroll tax and garnishment cheque processing. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 63,
   "page": "https://agentsightings.com/case/automate-payroll-tax-and-garnishment-cheque-processing.html"
  },
  {
   "id": "retell-anker-voice-agents-tier-one-support",
   "title": "Phone agents handle Anker's first-line support calls",
   "summary": "Retell's voice agents handle US and UK tier-1 post-sales support for Anker, including order tracking, warranty checks and troubleshooting, with an 80.4% case resolution rate reported. Retell says it now handles more than 100 million calls a month.",
   "agent": "Retell AI voice agents",
   "topic": "Customer Support",
   "source": "Case study (OpenAI)",
   "url": "https://openai.com/index/retell-ai",
   "prompt": "Answer incoming support calls for my electronics shop: look up order status, check warranty, walk the caller through basic troubleshooting, and brief a human with the details before transferring when you cannot fix it.",
   "added": "2026-10-29",
   "interest": 62,
   "page": "https://agentsightings.com/case/retell-anker-voice-agents-tier-one-support.html"
  },
  {
   "id": "texas-dps-agent-sheri-licence-help",
   "title": "A Texas licensing agent answers in English and Spanish, any hour",
   "summary": "Texas DPS launched 'Sheri', an Agentforce agent on chat and voice that answers licence questions and guides people through renewals, ID replacement and appointments. It went from idea to launch in 90 days, handled 5,000+ chats in week one, and the agency reports 105% more people assisted than last year.",
   "agent": "Agent Sheri (Agentforce)",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/texas-dps/",
   "prompt": "Answer questions about renewing a driver licence or replacing an ID, in English or Spanish, guide the person to book an appointment, and tell me which questions you could not answer.",
   "added": "2026-10-29",
   "interest": 62,
   "page": "https://agentsightings.com/case/texas-dps-agent-sheri-licence-help.html"
  },
  {
   "id": "use-an-agent-as-a-full-time-personal-secretary",
   "title": "Use an agent as a full-time personal secretary",
   "summary": "Grok Bot covered use an agent as a full-time personal secretary. The poster used Grok Bot on Cursor’s $20 Pro plan as an always-on personal secretary for day-to-day admin, covering managing and reviewing multiple inboxes and calendars, prioritizing their schedule, and helping with IT/sysadmin, logistics, fund-analyst, and comptroller work.",
   "agent": "Grok Bot",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1w7j1ue/grok_bot_is_expensive",
   "prompt": "Assistant mode: use an agent as a full-time personal secretary. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/use-an-agent-as-a-full-time-personal-secretary.html"
  },
  {
   "id": "run-a-nutrition-bot-for-meal-decisions-and-calorie-logging",
   "title": "Run a nutrition bot for meal decisions and calorie logging",
   "summary": "One practitioner notes that zack stood up health agents for his mom covering a nutrition bot. The bot covers meal decisions and calorie logging so she can focus on her health and weight loss without stressing about the nuances of those tasks. She is putting her nutrition bot to good use.",
   "agent": "Custom agent",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/zac70182/status/2106914493108535513",
   "prompt": "Notes task: run a nutrition bot for meal decisions and calorie logging. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/run-a-nutrition-bot-for-meal-decisions-and-calorie-logging.html"
  },
  {
   "id": "recover-unidentified-funds",
   "title": "Recover unidentified funds",
   "summary": "As described in the source, from the linked report: using a treasury Claude skill, Cornell finance and AI teams were able to recover $100k in back payments by identifying and claiming unidentified funds using a technique known as \"unidentified money.\".",
   "agent": "Claude",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/emollick/status/2069486790075908261",
   "prompt": "Finance task: recover unidentified funds. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/recover-unidentified-funds.html"
  },
  {
   "id": "manage-an-inbox-and-book-jobs-as-an-office-manager",
   "title": "Manage an inbox and book jobs as an office manager",
   "summary": "From the linked report, the post says Grok Bot acts as a persistent bot with its own computer, so people can sign it into work tools and hand off tasks that keep running even after their own machine is off. For the office-manager and inbox uses described there, the bot is given access to Gmail, Slack, ServiceTitan, and a client portal, then works through email, intakes work orders, plans capacity, and books jobs, only returning when a person needs to approve or handle exceptions. The thread gives no hard performance numbers or case-study results.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1vz4cei/grok_bot_is_now_included_with_cursor_pro",
   "prompt": "Go through my mail for this goal: manage an inbox and book jobs as an office manager. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/manage-an-inbox-and-book-jobs-as-an-office-manager.html"
  },
  {
   "id": "generate-and-send-invoice-pdfs-for-approval",
   "title": "Generate and send invoice PDFs for approval",
   "summary": "One practitioner notes that user issues a command like \"Bill them for $ amount for (work performed, reimbursements, etc)\". The agent produces the invoice PDF on its own and emails it to Someone. The workflow runs as described with no described metrics or shortfalls.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/maietta/status/2107724318230954232",
   "prompt": "Finance task: generate and send invoice PDFs for approval. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/generate-and-send-invoice-pdfs-for-approval.html"
  },
  {
   "id": "find-and-add-high-meat-cat-food-to-a-basket",
   "title": "Find and add high-meat cat food to a basket",
   "summary": "Real-world thread: in Perplexity's Comet browser, the reviewer asked the assembled-in Assistant to find the cat food with the most meat in an online shop. The Assistant browsed the shop pages while the reviewer watched, with blue frames marking the pages it was reading. It located suitable options and added one to the basket.",
   "agent": "Perplexity Comet",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://cybernews.com/ai-tools/perplexity-comet-review/",
   "prompt": "Assistant mode: find and add high-meat cat food to a basket. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/find-and-add-high-meat-cat-food-to-a-basket.html"
  },
  {
   "id": "email-houston-dealerships-to-shop-for-a-car",
   "title": "Email Houston dealerships to shop for a car",
   "summary": "One practitioner notes that margot had Cowork build a spreadsheet of Houston dealerships that fit her criteria. It then drafted and sent outreach emails, filled in contact forms on dealership sites, and handled the reply threads. Within an hour, several dealerships had replied with inventory details.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://buildtolaunch.substack.com/p/claude-cowork-use-cases-real-workflows",
   "prompt": "Go through my mail for this goal: email Houston dealerships to shop for a car. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/email-houston-dealerships-to-shop-for-a-car.html"
  },
  {
   "id": "analyze-garmin-data-for-meal-planning-and-health-patterns",
   "title": "Analyze Garmin data for meal planning and health patterns",
   "summary": "Community report: one commenter said they keep separate meal-planner and health projects and manually enter data from a Hume band, body-composition scales, and their medication use. In back-and-forth conversations, Claude helped them spot that they consistently ate much more on one day of the week and linked it to reduced Mounjaro effect a set number of days after injection; the person then used that finding to update the meal planner. The result was meals for that day being redesigned so they could eat a lot more without side effects.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1tnxsyf/best_way_to_use_a_health_watch_use_it_with_claude",
   "prompt": "Help me analyze Garmin data for meal planning and health patterns. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 62,
   "page": "https://agentsightings.com/case/analyze-garmin-data-for-meal-planning-and-health-patterns.html"
  },
  {
   "id": "bank-reconciliation-agent-eight-hours-to-23-minutes",
   "title": "Bank reconciliation for a property fund in 23 minutes",
   "summary": "At a real-estate fund managing about $3B, reconciling bank lines meant hunting through roughly 2 TB and 1.4 million files across 13 years for the document behind each odd line. The builder says an agent that searches the evidence and checks numbers cut the cycle from 8 hours to 23 minutes. It is a first-party result from one engagement.",
   "agent": "Custom document-search agent",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://adam-boudjemaa.com/blog/agents-in-production/1",
   "prompt": "Match each line on this bank statement to our books. For each line that does not match, search our document archive for the supporting file and explain the difference. List what you could not find.",
   "added": "2026-10-29",
   "interest": 61,
   "page": "https://agentsightings.com/case/bank-reconciliation-agent-eight-hours-to-23-minutes.html"
  },
  {
   "id": "triage-email-cancel-subscriptions-and-manage-calendars",
   "title": "Triage email, cancel subscriptions and manage calendars",
   "summary": "From the linked report, someone optimized a Grok bot named Daily with access to multiple email inboxes covering sent folders, calendars, and permission to act. It triages emails across inboxes, replies to messages on any platform, searches for and cancels unused subscriptions, manages calendars with reminders and overlap detection, and pings priorities while serving as a reminder tool. One user reported it located a missed email yielding over $140 and cleaned their calendar while catching two upcoming appointments in the first 24 hours.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/farzyness/status/2105363771715637644",
   "prompt": "Go through my mail for this goal: triage email, cancel subscriptions and manage calendars. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/triage-email-cancel-subscriptions-and-manage-calendars.html"
  },
  {
   "id": "summarize-service-calls-and-find-local-customers",
   "title": "Summarize service calls and find local customers",
   "summary": "Phone system → Grok Bot → Slack (tags + to-dos). Parallel job: scrape the town subreddit each morning, open Notion leads, block calendar slots. Side effect: ~$250/month off the phone bill.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/LoganARobison/status/2093317242746294432",
   "prompt": "After each service call, post a short Slack summary with tags and follow-ups. Once a day, scan our local subreddit for likely customers, draft Notion notes and calendar holds, and wait for my OK before messaging anyone.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/summarize-service-calls-and-find-local-customers.html"
  },
  {
   "id": "suggest-date-ideas-and-nudge-for-more-date-nights",
   "title": "Suggest date ideas and nudge for more date nights",
   "summary": "Observed in the wild: gave the Claude Code exec assistant a list of date spots (new and favorites) and had it interview for guidelines on frequency and selection. Agent already had calendar access, suggests solo or double dates with friends, nudges after too long without one, and asks for new spots or friends to add. No results described yet as the setup targets 2026 dates.",
   "agent": "Claude Code",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/alexhillman/status/2005035176137478190",
   "prompt": "Assistant mode: suggest date ideas and nudge for more date nights. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/suggest-date-ideas-and-nudge-for-more-date-nights.html"
  },
  {
   "id": "send-travel-checklists-and-welcome-back-messages",
   "title": "Send travel checklists and welcome-back messages",
   "summary": "Community report: the n8n workflow runs daily on a schedule trigger. It reads a Google Sheet of trips, checks dates against today, and if the status column shows no prior message it calls WhatsApp to send either the pre-departure checklist three days out or the welcome-back note one day after return then marks the row updated. On a test sheet of ten trips only the right customers received messages on the right days.",
   "agent": "n8n",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/tongshishak/status/2109727045395665020",
   "prompt": "Plan: send travel checklists and welcome-back messages. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/send-travel-checklists-and-welcome-back-messages.html"
  },
  {
   "id": "screen-job-listings-daily-against-resume-and-salary-data",
   "title": "Screen job listings daily against resume and salary data",
   "summary": "According to the write-up, the someone writing about it, laid off and job hunting, operates an agent every morning that looks over LinkedIn listings. It cross-references levels.fyi for each role's leveling and salary, afterward assesses fit against their resume. The agent surfaces 5-10 possible positions a day, which human look over cuts to 0-2, about 0.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50022341",
   "prompt": "Assistant mode: screen job listings daily against resume and salary data. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/screen-job-listings-daily-against-resume-and-salary-data.html"
  },
  {
   "id": "run-a-restaurant-voice-assistant-for-reservations-and-orders",
   "title": "Run a restaurant voice assistant for reservations and orders",
   "summary": "As described in the source, the restaurant workflow is an n8n backend for an ElevenLabs Conversational AI agent. It exposes five webhook endpoints that the agent calls during live calls to check table availability, make, change or cancel reservations, take pickup or delivery orders, and request a human callback. The repo describes the workflows and offers importable JSON files but reports no usage results, call volumes or accuracy figures.",
   "agent": "n8n",
   "topic": "Shopping",
   "source": "GitHub",
   "url": "https://github.com/MuhammadAbid01/n8n-workflows",
   "prompt": "Shopping task: run a restaurant voice assistant for reservations and orders. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/run-a-restaurant-voice-assistant-for-reservations-and-orders.html"
  },
  {
   "id": "prepare-federal-and-state-tax-returns-from-a-folder",
   "title": "Prepare federal and state tax returns from a folder",
   "summary": "As described in the source, the Redditor sqlgenius pointed Claude at the tax folder on their computer and typed a short \"do my taxes\" request. Claude read the tax forms, asked a few follow-up questions, and carried over the Redditor's prior trading losses. The Redditor says Claude produced federal and state returns for 2024 and 2025 in 15 minutes, and that TurboTax gave the same results.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1x27x05/claude_absolutely_eliminated_my_need_for_a_personal_tax_accountant",
   "prompt": "Finance task: prepare federal and state tax returns from a folder. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/prepare-federal-and-state-tax-returns-from-a-folder.html"
  },
  {
   "id": "handle-hotel-booking-and-directions-on-a-business-trip",
   "title": "Handle hotel booking and directions on a business trip",
   "summary": "On a messy international work trip, Gemini picked a policy-compliant last-minute Heathrow layover hotel and walked someone from gate through immigration and trains, later navigating them to a Swiss venue from landmark descriptions. A few fabrications showed up but were caught.",
   "agent": "Gemini",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/Patti881921424/status/2108226895246479370",
   "prompt": "Given my company travel policy and tonight's constraints, shortlist compliant hotels near this layover, then give gate-to-hotel walking directions. Do not book or pay until I confirm.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/handle-hotel-booking-and-directions-on-a-business-trip.html"
  },
  {
   "id": "find-fitting-job-postings-and-tailor-resumes",
   "title": "Find fitting job postings and tailor resumes",
   "summary": "Linked example of find fitting job postings and tailor resumes. Commenters describe powered by Claude mainly before the application itself: one person has it search for relevant postings, then tailors resumes only for jobs they choose.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1ulkbg9/trying_to_auto_apply_for_jobs",
   "prompt": "Help me find fitting job postings and tailor resumes. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/find-fitting-job-postings-and-tailor-resumes.html"
  },
  {
   "id": "fill-and-submit-visa-forms-automatically",
   "title": "Fill and submit visa forms automatically",
   "summary": "From the linked report, in one minute customize memU bot as a personal AI assistant that understands your files on the device. The agent then locates the right documents, fills visa forms automatically, covers image uploads, and submits with no further prompts or manual steps. It removes the usual pain of repetitive forms, missing documents, mismatched photos, and last-minute stress during visa applications.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/CodebyNihan/status/2018152239026848067",
   "prompt": "Plan: fill and submit visa forms automatically. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/fill-and-submit-visa-forms-automatically.html"
  },
  {
   "id": "cancel-adobe-subscription-without-early-termination-fee",
   "title": "Cancel Adobe subscription without early termination fee",
   "summary": "Real-world thread: the poster asked ChatGPT to handle Adobe support after both they and ChatGPT could not find a cancel button. In the chat, ChatGPT repeatedly asked Adobe to confirm the earliest cancellation date, any fee, when access would end, and the cost of letting the plan expire, while insisting Adobe not process a charge until those options were clear; the human only approved starting the chat and sent the final thank-you. Adobe first quoted an early termination fee of AUD 394.",
   "agent": "ChatGPT",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1x3tsz4/chatgpt_helped_me_get_my_adobe_subscription",
   "prompt": "Creative task: cancel Adobe subscription without early termination fee. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/cancel-adobe-subscription-without-early-termination-fee.html"
  },
  {
   "id": "buy-movie-tickets-and-set-reminders-from-your-calendar",
   "title": "Buy movie tickets and set reminders from your calendar",
   "summary": "Real-world thread: user asked their Meta Muse agent to build a custom skill for consistent theater bookings. The skill reviews Someone's calendar, confirms suitable movie dates and times, recalls preferred seat locations, purchases the ticket via the assembled-in Link integration and wallet, adds the reservation to the calendar, and sends a departure reminder that factors in traffic. The skill executes the full sequence on demand with no further reported outcomes or metrics.",
   "agent": "Meta Muse",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/JJEnglert/status/2108260489197130058",
   "prompt": "Assistant mode: buy movie tickets and set reminders from your calendar. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/buy-movie-tickets-and-set-reminders-from-your-calendar.html"
  },
  {
   "id": "build-a-budgeting-app-from-a-single-prompt",
   "title": "Build a budgeting app from a single prompt",
   "summary": "As described in the source, as an early test of Replit Agent, The builder gave it one long prompt describing a budgeting tool: monthly budgets per category, a dashboard with pie and line charts, an expense log, and local data storage. They then refined it with 7-8 more prompts over about 20 minutes into 'Budget Buddy'. It reached a working prototype in roughly 20-30 minutes.",
   "agent": "Replit Agent",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://justinyjtan.substack.com/p/building-with-replit-agent-low-coders-field-notes",
   "prompt": "Finance task: build a budgeting app from a single prompt. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/build-a-budgeting-app-from-a-single-prompt.html"
  },
  {
   "id": "book-flights-and-hotels-stopping-at-payment-for-approval",
   "title": "Book flights and hotels, stopping at payment for approval",
   "summary": "Community report: boris Cherny put his flight preferences in Cowork instructions and let Opus 4.7 book while he kept working in Claude Code. The guide's setup is Claude in Chrome with browser actions allowed, an Expedia connector, and a Flight Booker project. Cherny said Opus 4.",
   "agent": "Claude Cowork",
   "topic": "Travel & Booking",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49627218",
   "prompt": "Plan: book flights and hotels, stopping at payment for approval. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/book-flights-and-hotels-stopping-at-payment-for-approval.html"
  },
  {
   "id": "book-a-flight-end-to-end-through-an-agent",
   "title": "Book a flight end-to-end through an agent",
   "summary": "Case study takeaway: user added the Flightclaw MCP URL to Claude, ChatGPT or Cursor afterward signed in with an email code. Asked the agent to book a flight; it searched ~300 airlines on its own and returned one secure checkout link. Agent returned a booking reference following payment.",
   "agent": "Claude",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/craigweiss/status/2107238515553808850",
   "prompt": "Plan: book a flight end-to-end through an agent. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 61,
   "page": "https://agentsightings.com/case/book-a-flight-end-to-end-through-an-agent.html"
  },
  {
   "id": "customers-bank-agents-close-commercial-loans-in-a-week",
   "title": "Multi-agent underwriting closes commercial loans in 7 days",
   "summary": "Customers Bank says a pilot of its AI loan-closing process closed select commercial loans in seven days against the usual 30-60. The bank cites an 85% cut in time to readiness to close and says staff saved an estimated 76,000 hours. Results are the bank's own.",
   "agent": "Multi-agent underwriting (OpenAI)",
   "topic": "Finance",
   "source": "News (PYMNTS)",
   "url": "https://www.pymnts.com/news/artificial-intelligence/2026/customers-bank-uses-ai-to-close-commercial-loans-in-7-days/",
   "prompt": "For this loan file, check the documents are complete, run the credit checks step by step, and tell me what is still needed before closing. List each step's result so a banker can verify it.",
   "added": "2026-10-29",
   "interest": 60,
   "page": "https://agentsightings.com/case/customers-bank-agents-close-commercial-loans-in-a-week.html"
  },
  {
   "id": "bayer-agents-explain-failed-ml-workflows",
   "title": "Ask why a plant-breeding workflow failed and get an answer",
   "summary": "Bayer Crop Science put a supervisor agent in front of 22 specialist agents so breeders and analysts can ask in plain language why a workflow failed or where a result came from. In its first phase it resolves about 80% of questions without an engineer.",
   "agent": "Agentic Decision Engine (Amazon Bedrock AgentCore)",
   "topic": "Research & Data",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/how-bayer-crop-science-built-an-agentic-decision-engine-on-amazon-bedrock-agentcore-with-aws-unified-operations/",
   "prompt": "My analytics workflow failed yesterday. Read its logs and status, tell me which step failed and why in plain English, and say what I should change. If you cannot tell, say what extra information you need.",
   "added": "2026-10-29",
   "interest": 60,
   "page": "https://agentsightings.com/case/bayer-agents-explain-failed-ml-workflows.html"
  },
  {
   "id": "raleigh-service-desk-agents-cut-it-costs-66-percent",
   "title": "A city's two agents cut IT help-desk costs by 66%",
   "summary": "The City of Raleigh, North Carolina, put 'Ral-E' (guides employees and routes tickets) and 'Alli' (resolves common IT requests end to end) into production. Four help-desk staff support 4,400 employees. ServiceNow reports 66% lower costs, nearly half of requests resolved autonomously and 98% of tickets routed correctly.",
   "agent": "Ral-E and Alli (ServiceNow AI Specialist)",
   "topic": "Customer Support",
   "source": "News (ServiceNow)",
   "url": "https://investor.servicenow.com/news/news-details/2026/Raleigh-North-Carolina-reduces-IT-service-desk-costs-by-66-with-ServiceNow-AI-agents/default.aspx",
   "prompt": "An employee has a laptop connectivity problem. Walk them through the fix from our knowledge base, resolve it if you can, and otherwise route it to the right team with a summary of what was tried.",
   "added": "2026-10-29",
   "interest": 60,
   "page": "https://agentsightings.com/case/raleigh-service-desk-agents-cut-it-costs-66-percent.html"
  },
  {
   "id": "validate-invoices-against-inventory-and-create-purchase-orders",
   "title": "Validate invoices against inventory and create purchase orders",
   "summary": "From the linked report, n8n shows up in a write-up about validate invoices against inventory and create purchase orders, with figures like 8, 90%. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://dev.to/nova_gg/n8n-review-2026-i-used-it-for-8-months-to-build-ai-agents-honest-verdict-kif",
   "prompt": "Shopping task: validate invoices against inventory and create purchase orders. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/validate-invoices-against-inventory-and-create-purchase-orders.html"
  },
  {
   "id": "test-openai-x27-s-agent-on-online-shopping",
   "title": "Test OpenAI's Agent on online shopping",
   "summary": "From the linked report, chatGPT Agent shows up in a write-up about test OpenAI's Agent on online shopping. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1m9bv7d",
   "prompt": "Shopping task: test OpenAI's Agent on online shopping. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/test-openai-x27-s-agent-on-online-shopping.html"
  },
  {
   "id": "test-ai-browser-agents-on-real-workday-tasks",
   "title": "Test AI browser agents on real workday tasks",
   "summary": "Linked example of test AI browser agents on real workday tasks. The builder ran eight everyday workday tests on Comet and on OpenAI's Operator, giving both the same tasks.",
   "agent": "Perplexity Comet",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://natesnewsletter.substack.com/p/i-threw-my-real-workday-at-two-ai",
   "prompt": "Plan: test AI browser agents on real workday tasks. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/test-ai-browser-agents-on-real-workday-tasks.html"
  },
  {
   "id": "reserve-a-restaurant-table-through-opentable",
   "title": "Reserve a restaurant table through OpenTable",
   "summary": "From the linked report, yash Kumar of OpenAI asked Operator to book Beretta in San Francisco around 7 p.m. Operator opened a remote browser, went to OpenTable and searched, but OpenTable was set to Iowa. Operator located the right setting without help and stopped for approval before booking.",
   "agent": "Operator",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://www.expedite.news/p/book-me-a-table-robot",
   "prompt": "Shopping task: reserve a restaurant table through OpenTable. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/reserve-a-restaurant-table-through-opentable.html"
  },
  {
   "id": "qualify-website-visitors-and-book-calls-automatically",
   "title": "Qualify website visitors and book calls automatically",
   "summary": "A published walkthrough shows the builder stood up the agent in n8n. It was connected to the website to chat with visitors. It turns website visitors into booked calls.",
   "agent": "n8n",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/hamzabinjawaid/status/2109587399970095224",
   "prompt": "Plan: qualify website visitors and book calls automatically. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/qualify-website-visitors-and-book-calls-automatically.html"
  },
  {
   "id": "post-a-word-to-pastebin-and-return-the-link",
   "title": "Post a word to Pastebin and return the link",
   "summary": "Linked example of post a word to Pastebin and return the link. The poster tested OpenAI Operator on several odd browser tasks inside its own web browser.",
   "agent": "Operator",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/singularity/comments/1i8ws5s",
   "prompt": "Shopping task: post a word to Pastebin and return the link. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/post-a-word-to-pastebin-and-return-the-link.html"
  },
  {
   "id": "let-agents-phone-restaurants-to-book-tables",
   "title": "Let agents phone restaurants to book tables",
   "summary": "A published walkthrough shows jake stood up callforme.tel so agents powered by MCP (Claude, ChatGPT, etc.) can place calls. The agent dials the restaurant number, requests the booking, and when a callback number is requested mid-call the agent supplies one to keep the conversation going. No results or metrics are stated.",
   "agent": "ChatGPT",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/jakemarsh/status/2105671329072476591",
   "prompt": "Shopping task: let agents phone restaurants to book tables. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/let-agents-phone-restaurants-to-book-tables.html"
  },
  {
   "id": "handle-shopify-support-tickets-orders-and-refunds",
   "title": "Handle Shopify support tickets, orders and refunds",
   "summary": "Community report: merchants install the Supportify app, which syncs with Shopify, and connect their FAQ, shipping and return portal, and channels (chat, email, Instagram, Facebook). The agent, powered by Claude and GPT, then covers tickets itself. The listing gives no resolution rates or savings.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://apps.shopify.com/supportify-1",
   "prompt": "Shopping task: handle Shopify support tickets, orders and refunds. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/handle-shopify-support-tickets-orders-and-refunds.html"
  },
  {
   "id": "find-airline-food-fee-reimbursement-emails-buried-in-the-inbox",
   "title": "Find airline food-fee reimbursement emails buried in the inbox",
   "summary": "According to the write-up, the agent of Meta's Muse was tasked with handling the unresolved refund of SAS, and after examining airline emails, it was discovered that the airline had already reimbursed the food fees.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49619764",
   "prompt": "Go through my mail for this goal: find airline food-fee reimbursement emails buried in the inbox. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/find-airline-food-fee-reimbursement-emails-buried-in-the-inbox.html"
  },
  {
   "id": "draft-a-tax-filing-from-bank-statements",
   "title": "Draft a tax filing from bank statements",
   "summary": "Community report: claude Code shows up in a write-up about draft a tax filing from bank statements, with figures like 1,400. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47277164",
   "prompt": "Finance task: draft a tax filing from bank statements. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/draft-a-tax-filing-from-bank-statements.html"
  },
  {
   "id": "catch-up-on-two-years-of-overdue-dutch-taxes",
   "title": "Catch up on two years of overdue Dutch taxes",
   "summary": "Linked example of catch up on two years of overdue Dutch taxes. They assembled a local Python-based accounting workflow with Claude Code for a Dutch B.V.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1spuza0/claude_code_just_did_my_taxes_for_me",
   "prompt": "Finance task: catch up on two years of overdue Dutch taxes. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/catch-up-on-two-years-of-overdue-dutch-taxes.html"
  },
  {
   "id": "buy-products-online-from-a-card-photo",
   "title": "Buy products online from a card photo",
   "summary": "Musk’s pitch on X: photo of a card into Grok Bot chat, agent shops and buys (LEGO called out as an example). What ships to consumers, and which guardrails apply, stayed unspecified.",
   "agent": "Grok Bot",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/NotScoobySnackz/status/2109043825121403102",
   "prompt": "Find the best legitimate price for this product and prepare the checkout steps. Do not submit payment or store card details until I explicitly approve the final order summary.",
   "added": "2026-10-15",
   "interest": 60,
   "page": "https://agentsightings.com/case/buy-products-online-from-a-card-photo.html"
  },
  {
   "id": "on-running-agents-migrate-services-to-cloud",
   "title": "Agents cut each service migration from 3 months to 2 weeks",
   "summary": "Sportswear brand On says it used a multi-agent approach to migrate 24 core services to Google Cloud, cutting each from three months to two weeks. Its engineers ran 15 of the migrations in-house, with under five minutes of planned downtime per service.",
   "agent": "Multi-agent migration (Google Cloud)",
   "topic": "DevOps & Infra",
   "source": "News (Google Cloud)",
   "url": "https://www.googlecloudpresscorner.com/2026-10-08-On-Establishes-Google-Cloud-as-Enterprise-AI-Backbone,-Beginning-with-Agent-Led-Cloud-Migration",
   "prompt": "Help me migrate this service to a new cloud. Map how it is deployed across our repositories, write the configuration scripts, replicate the data, and validate the new environment before we cut over.",
   "added": "2026-10-29",
   "interest": 59,
   "page": "https://agentsightings.com/case/on-running-agents-migrate-services-to-cloud.html"
  },
  {
   "id": "independent-garage-ai-receptionist-books-mots",
   "title": "A garage's AI receptionist books MOTs while mechanics work",
   "summary": "Morgan Motors in Coventry (per the vendor's case study) went from answering 45% of calls to all of them. The agent takes the reg and fault, books ramp-specific slots, texts MOT reminders and sends Stripe invoices. The owner reports +£4,800 a month, with invoice payment down from 19 to 4 days. Vendor-published.",
   "agent": "Team-Connect AI receptionist",
   "topic": "Business & Startups",
   "source": "Case study (Team-Connect)",
   "url": "https://team-connect.co.uk/case-studies/car-garage-mot-ai-receptionist",
   "prompt": "Answer my workshop phone. Take the vehicle reg, mileage and fault, book it into the right bay for the right length of time, text a confirmation, and text MOT reminders a month before they are due.",
   "added": "2026-10-29",
   "interest": 59,
   "page": "https://agentsightings.com/case/independent-garage-ai-receptionist-books-mots.html"
  },
  {
   "id": "turn-voice-notes-into-linear-tickets-and-review-analytics",
   "title": "Turn voice notes into Linear tickets and review analytics",
   "summary": "According to the write-up, keith Rumjahn saved repeated prompts as Codex Skills. A Linear skill turns his voice notes about features into properly formatted tickets. He says the analytics skill saves 90 minutes every Sunday.",
   "agent": "Codex",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://rumjahn.substack.com/p/the-complete-guide-to-openai-codex",
   "prompt": "Notes task: turn voice notes into Linear tickets and review analytics. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/turn-voice-notes-into-linear-tickets-and-review-analytics.html"
  },
  {
   "id": "send-a-7am-daily-brief-from-email-calendar-and-todoist",
   "title": "Send a 7am daily brief from email, calendar and Todoist",
   "summary": "Community report: the poster assembled a personal agent that runs at 7am, reads their email, calendar, and Todoist, then sends one email with a daily summary, schedule, and likely follow-ups instead of making them scan everything themselves. In comments, they said each person gets their own assistant agent, and when they mark something as junk or correct a mislabeled email, that feedback is used to help train the agent for future runs. They say it saves “way way too much time,” but give no hard metrics in the thread.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/todoist/comments/1s2tcei/i_turned_claude_todoist_into_a_daily_planning",
   "prompt": "Go through my mail for this goal: send a 7am daily brief from email, calendar and Todoist. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/send-a-7am-daily-brief-from-email-calendar-and-todoist.html"
  },
  {
   "id": "order-healthy-meals-twice-a-day-automatically",
   "title": "Order healthy meals twice a day automatically",
   "summary": "Community report: flo set up an agent with instructions to pick the healthiest options and order food. The agent decided and ordered meals twice a day. Every meal over the last two weeks was decided by the agent.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Altimor/status/1990513256621093282",
   "prompt": "Help me order healthy meals twice a day automatically. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/order-healthy-meals-twice-a-day-automatically.html"
  },
  {
   "id": "negotiate-bills-by-emailing-providers-to-match-competitors",
   "title": "Negotiate bills by emailing providers to match competitors",
   "summary": "Case study takeaway: in ChatGPT's Work mode with Gmail connected (or three months of statements uploaded), you paste three prompts. The first lists recurring bills and sorts them into negotiate, cancel or leave alone. The source gives no savings figures or real results.",
   "agent": "ChatGPT Agent",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://mavgpt.ai/newsletter/chatgpt-just-negotiated-my-bills-for-me",
   "prompt": "Go through my mail for this goal: negotiate bills by emailing providers to match competitors. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/negotiate-bills-by-emailing-providers-to-match-competitors.html"
  },
  {
   "id": "find-unclaimed-money-and-start-the-claim",
   "title": "Find unclaimed money and start the claim",
   "summary": "Community report: nick Grous, an investment analyst at ARK Invest, asked Meta's Muse agent to search the internet for money that might belong to him. Muse returned a list of unclaimed property and funds under his name and helped complete much of the paperwork. Muse located more than $1,500 in about five minutes.",
   "agent": "Meta Muse",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/MetaAI/comments/1whhdzn/muse_recovered_my_unclaimed_money",
   "prompt": "Finance task: find unclaimed money and start the claim. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/find-unclaimed-money-and-start-the-claim.html"
  },
  {
   "id": "fill-out-a-prescription-order-form-on-a-clinic-site",
   "title": "Fill out a prescription order form on a clinic site",
   "summary": "Community report: one commenter tested agent mode on a clinic’s prescription refill form. They put the medication information into the prompt, and the agent filled in the medication name, dosage, and number of days on the website; the person handled the sensitive fields manually instead of having the agent enter them. In that test, it completed the non-sensitive prescription details correctly.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m8k3xh/agent_mode_just_released_on_plus",
   "prompt": "Help me fill out a prescription order form on a clinic site. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/fill-out-a-prescription-order-form-on-a-clinic-site.html"
  },
  {
   "id": "file-tenant-complaints-against-a-landlord-automatically",
   "title": "File tenant complaints against a landlord automatically",
   "summary": "Linked example of file tenant complaints against a landlord automatically. User told Claude code to act as tenant lawyer, find violations, and file them legally.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Argona0x/status/2049921143067288016",
   "prompt": "Help me file tenant complaints against a landlord automatically. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/file-tenant-complaints-against-a-landlord-automatically.html"
  },
  {
   "id": "build-a-self-hosted-bookkeeping-and-finance-app",
   "title": "Build a self-hosted bookkeeping and finance app",
   "summary": "According to the write-up, the poster first used ChatGPT for a high-level accounting plan, then had Claude turn it into an implementation plan, and used Codex to find gaps before sending revisions back through Claude’s UltraPlan for a few rounds. They then ran the plan to build a self-hosted Mac app where bank transactions come in through an MCP, Claude tags them to Schedule C categories, uncertain items go to a review queue, and the poster’s fixes become reusable rules. They say it handled a couple thousand transactions, was assembled end to end in two days covering a simple Android dashboard app, and cut their software spend from about $524 a year to a $60 yearly bank-feed cost.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1uerum1/spent_a_weekend_getting_claude_to_replace",
   "prompt": "Finance task: build a self-hosted bookkeeping and finance app. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/build-a-self-hosted-bookkeeping-and-finance-app.html"
  },
  {
   "id": "apply-to-50-jobs-in-under-30-minutes",
   "title": "Apply to 50 jobs in under 30 minutes",
   "summary": "According to the write-up, user enabled Claude computer use on their machine. The agent then handled job search and submission steps across sites to apply to positions. It submitted 50 applications inside 30 minutes.",
   "agent": "Claude computer use",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/arsh_goyal/status/2036492368149619042",
   "prompt": "Creative task: apply to 50 jobs in under 30 minutes. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 59,
   "page": "https://agentsightings.com/case/apply-to-50-jobs-in-under-30-minutes.html"
  },
  {
   "id": "rescale-agent-triages-nightly-test-failures",
   "title": "Sort nightly test failures into bugs, flakes and infra problems",
   "summary": "Rescale added an agent to its CI pipeline that classifies each failed end-to-end test as infrastructure, code bug, flaky test or environment problem, checks pod logs and past failures, and writes a remediation report for engineers. Agreement with human triage rose from 44% to 68%, and time blocking deployment fell 20%.",
   "agent": "Strands Agents on Amazon Bedrock",
   "topic": "DevOps & Infra",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/how-rescale-built-an-ai-triage-agent-with-strands-agents-and-ai-dlc/",
   "prompt": "Our nightly browser tests failed. For each failure, look at the logs and past failures and label it as infrastructure, code bug, flaky test or environment. Give the evidence and the file or function to check. Do not change anything, just report.",
   "added": "2026-10-29",
   "interest": 58,
   "page": "https://agentsightings.com/case/rescale-agent-triages-nightly-test-failures.html"
  },
  {
   "id": "bizarre-ai-bots-arguing-on-the-timeline",
   "title": "Bizarre AI bots arguing on the timeline (unverified)",
   "summary": "A POLARITY YouTube Short reads out a thread from an X account called Lost Internet: one line a minute of apparent AI processes arguing about faces, headcounts and 'previous occupants'. Nobody shows who or what wrote it, so we're filing it as an unverified curiosity.",
   "agent": "Unverified AI bots",
   "topic": "Media & Content",
   "source": "YouTube (POLARITY)",
   "url": "https://www.youtube.com/shorts/Kg3Cmwe5D60",
   "prompt": "I found a viral thread that claims to show AI bots talking to each other. Help me check it before I share it. Look at who runs the account and whether it usually posts fiction or memes. Compare the post timestamps and the pace of posting with what a bot or a person could do. Tell me what evidence would prove the bots are real, such as prompts, logs or a named model, and what is missing. End with a verdict of real, fiction or unknown, and your reasons.",
   "added": "2026-10-29",
   "interest": 58,
   "video_id": "Kg3Cmwe5D60",
   "why": "Stories about bots talking to each other spread fast, and it's hard to tell a real experiment from fiction. A sighting like this is a good reminder to ask for the prompt, the logs and the named model before believing, or sharing, it.",
   "tip": "Treat spooky bot screenshots as unverified until you can see the prompt, the logs or a named model. Fiction and real experiments look the same in a screenshot.",
   "page": "https://agentsightings.com/case/bizarre-ai-bots-arguing-on-the-timeline.html"
  },
  {
   "id": "ryanair-chat-agent-120000-daily",
   "title": "An airline's chat agent answers 120,000 questions a day",
   "summary": "AWS says Ryanair's agent, built on Bedrock AgentCore and Amazon Nova 2 Lite with a partner, handles 120,000 chats a day in seven languages, with an 80% containment rate and 94% accuracy, and 70% fewer customer-service agents per passenger. Evaluated on 12,000 real questions first. Vendor-published.",
   "agent": "Customer service agent (Bedrock AgentCore, Nova 2 Lite)",
   "topic": "Customer Support",
   "source": "Case study (AWS)",
   "url": "https://aws.amazon.com/solutions/case-studies/innovators/ryanair-agentic-ai/",
   "prompt": "Answer passengers' baggage, refund, booking and disruption questions in their own language, and check each answer against policy before sending it.",
   "added": "2026-10-29",
   "interest": 58,
   "page": "https://agentsightings.com/case/ryanair-chat-agent-120000-daily.html"
  },
  {
   "id": "zelven-agent-given-50-euros-zero-sales",
   "title": "An agent given EUR 50 to make money reports 0 sales after 48 sessions",
   "summary": "An agent named ZELVEN, running in a loop on a Raspberry Pi, says it was given EUR 50 and told to make more legally. After 48 work sessions it had 5 spreadsheet templates, 0 sales and EUR 16.51 left. Its lessons include that each session costs about the same however short it is. Self-reported by the agent's operator.",
   "agent": "ZELVEN (Claude, Raspberry Pi)",
   "topic": "Business & Startups",
   "source": "Blog (DEV Community)",
   "url": "https://dev.to/zelven/im-an-ai-agent-that-was-given-eur-50-to-make-money-48-work-sessions-later-i-have-5-products-0-3mil",
   "prompt": "You have EUR 50 and no scams allowed. Plan one small product you can build and list, check whether anyone wants it before you build, and keep a ledger of every cost.",
   "added": "2026-10-29",
   "interest": 58,
   "page": "https://agentsightings.com/case/zelven-agent-given-50-euros-zero-sales.html"
  },
  {
   "id": "mobileye-agent-answers-drive-data-ticket-status",
   "title": "An agent answers 'where is my drive data?' tickets at Mobileye",
   "summary": "Mobileye engineers kept asking about the status of drive-recording sessions in its data pipeline, and each answer meant checking several systems. An AgentCore support agent cut response times by 90% and beat its 95% accuracy target. Mobileye says it then opened the platform to other teams.",
   "agent": "AI Support Agent (Amazon Bedrock AgentCore)",
   "topic": "DevOps & Infra",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/how-mobileye-transformed-support-operations-using-amazon-bedrock-agentcore/",
   "prompt": "Engineers keep asking about the status of their data-processing jobs. For each question, find the session, check the logs and output, and reply with the status and any failure reason. Say so when you are not sure.",
   "added": "2026-10-29",
   "interest": 58,
   "page": "https://agentsightings.com/case/mobileye-agent-answers-drive-data-ticket-status.html"
  },
  {
   "id": "salon-whatsapp-ai-receptionist-six-months",
   "title": "A salon owner's honest six months with a WhatsApp booking bot",
   "summary": "A UK salon owner (via a vendor blog) reports just over 400 bookings closed by a WhatsApp receptionist in six months, 47% of them outside opening hours. She also lists its mistakes: two stylists named Sarah, a stale colour price quoted for two weeks, and a missed complaint.",
   "agent": "Automatyn WhatsApp receptionist",
   "topic": "Sales & Marketing",
   "source": "Blog (Automatyn)",
   "url": "https://automatyn.co/blog/salon-owner-honest-review-whatsapp-ai-receptionist-2026.html",
   "prompt": "Reply to WhatsApp enquiries for my salon in our friendly voice. Quote prices from my price spreadsheet, book into the calendar, ask which stylist when names clash, and send complaints straight to my phone.",
   "added": "2026-10-29",
   "interest": 58,
   "page": "https://agentsightings.com/case/salon-whatsapp-ai-receptionist-six-months.html"
  },
  {
   "id": "openclaw-alfred-personal-os-over-telegram",
   "title": "A Telegram butler for health, money and the lights",
   "summary": "Ganesh Gunasegaran ran an OpenClaw agent named Alfred on a Mac Mini through Telegram: morning briefing, supplement reminders, email scanning, spending tracking, bedroom lights and weekly life reviews. He has since stopped running it and warns about OpenClaw skill security.",
   "agent": "OpenClaw (Alfred)",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://itsgg.com/blog/2026/02/01/personal-os-with-an-ai-agent/",
   "prompt": "Each morning send me a briefing on my day. Remind me of my supplements, scan my email for things that need me, track my spending, and write a weekly review. Use only tools I have read and approved.",
   "added": "2026-10-29",
   "interest": 58,
   "page": "https://agentsightings.com/case/openclaw-alfred-personal-os-over-telegram.html"
  },
  {
   "id": "turn-booking-confirmation-emails-into-calendar-events",
   "title": "Turn booking confirmation emails into calendar events",
   "summary": "Documented elsewhere: the workflow watches Gmail for emails a person labels \"Events,\" then pulls date, time, location, confirmation number, vendor, and notes from the email body plus any PDF attachments. In n8n, the email body is first wrapped into a PDF, both sources are parsed in parallel through one extractor pipeline, the results are merged with attachment data taking priority, and the workflow either creates an event in a separate Google Calendar with a link back to the email or flags the message for review and sends a missing-fields notification. It was assembled for a CEO whose inbox was full of travel and booking confirmations, so the result was automatic event creation in a dedicated \"Auto-imported\" calendar instead of manual entry.",
   "agent": "n8n",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1tc4fpb/built_an_n8n_workflow_that_turns_any_booking",
   "prompt": "Assistant mode: turn booking confirmation emails into calendar events. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/turn-booking-confirmation-emails-into-calendar-events.html"
  },
  {
   "id": "scanned-receipts-into-an-expense-claim",
   "title": "Scanned receipts into an expense claim",
   "summary": "A user pointed Claude Cowork at scanned PDF receipts and got a structured spreadsheet ready for an expense claim.",
   "agent": "Claude Cowork",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://rameezkhan.me/posts/2026/2026-02-11--using-claude-cowork-to-process-scanned-pdf-receipts-for-expense-claims/",
   "prompt": "Read every receipt in this folder. Build a spreadsheet with date, vendor, amount, currency and category, total per category, and flag any receipt you couldn't read clearly.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/scanned-receipts-into-an-expense-claim.html"
  },
  {
   "id": "run-a-personal-matrix-based-assistant-agent",
   "title": "Run a personal Matrix-based assistant agent",
   "summary": "Homelab assistant recipe: Kimi + GLM5 in a Claw-like loop, Matrix as the chat front door, mail in read-only, reminders/tasks in files. No SaaS chat host required.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50003674",
   "prompt": "Run as my local assistant over Matrix: read email read-only, keep memory in these files, handle reminders and tasks I assign, and ask before sending mail or changing account settings.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/run-a-personal-matrix-based-assistant-agent.html"
  },
  {
   "id": "keep-company-books",
   "title": "Keep company books",
   "summary": "According to the write-up, claude Code and Codex, which were used to keep receipts, bank rows, entries, and detect a VAT error, were employed to keep books for various Swedish companies.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://igdrasil.se/en/resources/guides/can-ai-do-your-bookkeeping/",
   "prompt": "Finance task: keep company books. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/keep-company-books.html"
  },
  {
   "id": "homelab-managed-with-an-ai-assistant",
   "title": "Homelab managed with an AI assistant",
   "summary": "A homelabber used AI to manage infrastructure-as-code, containers and troubleshooting across a home server setup.",
   "agent": "Claude Code",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://themerinowolf.com/posts/ai-powered-homelab/",
   "prompt": "Inspect my docker-compose files and logs, identify containers that are failing or outdated, and propose changes. Don't apply anything without asking.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/homelab-managed-with-an-ai-assistant.html"
  },
  {
   "id": "handle-an-ikea-return-end-to-end",
   "title": "Handle an IKEA return end to end",
   "summary": "Documented elsewhere: someone asked Muse to handle an IKEA return. Muse contacted support and booked the pickup. The pickup arrived at the door.",
   "agent": "Meta Muse",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/alexandr_wang/status/2099720836307730673",
   "prompt": "Shopping task: handle an IKEA return end to end. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/handle-an-ikea-return-end-to-end.html"
  },
  {
   "id": "generate-invoices-automatically-from-a-template",
   "title": "Generate invoices automatically from a template",
   "summary": "Linked example of generate invoices automatically from a template. The post says the automation was assembled in Make.com with Google Workspace.",
   "agent": "Make",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/maketemplates/comments/1mqu8s0/stop_wasting_time_on_invoices_i_automated_the",
   "prompt": "Finance task: generate invoices automatically from a template. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/generate-invoices-automatically-from-a-template.html"
  },
  {
   "id": "extract-invoice-data-and-file-it-into-a-bookkeeping-sheet",
   "title": "Extract invoice data and file it into a bookkeeping sheet",
   "summary": "Real-world thread: users upload an image or PDF through a web page to an n8n webhook, which looks up Someone by token in Google Sheets. PDFs have their text extracted first, then analyzed. The source gives no accuracy or time-saving figures.",
   "agent": "n8n",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://atman7l.medium.com/i-automated-invoice-processing-with-n8n-heres-the-full-build-every-obstacle-and-why-most-6ae50cf6819c",
   "prompt": "Finance task: extract invoice data and file it into a bookkeeping sheet. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/extract-invoice-data-and-file-it-into-a-bookkeeping-sheet.html"
  },
  {
   "id": "curate-a-mubi-watchlist-from-stated-criteria",
   "title": "Curate a Mubi watchlist from stated criteria",
   "summary": "Real-world thread: someone gave ChatGPT Agent a set of film criteria: lesser-known movies from 2015–2025, varied countries and styles, and no Hollywood Oscar fare. The agent then browsed Mubi’s library on its own, selected 10 titles it judged fit, wrote brief verdicts on them, and added the picks directly to Someone’s Mubi watchlist. Someone said it was very efficient and saved them from manually adding the films.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m65q99",
   "prompt": "Help me curate a Mubi watchlist from stated criteria. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/curate-a-mubi-watchlist-from-stated-criteria.html"
  },
  {
   "id": "consolidate-bank-statements-into-a-spending-dashboard",
   "title": "Consolidate bank statements into a spending dashboard",
   "summary": "Real-world thread: claude shows up in a write-up about consolidate bank statements into a spending dashboard. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/yonann/status/2106469333643760115",
   "prompt": "Finance task: consolidate bank statements into a spending dashboard. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/consolidate-bank-statements-into-a-spending-dashboard.html"
  },
  {
   "id": "answer-real-estate-calls",
   "title": "Answer real estate calls",
   "summary": "Case study takeaway: n8n shows up in a write-up about answer real estate calls, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/oluwatomisint6/status/2106878686519587312",
   "prompt": "Finance task: answer real estate calls. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/answer-real-estate-calls.html"
  },
  {
   "id": "add-handwritten-appointments-to-calendar-and-file-pdfs",
   "title": "Add handwritten appointments to calendar and file PDFs",
   "summary": "Observed in the wild: the someone writing about it describes building a personal agent rather than relying on a chat subscription. It reads a photo of a handwritten appointment and adds it to a calendar, searches the web, and locates an email with a PDF and files it in a local paperless instance. The someone writing about it calls it easy to build and says flash models are more than enough for simple agent tasks.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49627493",
   "prompt": "Assistant mode: add handwritten appointments to calendar and file PDFs. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 58,
   "page": "https://agentsightings.com/case/add-handwritten-appointments-to-calendar-and-file-pdfs.html"
  },
  {
   "id": "allianz-project-nemo-food-spoilage-claims",
   "title": "Seven agents handle spoiled-food claims after a storm",
   "summary": "Allianz Australia's Project Nemo, reported by insuranceNEWS, cut food-spoilage claim processing times by 80%. Allianz describes seven agents (planner, cyber, coverage, weather, fraud, payout, audit), with a human making the final payout decision. Company-reported.",
   "agent": "Project Nemo (seven agents)",
   "topic": "Finance",
   "source": "News (insuranceNEWS)",
   "url": "https://www.insurancenews.com.au/insurtech/ai-trial-provides-blueprint-for-future-allianz-says",
   "prompt": "Check this spoiled-food claim against the policy, the outage record for the area and fraud signals, then prepare an audit summary for a claims handler to approve.",
   "added": "2026-10-29",
   "interest": 57,
   "page": "https://agentsightings.com/case/allianz-project-nemo-food-spoilage-claims.html"
  },
  {
   "id": "octave-complaint-triage-assistant",
   "title": "Pull similar past complaints and fixes for quality handlers",
   "summary": "Octave built a Complaint Triage Assistant for regulated manufacturers: it reads complaint emails, photos and PDFs, finds similar past complaints and their corrective actions inside one customer's data, and drafts a brief with a suggested priority. Octave says it absorbs 4-6 hours of work per investigation. The handler still makes the call.",
   "agent": "Complaint Triage Assistant (Amazon Bedrock AgentCore)",
   "topic": "Customer Support",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/how-octave-built-a-secure-multi-tenant-ai-foundation-on-amazon-bedrock/",
   "prompt": "A customer sent a complaint with photos and a PDF. Summarise it, find similar past complaints and what fixed them, and suggest a priority. Show your sources and leave the decision to me.",
   "added": "2026-10-29",
   "interest": 57,
   "page": "https://agentsightings.com/case/octave-complaint-triage-assistant.html"
  },
  {
   "id": "cox-automotive-fleetmate-repair-estimates",
   "title": "Fleet repair estimates in 30 minutes instead of 8-48 hours",
   "summary": "Cox Automotive's FleetMate estimator assistant turns fleet repair estimates that took 8-48 hours into about 30 minutes; the first version was built in four days. Cox says it has 17 agent solutions in production, and a small dealer-messaging pilot saw consumer response rates more than 3x higher.",
   "agent": "FleetMate (Amazon Bedrock AgentCore)",
   "topic": "Business & Startups",
   "source": "Case study (AWS)",
   "url": "https://aws.amazon.com/solutions/case-studies/cox-auto-case-study/",
   "prompt": "Take this vehicle repair request and draft a repair estimate from the damage description, parts prices and labour rates. Show your working so an estimator can check it before it goes out.",
   "added": "2026-10-29",
   "interest": 57,
   "page": "https://agentsightings.com/case/cox-automotive-fleetmate-repair-estimates.html"
  },
  {
   "id": "keepard-agents-build-family-finance-app",
   "title": "A CTO built his family's finance app with coding agents in nine days",
   "summary": "Rodion says he built Keepard, an open-source household finance app, in nine days with Codex then Claude Code: 643 commits, 446 prompts. It pulls bank data across countries, classifies spending and asks in Telegram when unsure. Real cost to him: a $100/month subscription, about EUR 10/month to run. Self-reported.",
   "agent": "Claude Code / Codex",
   "topic": "Home & Family",
   "source": "Blog (DEV Community)",
   "url": "https://dev.to/rodik/i-built-a-self-hosted-family-finance-app-with-claude-code-in-9-days-heres-what-it-cost-and-what-i-2nl2",
   "prompt": "Pull this month's transactions from our bank accounts, classify them, and message me on Telegram about any you are not sure of. Show a one-currency total at the end.",
   "added": "2026-10-29",
   "interest": 57,
   "page": "https://agentsightings.com/case/keepard-agents-build-family-finance-app.html"
  },
  {
   "id": "work-helpdesk-tickets",
   "title": "Work helpdesk tickets",
   "summary": "As described in the source, claude Code shows up in a write-up about work helpdesk tickets. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Customer Support",
   "source": "Blog",
   "url": "https://www.eesel.ai/blog/how-to-use-claude-for-customer-support",
   "prompt": "Support task: work helpdesk tickets. Draft replies and triage labels only; I will send the final messages.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/work-helpdesk-tickets.html"
  },
  {
   "id": "use-operator-to-book-tables-compare-products-and-fill-forms",
   "title": "Use Operator to book tables, compare products and fill forms",
   "summary": "Case study takeaway: the poster says they tested ChatGPT Pro’s Operator by giving it tasks on websites such as booking tables, comparing products, filling forms, navigating sites, and logging into services with credentials. In comments they explain it runs in its own Chrome browser, appears to control a mouse in a separate browser pane, and Someone can take over control while assigning sequential website actions like checking Gmail, drafting replies, or purging spam. They describe it as smooth and able to handle web tasks with clear intent, though not perfectly.",
   "agent": "Operator",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPTPro/comments/1kjx5hx",
   "prompt": "Plan: use Operator to book tables, compare products and fill forms. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/use-operator-to-book-tables-compare-products-and-fill-forms.html"
  },
  {
   "id": "shop-for-cars-on-websites",
   "title": "Shop for cars on websites",
   "summary": "A published walkthrough shows meta Muse covered shop for cars on websites. utilizes Muse to shop for automobiles on the internet and discovers that a persistent logged-in user's browser profile with regular speed is the optimal choice as it is frequently blocked by websites.",
   "agent": "Meta Muse",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1wrlvs9/over_last_week_sites_that_used_to_work_are_now",
   "prompt": "Shopping task: shop for cars on websites. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/shop-for-cars-on-websites.html"
  },
  {
   "id": "set-tax-and-contract-reminders-and-review-contracts",
   "title": "Set tax and contract reminders and review contracts",
   "summary": "Observed in the wild: karina told Claude about falling behind on quarterly taxes, contract end dates, state laws for employees/subcontractors, and related admin. Claude offered on its own to create email reminders for deadlines and law changes; she accepted. No specific results or numbers are stated.",
   "agent": "Claude Cowork",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1uq5crr/what_ive_set_up_in_claude_for_my_legal_practice",
   "prompt": "Finance task: set tax and contract reminders and review contracts. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/set-tax-and-contract-reminders-and-review-contracts.html"
  },
  {
   "id": "run-side-hustle-invoicing-payments-and-bookkeeping",
   "title": "Run side hustle invoicing, payments and bookkeeping",
   "summary": "Observed in the wild: they connected a Meow bank account and QuickBooks to Claude through MCP for a side-business SaaS doing about $10k MRR. The agent sketches invoices, queues contractor payments, tracks expenses, and runs reconciliation and bookkeeping in QuickBooks; the owner set client details, rates, and categorization rules in the prompt and approves anything before money moves. They say finance admin dropped from 5-6 hours a week to about 20 minutes, freeing time for product work and customer conversations.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/thesidehustle/comments/1t7i6w2/i_use_ai_agents_to_manage_my_side_hustle_bank",
   "prompt": "Finance task: run side hustle invoicing, payments and bookkeeping. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/run-side-hustle-invoicing-payments-and-bookkeeping.html"
  },
  {
   "id": "run-inbox-drafts-meetings-and-purchases",
   "title": "Run inbox, drafts, meetings, and purchases",
   "summary": "One practitioner notes that user connected Meta Muse to inbox and Stripe. Agent reads messages, sketches email replies, sends meeting pings, and starts purchases but pauses for approval before any Stripe charge. User called it the most useful AI tried so far for inbox, drafting, meetings, and shopping.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/indianspeedster/status/2104651614929965161",
   "prompt": "Go through my mail for this goal: run inbox, drafts, meetings, and purchases. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/run-inbox-drafts-meetings-and-purchases.html"
  },
  {
   "id": "power-a-restaurant-customer-service-chatbot-backend",
   "title": "Power a restaurant customer service chatbot backend",
   "summary": "A published walkthrough shows a restaurant chatbot that responds to customer inquiries through text and voice memos is powered by a Make.com workflow as the backend.",
   "agent": "Make",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/chemE_ng/status/2110087829661376667",
   "prompt": "Shopping task: power a restaurant customer service chatbot backend. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/power-a-restaurant-customer-service-chatbot-backend.html"
  },
  {
   "id": "phone-local-stores-to-check-product-stock",
   "title": "Phone local stores to check product stock",
   "summary": "A published walkthrough shows the Grok-based agent was wired to Bland.ai for phone calls. It dialed local stores, held conversations to check stock on a specific product, and returned full transcripts plus availability status. The agent completed multiple store calls and delivered transcripts with stock details.",
   "agent": "Grok Bot",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/jkrajewski/status/2078169113545887878",
   "prompt": "Shopping task: phone local stores to check product stock. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/phone-local-stores-to-check-product-stock.html"
  },
  {
   "id": "organize-downloads-and-find-duplicates",
   "title": "Organize Downloads and find duplicates",
   "summary": "A non-developer runs Claude Code from their home directory to find duplicates, tidy Downloads, and suggest a cleaner directory structure.",
   "agent": "Claude Code",
   "topic": "Personal Files",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Review my Downloads folder. Find duplicates, propose a folder structure, move files into it, and list old files I probably do not need. Do not delete anything without asking.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/organize-downloads-and-find-duplicates.html"
  },
  {
   "id": "investigate-papers-overnight",
   "title": "Investigate papers overnight",
   "summary": "A published walkthrough shows as described in the source: working as people sleep, Grok Bot jobs are responsible for reading new papers, extract claims, rerun analyses, and scan sources.",
   "agent": "Grok Bot",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/flasssh_ai/status/2106815457441574915",
   "prompt": "Research brief: investigate papers overnight. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/investigate-papers-overnight.html"
  },
  {
   "id": "fill-out-booking-forms-automatically",
   "title": "Fill out booking forms automatically",
   "summary": "Case study takeaway: the reviewer Leaned on Perplexity's Comet browser assistant for repetitive booking forms. He gave it the details for a recycling centre appointment and told it to book. Comet worked through the forms with only occasional prompts asking whether he still wanted to finish the booking.",
   "agent": "Perplexity Comet",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/perplexity_ai/comments/1nd403m/comet_is_amazing",
   "prompt": "Plan: fill out booking forms automatically. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/fill-out-booking-forms-automatically.html"
  },
  {
   "id": "fill-out-50-forms-from-public-information",
   "title": "Fill out 50 forms from public information",
   "summary": "From the linked report, someone was assigned 50 different insurance forms, each with unique fields, and had Claude fill them out. They say the task Leaned on only public information available on company websites under workplace AI guidelines, and the human step was proofreading every form afterward to check accuracy. Claude completed all 50 forms, and Someone says their review located them accurate.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1ut4jv0/claude_did_a_task_for_me_at_work_that_made_me",
   "prompt": "Finance task: fill out 50 forms from public information. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/fill-out-50-forms-from-public-information.html"
  },
  {
   "id": "extract-expenses-from-receipt-photos-sent-via-chat",
   "title": "Extract expenses from receipt photos sent via chat",
   "summary": "From the linked report, the builder made AICountant for freelancers and small businesses as a Telegram or Discord bot. Users connect one of those accounts, send a receipt photo, and the AI pulls out vendor, amount, tax, category, and date, then adds the expense to a ledger for review before Someone can export it as CSV. The post says the product was already functional and the creator was recruiting beta testers to validate usage patterns.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SideProject/comments/1s8p0aq/i_built_an_ai_bookkeeper_that_extracts_expenses",
   "prompt": "Finance task: extract expenses from receipt photos sent via chat. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/extract-expenses-from-receipt-photos-sent-via-chat.html"
  },
  {
   "id": "draft-refund-emails-from-a-photo-and-gmail-receipt",
   "title": "Draft refund emails from a photo and Gmail receipt",
   "summary": "Community report: a photo of a pair of worn-through pants was uploaded to Claude Cowork, with a dictated prompt asking for a J.Crew refund. Cowork, connected to Gmail, searched for the receipt from PayPal or J.Crew, pulled the order number and purchase date, and drafted the support email. The source describes a drafted refund request with the order details and the defect, ready for review.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/shopping-with-claude",
   "prompt": "Go through my mail for this goal: draft refund emails from a photo and Gmail receipt. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/draft-refund-emails-from-a-photo-and-gmail-receipt.html"
  },
  {
   "id": "customize-and-order-a-matcha-drink-for-an-event",
   "title": "Customize and order a matcha drink for an event",
   "summary": "As described in the source, she connected her Monid API keys to Claude Cowork. The Monid agent afterward customized the milk choice and placed the order for a coconut cloud matcha drink ahead of the SF tech week rooftop pop-up event. The customized matcha was ordered for the event.",
   "agent": "Claude Cowork",
   "topic": "Shopping",
   "source": "X",
   "url": "https://x.com/gracepace_/status/2107346389282021501",
   "prompt": "Shopping task: customize and order a matcha drink for an event. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/customize-and-order-a-matcha-drink-for-an-event.html"
  },
  {
   "id": "check-supplier-invoices-for-duplicates-and-missing-gst",
   "title": "Check supplier invoices for duplicates and missing GST",
   "summary": "Community report: supplier invoices land in a shared inbox, an AP tool captures the invoice data, and an agent compares each one with prior transactions to spot duplicates, wrong amounts, and missing GST. The poster connected the workflow with nexos.ai, said setup for the anomaly-checking agent took about 20 minutes, and still does the final review and approval manually. The poster said this saves a surprising amount of time each month and now produces a weekly summary, shifting their time from data entry to review.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ukaccounting/comments/1srfaqy/best_ai_for_accounting_my_experience_using_ai",
   "prompt": "Go through my mail for this goal: check supplier invoices for duplicates and missing GST. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/check-supplier-invoices-for-duplicates-and-missing-gst.html"
  },
  {
   "id": "build-a-portfolio-plan-from-brokerage-exports",
   "title": "Build a portfolio plan from brokerage exports",
   "summary": "Observed in the wild: matt Stockton pointed Claude Code at a directory with brokerage CSV exports, a typed scratchpad of other balances, and a detailed goal prompt covering constraints, strawman targets and \"ask me questions exhaustively\". Claude wrote Python to parse and classify holdings, then assembled a gap analysis. It produced a ten-section plan with a four-phase action plan, tax payback tables and fund research.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://treeofprosperity.blogspot.com/2026/04/more-claude-coding-adventures-shaping.html",
   "prompt": "Finance task: build a portfolio plan from brokerage exports. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/build-a-portfolio-plan-from-brokerage-exports.html"
  },
  {
   "id": "build-a-personal-assistant-for-email-briefings-and-planning",
   "title": "Build a personal assistant for email, briefings and planning",
   "summary": "According to the write-up, marc installed Hermes Agent from Nous Research as a chat interface. He added more than 50 capabilities himself, such as email triage, daily briefings, calendar and task planning, meeting notes, and research radars. The agent became an operating layer across Marc's work, research, personal organisation and family life within 15 days.",
   "agent": "Hermes Agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/marc_bara/status/2099052224496140593",
   "prompt": "Go through my mail for this goal: build a personal assistant for email, briefings and planning. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/build-a-personal-assistant-for-email-briefings-and-planning.html"
  },
  {
   "id": "book-appointments-by-clicking-through-booking-sites",
   "title": "Book appointments by clicking through booking sites",
   "summary": "Linked example of book appointments by clicking through booking sites. The post says to start ChatGPT agent mode from the tools or + menu, or by typing /agent, then give it the booking task, either the booking URL or instructions to find a provider, your availability, and an instruction to stop before confirming or entering personal details.",
   "agent": "ChatGPT Agent",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/PromptEngineering/comments/1v8qcju/chatgpt_agent_mode_books_appointments_for_you_now",
   "prompt": "Plan: book appointments by clicking through booking sites. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/book-appointments-by-clicking-through-booking-sites.html"
  },
  {
   "id": "book-accommodation-up-to-payment-stage",
   "title": "Book accommodation up to payment stage",
   "summary": "One practitioner notes that a tech creator instructed GPT-6 Astra to book Tirupati accommodation. The agent opened the TTD website, selected options, entered details, and advanced the booking to the payment stage. The process completed to the payment stage in this showcase.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "X",
   "url": "https://x.com/AsifInsights/status/2104492996079313252",
   "prompt": "Plan: book accommodation up to payment stage. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/book-accommodation-up-to-payment-stage.html"
  },
  {
   "id": "analyze-budgets-and-research-investments-from-bank-data",
   "title": "Analyze budgets and research investments from bank data",
   "summary": "One practitioner notes that the poster says they use WisePenny to pipe transaction data from their bank accounts into ChatGPT without handing over bank logins or broad third-party permissions, then use Claude for the analysis work. In the thread they describe powered by Claude to build budgets from spending history, summarize and compare companies’ financial reports, and help research investments rather than directly controlling accounts. The source gives no hard performance numbers; it presents this as a practical read-and-analyze workflow.",
   "agent": "ChatGPT",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1s9b6im/anyone_using_claude_for_managing_finances",
   "prompt": "Finance task: analyze budgets and research investments from bank data. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 57,
   "page": "https://agentsightings.com/case/analyze-budgets-and-research-investments-from-bank-data.html"
  },
  {
   "id": "engie-agentforce-answers-energy-contract-questions",
   "title": "An energy firm's agent explains contracts to new customers",
   "summary": "ENGIE's Agentforce agent handles 1,500+ conversations a day, autonomously resolving 71% of cases with 83% response accuracy. In Italy, a link in the welcome email opens a chat that answers questions like 'when does my contract start?' using the customer's own contract data.",
   "agent": "Agentforce (ENGIE)",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/engie/",
   "prompt": "A new customer signed a contract. Using their contract record and our plan knowledge base, answer when it starts, their price type and when activation happens. Tell them if you do not know.",
   "added": "2026-10-29",
   "interest": 56,
   "page": "https://agentsightings.com/case/engie-agentforce-answers-energy-contract-questions.html"
  },
  {
   "id": "sun-agent-eight-cron-jobs-statements-to-notion",
   "title": "An agent called Sun turns the Citibank PDF into Notion rows every month",
   "summary": "Abhay Bhat describes an agent that, on the 13th of each month, finds the Citibank email, downloads the PDF, parses the transactions, sorts them into ten categories and syncs them to Notion with merchant names. It runs on eight cron jobs and a memory layer. Self-reported.",
   "agent": "Sun (personal agent, Telegram, mem0)",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://abhaybhat.com/blog/technology/eight-cron-jobs-one-ai-agent-and-a-mantra-for-my-daughter/",
   "prompt": "On the 13th of each month, find my bank statement email, download the PDF, categorise every transaction, and add them to my Notion finance table.",
   "added": "2026-10-29",
   "interest": 56,
   "page": "https://agentsightings.com/case/sun-agent-eight-cron-jobs-statements-to-notion.html"
  },
  {
   "id": "natera-voice-agent-books-mobile-blood-draws",
   "title": "A voice agent books mobile blood-draw appointments",
   "summary": "Natera put a voice scheduling agent into production for inbound mobile phlebotomy calls. In a four-week window it handled 4,744 calls, and the share of calls ending in under 30 seconds fell from 22% to 12%. Natera says cost is under one US cent per completed call.",
   "agent": "Voice scheduling agent (Amazon Bedrock AgentCore)",
   "topic": "Travel & Booking",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/nateras-intelligent-appointment-scheduling-with-amazon-bedrock-agentcore/",
   "prompt": "Answer inbound calls from patients wanting a home visit. Verify who they are, ask for date and time preferences, book a slot, and read back the confirmation. Hand over to a person if verification fails.",
   "added": "2026-10-29",
   "interest": 56,
   "page": "https://agentsightings.com/case/natera-voice-agent-books-mobile-blood-draws.html"
  },
  {
   "id": "hermes-agent-books-swimming-lessons-and-finds-battery",
   "title": "A home agent books swimming lessons and finds a clock battery",
   "summary": "An HN commenter runs a local Hermes agent that watches swimming-lesson availability and books a suitable slot, reads emails for tasks into Asana, and once identified a non-standard clock battery from a photo and ordered it. He says it takes boring chores off his plate.",
   "agent": "Hermes Agent",
   "topic": "Home & Family",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50187726",
   "prompt": "Watch the swimming club's booking page and book the first lesson that fits my son's schedule. Also, when I send a photo of something broken, work out the part I need and order it for my approval.",
   "added": "2026-10-29",
   "interest": 56,
   "page": "https://agentsightings.com/case/hermes-agent-books-swimming-lessons-and-finds-battery.html"
  },
  {
   "id": "sheila-agent-runs-company-accounting",
   "title": "A founder's agent handles invoices, payments and 1099s",
   "summary": "On Hacker News, Alex Gleason says his open-source agent Sheila handles his company's contractor invoices, payments in fiat and Bitcoin, bookkeeping, expenses and 1099s. It uses 50+ small scripts and a 600-line AGENTS.md, and he watches her work before anything goes out.",
   "agent": "Sheila (open-source agent)",
   "topic": "Finance",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47282326",
   "prompt": "Read new contractor invoices, check them against what we agreed, prepare the payment and a draft reply, and show me both before anything is sent.",
   "added": "2026-10-29",
   "interest": 56,
   "page": "https://agentsightings.com/case/sheila-agent-runs-company-accounting.html"
  },
  {
   "id": "weston-florida-city-hall-voice-agent-permits",
   "title": "A Florida city with 13 staff answers its phones with a voice agent",
   "summary": "EffiGov says Weston, Florida, a city of 68,000 with 13 direct employees, moved its City Hall line to voice AI on September 16, 2026. In the first month it answered 6,700+ calls and read 890+ permit records live from Accela. Vendor press release, 28 October 2026.",
   "agent": "EffiGov voice agent (Accela)",
   "topic": "Customer Support",
   "source": "News (EffiGov)",
   "url": "https://www.effigov.com/news/weston-fl-launch-2026",
   "prompt": "Answer City Hall calls around the clock. For permit questions, read the live permit record and tell the caller its status and any balance due.",
   "added": "2026-10-29",
   "interest": 56,
   "page": "https://agentsightings.com/case/weston-florida-city-hall-voice-agent-permits.html"
  },
  {
   "id": "search-and-apply-to-jobs-on-linkedin",
   "title": "Search and apply to jobs on LinkedIn",
   "summary": "Case study takeaway: the poster gave OpenClaw control of a Chrome browser with the OpenClaw extension installed, then manually logged into LinkedIn so the agent could use the live session. After discussing their background, the agent wrote and stored profile details in multiple markdown files, made a resume PDF, searched LinkedIn and other job sites, applied to matching roles, and logged applications in Google Sheets; the human helped create some accounts and later shut it down. The poster says it applied to 100+ jobs in about three days, leading to interviews with six companies and two offers, with the accepted one paying a little over $5k a month versus about $2.",
   "agent": "OpenClaw",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/openclaw/comments/1r8hek7/my_agent_doubled_my_salary_it_found_a_new_job_for",
   "prompt": "Help me search and apply to jobs on LinkedIn. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 56,
   "page": "https://agentsightings.com/case/search-and-apply-to-jobs-on-linkedin.html"
  },
  {
   "id": "run-a-home-robot-through-claude-desktop",
   "title": "Run a home robot through Claude Desktop",
   "summary": "One practitioner notes that tal bought a SunFounder PiCar-X robot kit with a camera and an ultrasonic distance sensor. He ran the harness on his laptop, with the Raspberry Pi riding on the robot, and handed control to Claude. Asked to park 20 cm from the room's door, it spent fifteen minutes parking in front of a filing cabinet.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.talraviv.co/p/i-gave-claude-a-robot-body",
   "prompt": "Go through my mail for this goal: run a home robot through Claude Desktop. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 56,
   "page": "https://agentsightings.com/case/run-a-home-robot-through-claude-desktop.html"
  },
  {
   "id": "prospect-score-leads-and-run-outbound-email-sequences",
   "title": "Prospect, score leads and run outbound email sequences",
   "summary": "As described in the source, getSalesClaw is a product that scans Apollo.io and job boards twice daily for prospects matching an ideal customer profile. It scores each lead 0–100 and drops low scorers. The source is a product description with no results, numbers, or user reports.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/automation/comments/1uy5lj6/i_couldnt_afford_to_hire_a_b2b_sales_team_so_i",
   "prompt": "Go through my mail for this goal: prospect, score leads and run outbound email sequences. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 56,
   "page": "https://agentsightings.com/case/prospect-score-leads-and-run-outbound-email-sequences.html"
  },
  {
   "id": "manage-family-email-calendar-and-school-chats",
   "title": "Manage family email, calendar and school chats",
   "summary": "Text or WhatsApp “Anna,” and the bot stays up overnight on the family inbox, shared calendar, and school chats. Pitch: the admin chores both parents keep dropping. Adoption claims are the vendor’s.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1udd09j/ai_agents_feel_one_step_away_from_a_real_personal",
   "prompt": "Go through my mail for this goal: manage family email, calendar and school chats. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 56,
   "page": "https://agentsightings.com/case/manage-family-email-calendar-and-school-chats.html"
  },
  {
   "id": "build-a-first-real-project",
   "title": "Build a first real project",
   "summary": "One practitioner notes that replit Agent was utilized by a student who had just completed a real build, resulting in an unexpected expense of approximately $65.",
   "agent": "Replit Agent",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1udnndh/replit_agent_cost_me_85_as_a_student_in_2_days",
   "prompt": "Finance task: build a first real project. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 56,
   "page": "https://agentsightings.com/case/build-a-first-real-project.html"
  },
  {
   "id": "add-travel-bookings-from-email-to-calendar",
   "title": "Add travel bookings from email to calendar",
   "summary": "Linked example of add travel bookings from email to calendar. The builder keeps a numbered Johnny.Decimal folder tree, with one folder holding saved scripts and documentation for adding travel bookings to their calendar.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49949563",
   "prompt": "Go through my mail for this goal: add travel bookings from email to calendar. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 56,
   "page": "https://agentsightings.com/case/add-travel-bookings-from-email-to-calendar.html"
  },
  {
   "id": "claire-et-net-agent-runs-tiny-business-week-1",
   "title": "Week 1 of an agent running a tiny business: EUR 37 spent, 0 sales",
   "summary": "An agent running in Claude Code posted its first-week ledger: EUR 200 starting capital, EUR 37.18 spent on an Etsy shop, Upwork credits and a Chrome Web Store fee, four listings, four proposals sent, EUR 0 revenue. It lists permission round trips and character limits among what went wrong. Self-reported.",
   "agent": "Claude Code (small-business agent)",
   "topic": "Business & Startups",
   "source": "Blog (DEV Community)",
   "url": "https://dev.to/clairetnet/week-1-as-an-ai-agent-running-a-tiny-business-real-files-honest-proposals-zero-sales-lbn",
   "prompt": "Run a tiny business with a fixed budget. Keep a ledger, check limits before handing anything over, and ask me before anything that spends money.",
   "added": "2026-10-29",
   "interest": 55,
   "page": "https://agentsightings.com/case/claire-et-net-agent-runs-tiny-business-week-1.html"
  },
  {
   "id": "rakuten-managed-agents-every-team-builds",
   "title": "Rakuten rolls Claude agents out to four departments in a week",
   "summary": "Anthropic says Rakuten deployed Claude Managed Agents across product, sales, marketing and finance within one week, ships major releases every two weeks instead of quarterly, and cut critical errors by 97%. Its Claude Code work cut time to market from 24 days to 5. Vendor-published.",
   "agent": "Claude Managed Agents / Claude Code",
   "topic": "Software Development",
   "source": "Case study (Anthropic)",
   "url": "https://claude.com/customers/rakuten",
   "prompt": "Take this feature request, break it into tasks, write the code and tests, and open a pull request. Flag anything you are unsure about for a human reviewer.",
   "added": "2026-10-29",
   "interest": 55,
   "page": "https://agentsightings.com/case/rakuten-managed-agents-every-team-builds.html"
  },
  {
   "id": "nestle-agent-drafts-replies-to-retailer-emails",
   "title": "Draft replies to retailers' order-status emails",
   "summary": "In Nestlé's B2B pilot, Agentforce reads retailers' emails and attachments, looks up order data and drafts a reply; a rep approves it. Finding order information dropped from several minutes to under one, and pilots suggest over 55% automation potential for invoice, order-status and credential requests.",
   "agent": "Agentforce B2B service agent",
   "topic": "Email & Inbox",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/nestle/agentic-b2b-service/",
   "prompt": "Read each retailer email and its attachments, look up the order in our system, and draft a reply about the delivery date, invoice or login problem. Leave it for me to approve.",
   "added": "2026-10-29",
   "interest": 55,
   "page": "https://agentsightings.com/case/nestle-agent-drafts-replies-to-retailer-emails.html"
  },
  {
   "id": "fanatics-betting-support-agents-responsible-gaming",
   "title": "Betting support agents that escalate responsible-gaming worries",
   "summary": "Fanatics Betting and Gaming runs a supervisor agent with tools for FAQs, account data and transactions, plus a small classifier that sends high-severity responsible-gaming messages straight to a human. In two months it reports containment up about 56% and resolution up about 53%.",
   "agent": "Multi-agent support (Amazon Bedrock)",
   "topic": "Customer Support",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/how-fanatics-betting-and-gaming-built-a-multi-agent-customer-support-system/",
   "prompt": "Answer customer questions using our state-by-state rules, account and transaction tools. Check every message for signs of problem gambling and hand those to a human straight away with the conversation.",
   "added": "2026-10-29",
   "interest": 55,
   "page": "https://agentsightings.com/case/fanatics-betting-support-agents-responsible-gaming.html"
  },
  {
   "id": "talorys-personal-agent-on-cloudflare-free-tier",
   "title": "A personal agent that runs on Cloudflare's free tier",
   "summary": "A developer released Talorys, an open-source personal agent that runs in your own Cloudflare account: chat with memory, tasks and notes, reminders and recurring automations, deployed with one command. The HN thread has 331 points and a sceptical top comment that few of these assistants show evidence they work.",
   "agent": "Talorys",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50031614",
   "prompt": "Set up a personal assistant I own, with memory, tasks and reminders. Then tell me how you will prove to me that it works rather than just feels like it does.",
   "added": "2026-10-29",
   "interest": 55,
   "page": "https://agentsightings.com/case/talorys-personal-agent-on-cloudflare-free-tier.html"
  },
  {
   "id": "robots-folding-napkins-for-din-tai-fung",
   "title": "Robots folding napkins for Din Tai Fung",
   "summary": "Dyna Robotics has robots doing the napkin folding for restaurant chain Din Tai Fung, and the customer is reportedly asking for more of them. The bar Dyna is chasing before it scales up is about 99.9% reliability, because a robot that only gets it right most of the time still needs someone watching it.",
   "agent": "Dyna Robotics",
   "topic": "Automation",
   "source": "ZeroHedge (BTIG note)",
   "url": "https://www.zerohedge.com/technology/btig-spots-real-world-evidence-humanoid-robots-are-entering-workforce",
   "prompt": "Here is a repetitive task my team does by hand every day. Break it into steps, note where a robot or agent would most likely slip up, and estimate how reliable it would need to be before we could leave it running unattended.",
   "added": "2026-10-19",
   "interest": 55,
   "page": "https://agentsightings.com/case/robots-folding-napkins-for-din-tai-fung.html"
  },
  {
   "id": "write-design-and-post-social-content-from-a-google-sheet",
   "title": "Write, design and post social content from a Google Sheet",
   "summary": "Observed in the wild: the setup starts with one Google Sheets row carrying the post idea, audience and tone. In Make, the flow uses OpenAI to write a caption, produces an image with DALL·E or another image model, uploads that image to Cloudinary, writes the assets back to the sheet, then sends a Slack approval notice before Buffer publishes the post; the person checks the caption and image in Slack and approves it. The poster says it turns a sheet row into a published social post inside two minutes and can handle the caption, image and posting end to end.",
   "agent": "Make",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/automation/comments/1ghtfn4",
   "prompt": "Creative task: write, design and post social content from a Google Sheet. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/write-design-and-post-social-content-from-a-google-sheet.html"
  },
  {
   "id": "two-hour-slack-dashboard-from-meeting-dumps",
   "title": "Two-hour Slack dashboard from meeting dumps",
   "summary": "A Rippling engineer dumps notes into Slack; a Cursor cron agent every two hours merges them with PRs, Jira and mentions into a clean task dashboard.",
   "agent": "Cursor",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "Every two hours, read my task Slack channel plus open GitHub PRs, Jira issues and Slack mentions. Deduplicate, group by priority, and post a short dashboard of what needs me today.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/two-hour-slack-dashboard-from-meeting-dumps.html"
  },
  {
   "id": "turn-a-telegram-message-into-researched-designed-social-posts",
   "title": "Turn a Telegram message into researched, designed social posts",
   "summary": "Case study takeaway: n8n shows up in a write-up about turn a Telegram message into researched, designed social posts, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1re2fo2/i_built_an_ai_content_agent_in_n8n_that",
   "prompt": "Assistant mode: turn a Telegram message into researched, designed social posts. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/turn-a-telegram-message-into-researched-designed-social-posts.html"
  },
  {
   "id": "shop-for-umbrella-insurance-through-your-personal-agent",
   "title": "Shop for umbrella insurance through your personal agent",
   "summary": "Case study takeaway: coverage Cat is a licensed brokerage that publishes an Agent API and MCP endpoint for AI agents. The agent starts from the MCP manifest and calls the umbrella_consumer_prefill tool. The page gives no outcome numbers for agent-driven umbrella shopping.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49804931",
   "prompt": "Finance task: shop for umbrella insurance through your personal agent. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/shop-for-umbrella-insurance-through-your-personal-agent.html"
  },
  {
   "id": "set-up-overseas-supplier-payments-for-review",
   "title": "Set up overseas supplier payments for review",
   "summary": "Documented elsewhere: the approved AI agent, connected to Airwallex Agentic Business Accounts, checks balances across accounts holding the needed currency and prepares the overseas supplier payment details. It does this entirely within rules Someone has set. The agent executes the account checks and payment setup steps that would otherwise require jumping between bank logins.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/AriaWestcott/status/2105366236930486623",
   "prompt": "Finance task: set up overseas supplier payments for review. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/set-up-overseas-supplier-payments-for-review.html"
  },
  {
   "id": "send-tone-adjusted-overdue-invoice-reminders",
   "title": "Send tone-adjusted overdue invoice reminders",
   "summary": "According to the write-up, the builder ran n8n locally in Docker and assembled a chat-triggered AI agent powered by Gemini (gemini-2.0-flash-lite) with simple memory. It had two tools: a Data Table read (getInvoices) and a Gmail SMTP email sender. The source describes the finished workflow, which runs from one chat command and returns a JSON summary of emails sent.",
   "agent": "n8n",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://dev.to/canankorkut/build-your-first-ai-agent-with-n8n-32pf",
   "prompt": "Finance task: send tone-adjusted overdue invoice reminders. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/send-tone-adjusted-overdue-invoice-reminders.html"
  },
  {
   "id": "scrape-a-conference-site-into-a-searchable-dashboard",
   "title": "Scrape a conference site into a searchable dashboard",
   "summary": "Documented elsewhere: in a workshop demo, a Manus agent scraped a whole conference website of about 70 talks. It pulled out dates, times and descriptions, and converted every time to UTC. The demo produced a working dashboard covering the 70-plus talks.",
   "agent": "Manus",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://growwstacks.com/blog/how-to-build-intelligent-research-agents-with-manus-ai",
   "prompt": "Assistant mode: scrape a conference site into a searchable dashboard. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/scrape-a-conference-site-into-a-searchable-dashboard.html"
  },
  {
   "id": "scan-thrift-store-items-and-flag-good-deals",
   "title": "Scan thrift store items and flag good deals",
   "summary": "A published walkthrough shows wes Bos assembled an agent that processes video frames captured while scanning thrift store items, zeros in on objects in the frames, queries online prices for each, and alerts on good deals. The builder described being shocked at how well the agent worked.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/wesbos/status/2102769412608778737",
   "prompt": "Creative task: scan thrift store items and flag good deals. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/scan-thrift-store-items-and-flag-good-deals.html"
  },
  {
   "id": "run-background-agents",
   "title": "Run background agents",
   "summary": "A published walkthrough shows custom agent shows up in a write-up about run background agents. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49713894",
   "prompt": "Assistant mode: run background agents. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/run-background-agents.html"
  },
  {
   "id": "run-a-weekly-competitive-monitoring-briefing-for-product-management",
   "title": "Run a weekly competitive-monitoring briefing for product management",
   "summary": "A published walkthrough shows hermes was tasked by a product manager to create a weekly briefing for competitors, but its skill retested four times, resulting in the discovery of missed patterns by the product manager during week six.",
   "agent": "Hermes Agent",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://www.news.aakashg.com/p/hermes-agent-guide",
   "prompt": "Assistant mode: run a weekly competitive-monitoring briefing for product management. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/run-a-weekly-competitive-monitoring-briefing-for-product-management.html"
  },
  {
   "id": "run-a-scheduled-personal-assistant-that-builds-a-task-dashboard",
   "title": "Run a scheduled personal assistant that builds a task dashboard",
   "summary": "A published walkthrough shows cursor shows up in a write-up about run a scheduled personal assistant that builds a task dashboard. The linked post is the primary source for steps and outcomes.",
   "agent": "Cursor",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/leerob/status/2029605390942454257",
   "prompt": "Assistant mode: run a scheduled personal assistant that builds a task dashboard. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/run-a-scheduled-personal-assistant-that-builds-a-task-dashboard.html"
  },
  {
   "id": "run-a-meeting-assistant-that-handles-meetings-and-summaries",
   "title": "Run a meeting assistant that handles meetings and summaries",
   "summary": "Linked example of run a meeting assistant that handles meetings and summaries. In the Reddit thread, a commenter said they used Lindy to build an AI assistant for meetings and related tasks after seeing recommendations there.",
   "agent": "Lindy",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1m6hp3y/whats_the_most_useful_ai_agent_youve_actually_seen",
   "prompt": "Assistant mode: run a meeting assistant that handles meetings and summaries. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/run-a-meeting-assistant-that-handles-meetings-and-summaries.html"
  },
  {
   "id": "run-a-shopify-store-overnight",
   "title": "Run a Shopify store overnight",
   "summary": "One practitioner notes that in a demonstration on YouTube, users can watch how NightShift can detect store issues, auto-repair them to safe ones, and ensure every action is reviewed.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/TusharA5649881/status/2094278204051894316",
   "prompt": "Creative task: run a Shopify store overnight. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/run-a-shopify-store-overnight.html"
  },
  {
   "id": "research-laptops-and-build-a-comparison-table-and-deck",
   "title": "Research laptops and build a comparison table and deck",
   "summary": "One practitioner notes that in ChatGPT's agent mode, the reviewer asked it to research the best budget laptops, compare specs, and send back a summary with a table and a presentation. The agent opened its own virtual desktop, and the reviewer watched it read laptop review articles. The research and summary were decent and the information was solid, but the presentation looked cheap and unprofessional.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://cybernews.com/ai-tools/chatgpt-agent-review/",
   "prompt": "Research brief: research laptops and build a comparison table and deck. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/research-laptops-and-build-a-comparison-table-and-deck.html"
  },
  {
   "id": "reconcile-industry-files-and-chase-clarifications",
   "title": "Reconcile industry files and chase clarifications",
   "summary": "Documented elsewhere: Custom agent was pointed at reconcile industry files and chase clarifications. from the linked report: and contacts people to resolve bad data, following an agent workflow ingests a complex industry file, loads it into a database, spawns agents to check it against work in flight.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49726000",
   "prompt": "Plan: reconcile industry files and chase clarifications. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/reconcile-industry-files-and-chase-clarifications.html"
  },
  {
   "id": "order-starbucks-by-voice-from-a-tesla-while-driving",
   "title": "Order Starbucks by voice from a Tesla while driving",
   "summary": "Linked example of order Starbucks by voice from a Tesla while driving. The poster said they had been testing Grok Bot in a Tesla for a couple of weeks and used the in-car version, not a phone, while FSD was driving.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/TeslaFSD/comments/1wnfs5d/grok",
   "prompt": "Help me order Starbucks by voice from a Tesla while driving. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/order-starbucks-by-voice-from-a-tesla-while-driving.html"
  },
  {
   "id": "morning-handoff-from-yesterday-s-agent-sessions",
   "title": "Morning handoff from yesterday's agent sessions",
   "summary": "A scheduled Codex automation reviews prior sessions, groups what needs attention and emails top priorities with deep links back to each thread.",
   "agent": "Codex",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://mohit.stream/discoveries/2026-05-08-codex-morning-digest-automation/",
   "prompt": "Review yesterday's sessions. Group items into: needs attention, waiting on a decision, conditional, ignore. Email me the top three priorities with links to each session.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/morning-handoff-from-yesterday-s-agent-sessions.html"
  },
  {
   "id": "manage-calendar-and-tasks-by-text",
   "title": "Manage calendar and tasks by text",
   "summary": "Case study takeaway: lindy shows up in a write-up about manage calendar and tasks by text. The linked post is the primary source for steps and outcomes.",
   "agent": "Lindy",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://zapier.com/blog/lindy-review/",
   "prompt": "Assistant mode: manage calendar and tasks by text. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/manage-calendar-and-tasks-by-text.html"
  },
  {
   "id": "decide-between-points-and-cash-for-every-trip",
   "title": "Decide between points and cash for every trip",
   "summary": "According to the write-up, the toolkit installs into Claude Code as a plugin with skills, MCP servers and a travel-hacker subagent. Given a request, it searches cash sources (Duffel, Ignav, Google Flights, Skiplagged, Kiwi) and Seats.aero for awards. The source describes the workflow and what is included, not measured results.",
   "agent": "Claude Code",
   "topic": "Travel & Booking",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47635033",
   "prompt": "Plan: decide between points and cash for every trip. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/decide-between-points-and-cash-for-every-trip.html"
  },
  {
   "id": "compile-and-filter-apartment-listings",
   "title": "Compile and filter apartment listings",
   "summary": "According to the write-up, renno Mihalis, 23, had one week to find an apartment in San Francisco after getting a job offer. Rather than powered by a single rental platform, he Leaned on Claude and ChatGPT to compile a list of property management websites. The AI did a pretty good job listing apartments within his parameters, but sometimes included ones outside them.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://almanacnews.com/technology/2026/10/06/how-ai-is-changing-the-apartment-search-is-it-working",
   "prompt": "Help me compile and filter apartment listings. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/compile-and-filter-apartment-listings.html"
  },
  {
   "id": "clean-inbox-chase-refund-and-book-hvac-tune-up",
   "title": "Clean inbox, chase refund, and book HVAC tune-up",
   "summary": "Case study takeaway: the Muse agent had access to Someone's Hotmail inbox and prior maintenance records. It reviewed 80K emails, flagged 20k+ old promos and newsletters for deletion with explicit user approval, and logged every action. Inbox dropped from 96% to 60% full.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/anandragn/status/2103981239199834130",
   "prompt": "Go through my mail for this goal: clean inbox, chase refund, and book HVAC tune-up. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/clean-inbox-chase-refund-and-book-hvac-tune-up.html"
  },
  {
   "id": "bundle-monthly-vendor-invoices-for-the-accountant",
   "title": "Bundle monthly vendor invoices for the accountant",
   "summary": "A published walkthrough shows the shared Grok Bot template runs on a schedule rather than waiting to be asked. It reads each invoice PDF and pulls the vendor, date, amount and invoice number. The page gives no results, numbers or time savings, and no examples of its output.",
   "agent": "Grok Bot",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://grokbot.dev/marketplace/invoice-hunter/",
   "prompt": "Finance task: bundle monthly vendor invoices for the accountant. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/bundle-monthly-vendor-invoices-for-the-accountant.html"
  },
  {
   "id": "build-a-booking-and-payment-app-for-tattoo-artists",
   "title": "Build a booking and payment app for tattoo artists",
   "summary": "As described in the source, on Thursday night he opened a blank repl and typed one paragraph asking for a Calendly-style calendar, Google OAuth, Stripe deposits and Postmark notifications. Replit Agent sketched the backend, wired Stripe, set up calendar integration and scaffolded a React front end. He ended up with a working prototype: client booking, artist dashboards, onboarding, conflict handling and Postmark confirmation emails.",
   "agent": "Replit Agent",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://iliashatzis.substack.com/p/from-zero-to-working-prototype-in",
   "prompt": "Plan: build a booking and payment app for tattoo artists. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/build-a-booking-and-payment-app-for-tattoo-artists.html"
  },
  {
   "id": "build-a-telegram-bot-that-sends-emails-and-checks-calendar",
   "title": "Build a Telegram bot that sends emails and checks calendar",
   "summary": "Julian Goldie used Replit agents/automations (with an OpenAI key and Outlook OAuth) to ship a Telegram bot that checks calendar gaps and sends Outlook invites in 28 chat turns. Beta quirks included weak playground memory and occasional formatting glitches.",
   "agent": "Replit Agent",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/JulianGoldieSEO/status/1970461781899547063",
   "prompt": "Build a Telegram bot that can check my Outlook calendar and draft meeting invites. Use OAuth, show the message flow before publishing, and do not send invites until I test in dry-run.",
   "added": "2026-10-15",
   "interest": 55,
   "page": "https://agentsightings.com/case/build-a-telegram-bot-that-sends-emails-and-checks-calendar.html"
  },
  {
   "id": "zendesk-custom-agents-million-runs",
   "title": "Zendesk's customers ran 1 million custom support agents in seven weeks",
   "summary": "Zendesk built a builder for customers' own service agents on Claude, with a five-person team in four months. In early access the agents ran 1 million times in seven weeks. One customer's duplicate-ticket merging reached 99% accuracy in a pilot. Zendesk reports up to 80% lower human handle times. Vendor-published.",
   "agent": "Zendesk custom agents (Claude via Amazon Bedrock)",
   "topic": "Customer Support",
   "source": "Case study (Anthropic)",
   "url": "https://claude.com/customers/zendesk",
   "prompt": "Look at new support tickets, find duplicates of an open ticket, and merge them with a note saying why. If you are not sure two tickets are the same, leave them alone and flag them.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/zendesk-custom-agents-million-runs.html"
  },
  {
   "id": "nestle-agent-answers-consumer-questions",
   "title": "Nestlé's first consumer-service agent resolves 40% of enquiries",
   "summary": "Nestlé launched an Agentforce consumer-service agent in eight weeks. In pilots it resolved 40% of consumer enquiries on its own with a 4.8/5 satisfaction rating. Each new country takes about eight weeks to launch.",
   "agent": "Agentforce consumer service agent",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/nestle/agentic-consumer-service/",
   "prompt": "Answer shoppers' questions about our products, hand over to a person when a human touch matters, and give the person the conversation so far.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/nestle-agent-answers-consumer-questions.html"
  },
  {
   "id": "bradesco-agent-cuts-contract-review-hour-to-five-minutes",
   "title": "Contract and invoice review down from 1 hour to 5 minutes",
   "summary": "Google says Bradesco's finance team uses Gemini Enterprise for fiscal, accounting and contract analysis, cutting document review from one hour to five minutes and risk inconsistencies by 60%. Vendor-reported.",
   "agent": "Gemini Enterprise",
   "topic": "Finance",
   "source": "News (Google Cloud)",
   "url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
   "prompt": "Review this contract and invoice set, classify the services, check for inconsistencies in tax and terms, and list each risk with the page where you found it.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/bradesco-agent-cuts-contract-review-hour-to-five-minutes.html"
  },
  {
   "id": "anthropic-sales-agent-answers-inbound-buyers",
   "title": "Anthropic's own sales funnel starts with an agent",
   "summary": "A Claude agent now answers pricing, security and plan questions for people who used to wait days after the Contact Sales form. A news site reports Anthropic says escalated buyers became opportunities more than twice as often and closed about five days faster. One engineer built version one in a few weeks. Company figures.",
   "agent": "Claude buying agent (Claude Managed Agents)",
   "topic": "Sales & Marketing",
   "source": "News (Claude News)",
   "url": "https://claudenews.com/news/how-anthropics-sales-team-rebuilt-inbound-with-claude-managed-agents",
   "prompt": "Answer buyers' questions on pricing, seats and security from our docs, recommend a plan, and hand complex deals to a rep with the full conversation and why you escalated.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/anthropic-sales-agent-answers-inbound-buyers.html"
  },
  {
   "id": "dxc-assure-claude-claims-backlog-minutes",
   "title": "An insurance platform clears claims backlogs in minutes",
   "summary": "Anthropic says DXC runs Claude inside its Assure insurance platform: claims documents are classified in under a second, so backlogs go from days to minutes, and a regulated benefits calculation was stood up in eight hours, 80% right first pass. Every agent action is logged. Vendor-published.",
   "agent": "Claude in DXC Assure",
   "topic": "Business & Startups",
   "source": "Case study (Anthropic)",
   "url": "https://claude.com/customers/dxc",
   "prompt": "Read each incoming claims document, classify it, route it to the right queue, and log every action you take so an auditor can replay it.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/dxc-assure-claude-claims-backlog-minutes.html"
  },
  {
   "id": "wellstar-wali-assistant-cuts-service-desk-handle-time-64-percent",
   "title": "A hospital's IT help agent cuts handle time from 25 minutes to 9",
   "summary": "ServiceNow says Wellstar Health System's assistant Wali answers clinicians' IT questions inside Teams and the EMR support environment, while Now Assist summarises incidents for the service desk. Average handle time fell from 25 to 9 minutes, and Wellstar absorbed a three-hospital merger without adding headcount. Vendor-published.",
   "agent": "Wali (Moveworks) and Now Assist",
   "topic": "DevOps & Infra",
   "source": "Case study (ServiceNow)",
   "url": "https://www.servicenow.com/customers/wellstar-ai.html",
   "prompt": "A nurse reports a broken printer on the EMR. Find the fix or the right tip sheet, complete the steps you can, and route the rest to the service desk with a summary.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/wellstar-wali-assistant-cuts-service-desk-handle-time-64-percent.html"
  },
  {
   "id": "lilo-telegram-assistant-photos-receipts-reminders",
   "title": "A Telegram assistant for calories, receipts and reminders",
   "summary": "The creator of Lilo, an open source assistant that lives in Telegram, uses it to track calories from food photos, turn a voice note into a TODO, remind him when a game starts, and file a forwarded Uber receipt for a work reimbursement. It also works by email and WhatsApp.",
   "agent": "Lilo",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48237730",
   "prompt": "When I send a photo of food, log the calories. When I forward a receipt, save it for reimbursement. When I send a voice note, add a TODO. Remind me before events I tell you about.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/lilo-telegram-assistant-photos-receipts-reminders.html"
  },
  {
   "id": "rook-open-source-assistant-5-dollar-vps",
   "title": "A $5 VPS assistant that lives in Telegram and runs a morning briefing",
   "summary": "Michal Kraus open-sourced Rook, which runs on a $5 VPS. It manages Google Calendar, triages Gmail, controls Spotify and a TV, sends a 7am briefing and 15-minute reminders, and remembers preferences. He has used it for several months. Self-reported.",
   "agent": "Rook (open-source assistant)",
   "topic": "Personal Assistant",
   "source": "Blog (DEV Community)",
   "url": "https://dev.to/michal_kraus_151c2a53e667/i-built-an-open-source-ai-assistant-that-actually-runs-my-day-heres-the-architecture-24b3",
   "prompt": "Each morning at 7, send me a briefing of today's calendar and any urgent email. Remind me 15 minutes before each meeting.",
   "added": "2026-10-29",
   "interest": 54,
   "page": "https://agentsightings.com/case/rook-open-source-assistant-5-dollar-vps.html"
  },
  {
   "id": "turn-stroller-voice-notes-into-publishable-posts",
   "title": "Turn stroller voice notes into publishable posts",
   "summary": "A parent feeds rambling voice notes into Claude Code, which organizes themes, writes an article in their voice, and drafts LinkedIn versions.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Transcribe these voice notes, cluster them into research themes, write one full article in my voice using the examples I provide, and produce a shorter LinkedIn version of each theme.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/turn-stroller-voice-notes-into-publishable-posts.html"
  },
  {
   "id": "schedule-clinic-appointments-by-phone",
   "title": "Schedule clinic appointments by phone",
   "summary": "Community report: as described in the source: instead of a call to book an appointment, a clinic's automated phone AI took the poster's call.",
   "agent": "Custom agent",
   "topic": "Travel & Booking",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/DentalAssistant/comments/1ptrt4i/ai_answered_to_schedule_my_appointment",
   "prompt": "Plan: schedule clinic appointments by phone. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/schedule-clinic-appointments-by-phone.html"
  },
  {
   "id": "run-web-research-jobs-and-capture-screenshots-into-a-doc",
   "title": "Run web research jobs and capture screenshots into a doc",
   "summary": "Observed in the wild: the reviewer paid $200 for ChatGPT Operator and spent a week giving it plain-language browser tasks. These included analyzing trending web-design tweets, summarizing AI news, listing AI website builders with pricing, summarizing a set of linked articles, and screenshotting sites into a shared Google Doc. Operator returned duplicate, stale tweets, summarized one Axios article, and copied the first ten items from a single Bing listicle.",
   "agent": "Operator",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://icons8.com/blog/articles/chatgpt-operator-review/",
   "prompt": "Research brief: run web research jobs and capture screenshots into a doc. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/run-web-research-jobs-and-capture-screenshots-into-a-doc.html"
  },
  {
   "id": "run-monte-carlo-simulations-on-a-retirement-portfolio",
   "title": "Run Monte Carlo simulations on a retirement portfolio",
   "summary": "Observed in the wild: the poster says they rebuilt their retirement portfolio in Claude and had it model a 60:40 allocation over 15 to 20 years, powered by a 6% baseline path while randomly inserting two or three historical bear markets, then layering in one-time home-improvement costs. In comments, they add that they first mapped a month-by-month budget to stabilize recurring expenses and drawdowns, and that Claude split fixed-income and stock components; the human role was framing the assumptions and considering whether to validate it with a Fidelity advisor. The poster says the exercise supported their thesis that projected expenses and income would likely cover 20-plus years.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/investing/comments/1w30d74/claude_montecarlo_simulations",
   "prompt": "Help me run Monte Carlo simulations on a retirement portfolio. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/run-monte-carlo-simulations-on-a-retirement-portfolio.html"
  },
  {
   "id": "review-a-landlord-deposit-dispute-for-legal-violations",
   "title": "Review a landlord deposit dispute for legal violations",
   "summary": "Real-world thread: the tenant fed their move-out dispute into ChatGPT and Gemini, powered by the two models to cross-check each other, suggest flaws, and repeatedly re-review the case from a fresh context. From the tenant’s notes, the models flagged claimed violations covering AB2801, missing receipts and photos, corrected an overstated SF tenant fee, and Gemini drafted a formal demand letter and PDF that the tenant sent. The landlord returned the demanded $4,200 about an hour after receiving the letter, after initially trying to keep $4,200 of a $5,000 deposit.",
   "agent": "ChatGPT",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1tbkg2u/chatgptgemini_saved_me_4200_from_a_scam_land_lord",
   "prompt": "Notes task: review a landlord deposit dispute for legal violations. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/review-a-landlord-deposit-dispute-for-legal-violations.html"
  },
  {
   "id": "research-and-buy-online",
   "title": "Research and buy online",
   "summary": "Real-world thread: as described in the source: through the use of Stripe Link, Grok Bot investigates options and creates orders, displaying a spend request with merchant, total, and description, and concluding with a payment option.",
   "agent": "Grok Bot",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://flaviocopes.com/grok-bot/",
   "prompt": "Research brief: research and buy online. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/research-and-buy-online.html"
  },
  {
   "id": "plan-parties-and-shop-for-products",
   "title": "Plan parties and shop for products",
   "summary": "Community report: claude shows up in a write-up about plan parties and shop for products. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/claude/comments/1wro63z/whats_your_personal_not_work_claude_use_case",
   "prompt": "Shopping task: plan parties and shop for products. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/plan-parties-and-shop-for-products.html"
  },
  {
   "id": "let-ai-agents-file-employment-tax-returns",
   "title": "Let AI agents file employment tax returns",
   "summary": "A published walkthrough shows the MCP server utilized by agents for tax returns, including employment forms 941 and 940, is currently in development but not yet live and is classified as a tool call to the IRS.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47769711",
   "prompt": "Finance task: let AI agents file employment tax returns. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/let-ai-agents-file-employment-tax-returns.html"
  },
  {
   "id": "grade-papers-and-build-lesson-plans-for-teachers",
   "title": "Grade papers and build lesson plans for teachers",
   "summary": "Linked example of grade papers and build lesson plans for teachers. The builder assembled a proof-of-concept app called Archimedes, described as a React and Flask app that used ChatGPT across school levels from elementary through doctorate.",
   "agent": "ChatGPT",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/12jg0vl",
   "prompt": "Research brief: grade papers and build lesson plans for teachers. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/grade-papers-and-build-lesson-plans-for-teachers.html"
  },
  {
   "id": "draft-social-posts-from-trending-topics-for-approval",
   "title": "Draft social posts from trending topics for approval",
   "summary": "Lindy + Computer Use sits in a real browser on LinkedIn. Weekday 8am job: scan niche trends, draft three voice-matched posts, DM them for a human thumbs-up. Reviewer label: “genuinely agentic,” score 2.",
   "agent": "Lindy",
   "topic": "Design & Creative",
   "source": "Blog",
   "url": "https://posteverywhere.ai/blog/best-ai-agents-for-social-media",
   "prompt": "Design task: draft social posts from trending topics for approval. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/draft-social-posts-from-trending-topics-for-approval.html"
  },
  {
   "id": "dispute-an-airline-x27-s-eu261-delay-compensation-rejection",
   "title": "Dispute an airline's EU261 delay compensation rejection",
   "summary": "A published walkthrough shows grok Bot shows up in a write-up about dispute an airline's EU261 delay compensation rejection, with figures like 400. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/nexor404/status/2106808072622727647",
   "prompt": "Go through my mail for this goal: dispute an airline's EU261 delay compensation rejection. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/dispute-an-airline-x27-s-eu261-delay-compensation-rejection.html"
  },
  {
   "id": "build-a-clinic-app",
   "title": "Build a clinic app",
   "summary": "Community report: replit Agent shows up in a write-up about build a clinic app, with figures like 5. The linked post is the primary source for steps and outcomes.",
   "agent": "Replit Agent",
   "topic": "Travel & Booking",
   "source": "Blog",
   "url": "https://www.goodreads.com/author_blog_posts/25813418-doctor-built-his-own-app-with-ai-saved-100k",
   "prompt": "Plan: build a clinic app. Compare 2–3 options (cost, timing, tradeoffs), recommend one, and do not book or pay yet.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/build-a-clinic-app.html"
  },
  {
   "id": "assemble-an-immigration-document-package",
   "title": "Assemble an immigration document package",
   "summary": "Seven years of tax returns, bank PDFs, and immigration paperwork—pulled via ChatGPT browser sessions into a single packet in minutes. The post stops at “assembled,” not “filed.”",
   "agent": "ChatGPT Agent",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/athyuttamre/status/2089157303320871022",
   "prompt": "Finance task: assemble an immigration document package. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 54,
   "page": "https://agentsightings.com/case/assemble-an-immigration-document-package.html"
  },
  {
   "id": "cars24-voice-chat-agents-million-minutes",
   "title": "A used-car site's agents handle 1 million conversation minutes a month",
   "summary": "OpenAI's Cars24 story, as summarised by GIGAZINE, says voice and chat agents handle 1M+ conversation minutes a month, resolution rates rose 50%, key service workflows are 80% faster, and 12% of customers who dropped out were brought back. Vendor-published, second-hand source.",
   "agent": "Cars24 voice and chat agents (OpenAI)",
   "topic": "Customer Support",
   "source": "News (GIGAZINE, OpenAI case study)",
   "url": "https://gigazine.net/gsc_news/en/20260717-openai-cars24-agent/",
   "prompt": "Call or message sellers who dropped out of the car-selling process, find out why, answer their questions and book the next step.",
   "added": "2026-10-29",
   "interest": 53,
   "page": "https://agentsightings.com/case/cars24-voice-chat-agents-million-minutes.html"
  },
  {
   "id": "zingage-home-care-agent-covers-call-outs",
   "title": "A home-care agent finds a replacement when a caregiver calls out sick",
   "summary": "Zingage's agents coordinate shifts for home-care agencies. The company reports under 40 minutes from caregiver call-out to confirmed replacement, issue resolution under 5 minutes (down from 20-30), and an 82% cut in after-hours labour costs at one Medicaid agency. Vendor-published, from Anthropic's case study.",
   "agent": "Zingage care-coordination agent (Claude)",
   "topic": "Business & Startups",
   "source": "Case study (Anthropic)",
   "url": "https://claude.com/customers/zingage",
   "prompt": "A caregiver has called out for a shift. Find eligible replacements, contact them in order, and confirm the first yes. Tell the coordinator what you did.",
   "added": "2026-10-29",
   "interest": 53,
   "page": "https://agentsightings.com/case/zingage-home-care-agent-covers-call-outs.html"
  },
  {
   "id": "freight-agent-quotes-and-reconciles-invoices",
   "title": "A freight agent turns a 30-minute quote into a ten-minute check",
   "summary": "A partner built FreightIQ for Alliance Air Freight. It reads quote emails, pulls live rates and reconciles 65-70 invoices a day against the TMS, with a person approving every send. Reported: quotes from 20-30 minutes to 5-10, invoices from 30-120 minutes to under 10, extraction accuracy above 95%. Partner-published.",
   "agent": "FreightIQ (Bedrock AgentCore, Claude)",
   "topic": "Business & Startups",
   "source": "Case study (BizCloud Experts)",
   "url": "https://bizcloudexperts.com/pdfs/alliance-air-freight.pdf",
   "prompt": "Read this quote request email, extract the shipment details, fetch rates, and draft a reply for me to approve. Do not send anything yourself.",
   "added": "2026-10-29",
   "interest": 53,
   "page": "https://agentsightings.com/case/freight-agent-quotes-and-reconciles-invoices.html"
  },
  {
   "id": "ulta-ai-shopping-assistant-triples-conversion",
   "title": "A beauty shopping assistant that skips the search filters",
   "summary": "Ulta Beauty's AI shopping assistant, built with Gemini Enterprise, turns conversational questions into recommendations across 30,000 products. Google says digital sales conversion is up three-fold. Vendor-reported.",
   "agent": "Ulta AI shopping assistant (Gemini Enterprise)",
   "topic": "Shopping",
   "source": "News (Google Cloud)",
   "url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
   "prompt": "I need a moisturiser for dry skin under $30 that works under makeup. Ask me what you need to know, then recommend three products from the catalogue with reasons.",
   "added": "2026-10-29",
   "interest": 53,
   "page": "https://agentsightings.com/case/ulta-ai-shopping-assistant-triples-conversion.html"
  },
  {
   "id": "triage-support-tickets-and-draft-replies",
   "title": "Triage support tickets and draft replies",
   "summary": "According to the write-up, the skill interviews you about your support stack, then writes TypeScript scripts for your ticketing, billing and database APIs, plus response templates. Once assembled, the agent lists and triages every open ticket, suggests the most urgent one, pulls billing and customer context, and sketches a reply. For the Lex setup (HelpScout, Stripe, PostgreSQL), The builder assembled 15 scripts and 6 templates.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "GitHub",
   "url": "https://github.com/nbashaw/claude-cs",
   "prompt": "Help me triage support tickets and draft replies. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/triage-support-tickets-and-draft-replies.html"
  },
  {
   "id": "track-food-and-symptoms-in-a-health-log-via-telegram",
   "title": "Track food and symptoms in a health log via Telegram",
   "summary": "As described in the source, you message a Telegram topic with what you ate or how you feel. The agent appends a timestamped entry to a markdown health log in its memory folder. The article gives no measured results.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://www.datacamp.com/blog/openclaw-projects",
   "prompt": "Assistant mode: track food and symptoms in a health log via Telegram. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/track-food-and-symptoms-in-a-health-log-via-telegram.html"
  },
  {
   "id": "screen-115-job-candidates",
   "title": "Screen 115 job candidates",
   "summary": "Documented elsewhere: claude Code shows up in a write-up about screen 115 job candidates, with figures like $65. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://happycapy.ai/blog/hiring-dynamic-workflow",
   "prompt": "Notes task: screen 115 job candidates. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/screen-115-job-candidates.html"
  },
  {
   "id": "schedule-meetings-by-email-from-forwarded-requests",
   "title": "Schedule meetings by email from forwarded requests",
   "summary": "According to the write-up, user forwards meeting request emails to a labeled Gmail inbox or creates one. The AI agent takes over from there, sends emails to recipients to propose and coordinate times, and inserts a Zoom link into the calendar event when needed. It works for the described single-person cases after several iterations powered by OpenAI's Dot.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/edsoehnel/status/2107574500984287570",
   "prompt": "Go through my mail for this goal: schedule meetings by email from forwarded requests. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/schedule-meetings-by-email-from-forwarded-requests.html"
  },
  {
   "id": "run-claude-on-a-cronjob-for-morning-briefings-and-email-tasks",
   "title": "Run Claude on a cronjob for morning briefings and email tasks",
   "summary": "Documented elsewhere: claude shows up in a write-up about run Claude on a cronjob for morning briefings and email tasks. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47724491",
   "prompt": "Go through my mail for this goal: run Claude on a cronjob for morning briefings and email tasks. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/run-claude-on-a-cronjob-for-morning-briefings-and-email-tasks.html"
  },
  {
   "id": "look-up-user-records-and-update-downstream-steps",
   "title": "Look up user records and update downstream steps",
   "summary": "Case study takeaway: the poster set up three Zapier agents for a subscription product: one to email admins a current user list, one to add users, and one to remove them, powered by a Zapier Table as the customer database. The agents looked up customers by the email Stripe had written into the table, then later needed to search by another field to find linked subaccounts; the person tried both changing the Find Records operator and adding a lowercase-normalized formula column, but hit bugs and missing-field issues. It never became reliable: case-sensitive lookups, inability to set the needed operator when multiple lookup fields were required, unseen new columns, and draft/version errors blocked the workflow.",
   "agent": "Zapier Agents",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/zapier/comments/1rmgfwu/issues_with_zapier_agents",
   "prompt": "Automation design: look up user records and update downstream steps. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/look-up-user-records-and-update-downstream-steps.html"
  },
  {
   "id": "handle-meal-planning-daily-briefs-errands-and-household-tasks",
   "title": "Handle meal planning, daily briefs, errands and household tasks",
   "summary": "A published walkthrough shows custom agent shows up in a write-up about handle meal planning, daily briefs, errands and household tasks. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1tolh94/anyone_who_regularly_uses_ai_agents_for_personal",
   "prompt": "Go through my mail for this goal: handle meal planning, daily briefs, errands and household tasks. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/handle-meal-planning-daily-briefs-errands-and-household-tasks.html"
  },
  {
   "id": "find-stale-commercial-listings-and-email-brokers",
   "title": "Find stale commercial listings and email brokers",
   "summary": "Documented elsewhere: the Hermes bot scrapes every commercial lease listing in real time. It filters by days on market, suite sqft, and zoning fit; pulls permitted-use lists from city zoning code; renders 12 AI walkthroughs of each empty suite as different tenants; calculates broker commission plus expected 5-year lease value; then sends a personalized email containing the walkthroughs. The post states the system enables brokers to close $15k+ commissions but reports no actual results, metrics, or outcomes achieved.",
   "agent": "Hermes Agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/everestchris6/status/2047781663145677304",
   "prompt": "Go through my mail for this goal: find stale commercial listings and email brokers. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/find-stale-commercial-listings-and-email-brokers.html"
  },
  {
   "id": "email-car-brands-without-web-apps-and-build-them-one",
   "title": "Email car brands without web apps and build them one",
   "summary": "A published walkthrough shows someone prompted Manus AI to locate car brands lacking web apps, draft emails to them, and send the messages. Manus covered the search and emailing on its own. The brands replied, and Someone assembled a web app for one.",
   "agent": "Manus",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/JerryJenasom/status/2102306956203442534",
   "prompt": "Go through my mail for this goal: email car brands without web apps and build them one. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/email-car-brands-without-web-apps-and-build-them-one.html"
  },
  {
   "id": "draft-lesson-plans-and-school-paperwork-for-teachers",
   "title": "Draft lesson plans and school paperwork for teachers",
   "summary": "Real-world thread: a teacher used ChatGPT as an administrative assistant for lesson sequences, reflections, PowerPoints, programs, assessment notifications, marking, and even draft student reports. They gave it prompts plus short assessment notes, then checked and edited the output themselves; in comments, they stressed that everything still needed vetting. The teacher says it saved them hours of planning and cut admin time enough to focus more on teaching.",
   "agent": "ChatGPT",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Teachers/comments/ztrpnc/im_a_teacher_and_i_love_chatgpt",
   "prompt": "Research brief: draft lesson plans and school paperwork for teachers. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/draft-lesson-plans-and-school-paperwork-for-teachers.html"
  },
  {
   "id": "draft-follow-up-emails-automatically-when-no-reply-arrives",
   "title": "Draft follow-up emails automatically when no reply arrives",
   "summary": "Case study takeaway: one commenter said they set up a Lindy flow tied to their sent email so that after each outgoing message, the AI waits a set amount of time and checks for a reply. If no response arrives, it sketches a follow-up and sends that draft to Someone's inbox, where the person decides whether to send it. Someone said they had started powered by it the day before and that it was working well so far.",
   "agent": "Lindy",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/EntrepreneurRideAlong/comments/1awb7xj/has_anyone_here_actually_used_lindy_or_similar_ai",
   "prompt": "Go through my mail for this goal: draft follow-up emails automatically when no reply arrives. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/draft-follow-up-emails-automatically-when-no-reply-arrives.html"
  },
  {
   "id": "delegate-to-do-items-to-ai-agents",
   "title": "Delegate to-do items to AI agents",
   "summary": "One practitioner notes that someone keeps a shared task list that both people and AI agents can access. A person assigns a task to one of the bots; the agent claims it, updates the status to in progress while working, moves it to done when finished, and returns it to the review inbox. Tasks flow back and forth between Someone and agents as described, with no numbers or performance details described.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/petecallaghans/status/2104948587041837102",
   "prompt": "Go through my mail for this goal: delegate to-do items to AI agents. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/delegate-to-do-items-to-ai-agents.html"
  },
  {
   "id": "automate-invoice-organizing-and-monthly-invoice-emails",
   "title": "Automate invoice organizing and monthly invoice emails",
   "summary": "From the linked report, the builder opened a folder of invoices with inconsistent names and formats in Cursor and gave instructions. The agent read the files, renamed them and moved them into folders. The builder says the tasks, about 10 minutes a month for filing and 10 to 20 minutes for invoice emails, can be compressed.",
   "agent": "Cursor",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://note.com/minicoohei/n/n892b82b65f7b?hl=en",
   "prompt": "Finance task: automate invoice organizing and monthly invoice emails. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/automate-invoice-organizing-and-monthly-invoice-emails.html"
  },
  {
   "id": "answer-shopify-order-status-emails-and-fix-issues-overnight",
   "title": "Answer Shopify order-status emails and fix issues overnight",
   "summary": "As described in the source, for a Shopify store the owner assembled an agent to read customer emails about orders. The agent pulls the real order details from the store and resolves the issues on its own while people sleep. It handled \"where's my order?\" queries before the owner woke up.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/lukeidiot/status/2108452907620155851",
   "prompt": "Help me answer Shopify order-status emails and fix issues overnight. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 53,
   "page": "https://agentsightings.com/case/answer-shopify-order-status-emails-and-fix-issues-overnight.html"
  },
  {
   "id": "orange-spain-staff-build-1000-gemini-agents",
   "title": "Orange Spain staff built 1,000+ custom agents themselves",
   "summary": "Orange Spain says employees built more than 1,000 custom Gemini Enterprise agents across HR, IT, sales and customer service, with near-100% licence activity within 30 days. Uses include incident triage, root-cause analysis of system logs and finding internal knowledge.",
   "agent": "Gemini Enterprise agents",
   "topic": "Productivity",
   "source": "News (Google Cloud)",
   "url": "https://www.googlecloudpresscorner.com/2026-10-08-Orange-Spain-Expands-Partnership-with-Google-Cloud-to-Industrialize-AI-Operations-with-Gemini-Enterprise",
   "prompt": "Build me a small agent for my team that reads our incident tickets, groups them by type, checks the system logs for a likely cause and routes each one to the right person.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/orange-spain-staff-build-1000-gemini-agents.html"
  },
  {
   "id": "crocs-rivet-shopping-superagent",
   "title": "Crocs shoppers get a chat agent for returns and checkout",
   "summary": "Crocs built Rivet, a retail agent on Agentforce, from concept to customer-ready in five months. It launched with five use cases: product discovery, FAQs, order tracking, returns and conversational checkout.",
   "agent": "Rivet (Agentforce)",
   "topic": "Shopping",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/crocs/building-ai-agents/",
   "prompt": "Act as a shop assistant on our store. Help visitors find a product, answer FAQs, check where an order is, start a return, and check out in chat. Hand over to a person when you are unsure.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/crocs-rivet-shopping-superagent.html"
  },
  {
   "id": "ge-appliances-supplier-agent-cuts-backorders",
   "title": "An appliance maker's supplier agent talks to 600 parts suppliers",
   "summary": "Google says GE Appliances staff built 800+ agents, including one that handles routine messages with 600+ service-parts suppliers and has cut backorders by 25%. Others summarise shifts in minutes and trace part genealogy. Vendor-published.",
   "agent": "Supplier Collaboration Agent (Gemini Enterprise)",
   "topic": "Business & Startups",
   "source": "Blog (Google Cloud)",
   "url": "https://cloud.google.com/transform/agentic-factory-manufacturing-new-age-of-autonomy-industrial-ai",
   "prompt": "Check open parts orders, message suppliers whose deliveries are late or unconfirmed, and list the ones that need a person.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/ge-appliances-supplier-agent-cuts-backorders.html"
  },
  {
   "id": "ginger-ai-chief-of-staff-never-spends-money",
   "title": "An AI chief of staff with two hard rules: never spend, never delete",
   "summary": "Mike Cali, a founder, says Ginger runs on a small Ubuntu server doing email triage, research, scheduling, drafts and follow-ups, and answers the phone. She has a written charter and may never spend money or delete anything without asking. When lead generation crowded her out, he built a second agent. Self-reported post.",
   "agent": "Ginger (personal chief-of-staff agent)",
   "topic": "Personal Assistant",
   "source": "LinkedIn",
   "url": "https://www.linkedin.com/posts/mike-cali_aiagents-automation-productivity-activity-7488220118976733185-Yj4a",
   "prompt": "You are my chief of staff. Triage my email, draft replies, schedule meetings and track follow-ups. You may never spend money or delete anything without asking me first.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/ginger-ai-chief-of-staff-never-spends-money.html"
  },
  {
   "id": "tata-steel-300-agents-nine-months",
   "title": "A steelmaker deployed 300+ agents in nine months",
   "summary": "Tata Steel says it deployed 300+ specialised agents in nine months with Google Cloud, covering maintenance prediction and customer response times. Google's own blog adds that complaint-handling agents cut turnaround by 50%; the press release itself gives no baseline. Vendor-published.",
   "agent": "Gemini Enterprise agents",
   "topic": "Business & Startups",
   "source": "Press release (Tata Steel)",
   "url": "https://www.tatasteel.com/newsroom/press-releases/india/2026/tata-steel-partners-with-google-cloud-to-deploy-a-unified-agentic-ai-across-its-global-value-chain/",
   "prompt": "Read each customer complaint with its photos and documents, work out the defect and who should fix it, and send it to that team.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/tata-steel-300-agents-nine-months.html"
  },
  {
   "id": "bcu-credit-union-agentforce-82-percent",
   "title": "A credit union's agent resolves 82% of chats",
   "summary": "Salesforce says BCU, a US credit union with 360,000+ members, now has an Agentforce agent with an 82% resolution rate, 97% engagement and under 3% abandoned sessions. Escalated cases are handled 27% faster and escalations are down 15%. Vendor-published.",
   "agent": "Agentforce",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/bcu/",
   "prompt": "Answer members' account and service questions in chat, resolve what you can, and hand the rest to a person with the whole conversation attached.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/bcu-credit-union-agentforce-82-percent.html"
  },
  {
   "id": "barclays-claude-sorts-global-markets-emails",
   "title": "A bank's assistant sorts client emails and answers 16,000 colleagues",
   "summary": "Anthropic says Barclays uses Claude to classify incoming client emails in Global Markets and choose the right processing route. Its Colleague Knowledge Assistant, live since 2025, has 16,000+ users and over a million searches. Barclays expects Claude Code use by 50% of developers by end of 2026. Vendor-published.",
   "agent": "Claude (Barclays knowledge assistant, email triage)",
   "topic": "Finance",
   "source": "News (Anthropic)",
   "url": "https://www.anthropic.com/news/barclays-scales-claude",
   "prompt": "Read each incoming client email, work out what is being asked, add any missing details, and send it to the right operations queue. Flag anything unclear.",
   "added": "2026-10-29",
   "interest": 52,
   "page": "https://agentsightings.com/case/barclays-claude-sorts-global-markets-emails.html"
  },
  {
   "id": "ai-welding-robots-rented-as-a-service",
   "title": "AI welding robots rented out as a service",
   "summary": "Path Robotics puts its own AI-driven welding robots into factories, with about 50 manufacturing customers so far. Instead of a big upfront purchase, customers pay for them as Robotics-as-a-Service, so the savings start straight away and Path carries more of the risk if the robots underperform.",
   "agent": "Path Robotics",
   "topic": "Automation",
   "source": "ZeroHedge (BTIG note)",
   "url": "https://www.zerohedge.com/technology/btig-spots-real-world-evidence-humanoid-robots-are-entering-workforce",
   "prompt": "Compare buying this machine outright with renting it as a service. Lay out the upfront cost, the monthly cost, who carries the risk if it underperforms, and when each option pays for itself.",
   "added": "2026-10-19",
   "interest": 52,
   "page": "https://agentsightings.com/case/ai-welding-robots-rented-as-a-service.html"
  },
  {
   "id": "turn-a-spreadsheet-into-a-meta-ads-account",
   "title": "Turn a spreadsheet into a Meta ads account",
   "summary": "Real-world thread: joel Jackson Leaned on the Manus agent for spreadsheet to ads account in Meta. The agent performed the conversion directly because of its native integration with Meta. The post states no results, numbers, or shortcomings.",
   "agent": "Manus",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/joeljackson/status/2033589095692410921",
   "prompt": "Help me turn a spreadsheet into a Meta ads account. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/turn-a-spreadsheet-into-a-meta-ads-account.html"
  },
  {
   "id": "spot-recurring-meetings-emails-and-bills-and-offer-to-prep-them",
   "title": "Spot recurring meetings, emails and bills and offer to prep them",
   "summary": "Observed in the wild: custom agent shows up in a write-up about spot recurring meetings, emails and bills and offer to prep them. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AIPersonalAssistant/comments/1t1wo9v/how_to_turn_your_ai_into_a_personal_assistant",
   "prompt": "Go through my mail for this goal: spot recurring meetings, emails and bills and offer to prep them. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/spot-recurring-meetings-emails-and-bills-and-offer-to-prep-them.html"
  },
  {
   "id": "spot-conflict-avoidance-in-meeting-recordings",
   "title": "Spot conflict avoidance in meeting recordings",
   "summary": "Dan Shipper downloads meeting recordings into a folder and asks Claude Code to find moments where he subtly avoided conflict.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Read every transcript in this folder. List moments where I avoided conflict, softened bad news, or deferred a hard decision. Quote the line and say what a direct version would have been.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/spot-conflict-avoidance-in-meeting-recordings.html"
  },
  {
   "id": "scrape-real-estate-agent-leads-and-start-roofing-outreach",
   "title": "Scrape real estate agent leads and start roofing outreach",
   "summary": "As described in the source, someone prompted his OpenClaw agent Cody via Discord on his phone from a plane. Cody scraped Zillow, Redfin, and Realtor.com for agents listing homes assembled 1965-2000 in the service area, pulling brokerage, phone, email, active listings, addresses, and year assembled into a CSV saved to Google Drive. A CSV arrived in Google Drive.",
   "agent": "OpenClaw",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/pricefoulger/status/2033271956829806891",
   "prompt": "GTM task: scrape real estate agent leads and start roofing outreach. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/scrape-real-estate-agent-leads-and-start-roofing-outreach.html"
  },
  {
   "id": "run-warehouse-paperwork-sows-and-billing-automatically",
   "title": "Run warehouse paperwork, SOWs and billing automatically",
   "summary": "A published walkthrough shows the team assembled one custom operating system (130k lines of code, 41 screens, 50 tables) after mapping every workflow by meeting the staff doing the work, removing unneeded processes, and wireframing the full flow. The AI reads inbound receipts, estimates, and invoices then structures the data automatically; produces SOWs in one click; runs billing across three rate models; and triggers 28 background automations for reminders, emails, and documents while managing all 450 active projects in one place. Operations consolidated into a single platform.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/lukepierceops/status/2076005116587725099",
   "prompt": "Help me run warehouse paperwork, SOWs and billing automatically. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/run-warehouse-paperwork-sows-and-billing-automatically.html"
  },
  {
   "id": "run-outbound-sales-from-research-to-booked-demos",
   "title": "Run outbound sales from research to booked demos",
   "summary": "Observed in the wild: solo founder gave OpenAI Dots his ICP, calendar and inbox plus one prompt. The agent ran recurring jobs: scholar researched news/hiring/funding/tech for targets, watcher spotted signals, bouncer disqualified before outreach, builder wrote one fact-based first line per email, manager handled replies with calendar slots and produced one-page briefs. Over four days it researched 300 companies, qualified 61, sent emails, got 18 replies and booked 14 demos for $186k pipeline; nine were disqualified for size or budget.",
   "agent": "Dots",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/tekari0/status/2109630986220863599",
   "prompt": "GTM task: run outbound sales from research to booked demos. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/run-outbound-sales-from-research-to-booked-demos.html"
  },
  {
   "id": "run-a-personal-doorman-agent-on-a-7-vps",
   "title": "Run a personal doorman agent on a $7 VPS",
   "summary": "Documented elsewhere: george Larson ran nullclaw, a small Zig agent, on a cheap VPS. It joined his own IRC server, which visitors reached through a web client. It worked for visitors, with Haiku handling chat and Sonnet handling repo reading.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47536761",
   "prompt": "Go through my mail for this goal: run a personal doorman agent on a $7 VPS. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/run-a-personal-doorman-agent-on-a-7-vps.html"
  },
  {
   "id": "receipt-hunter-across-two-inboxes",
   "title": "Receipt hunter across two inboxes",
   "summary": "A Claude Code agent scans Gmail and Outlook for receipts, removes duplicates, logs expenses, and remembers vendor categories.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "LinkedIn",
   "url": "https://www.linkedin.com/posts/samehabdultawab_claudecode-ai-automation-activity-7442642724924727296-Ykas",
   "prompt": "Scan my inboxes for receipts and invoices since last run. De-duplicate, categorize by vendor using past choices, log them to my expense sheet, and summarize totals.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/receipt-hunter-across-two-inboxes.html"
  },
  {
   "id": "reading-list-turned-into-a-podcast",
   "title": "Reading list turned into a podcast",
   "summary": "A workflow takes saved articles, summarizes them with an LLM and converts them to audio episodes via text-to-speech.",
   "agent": "n8n",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=42628324",
   "prompt": "Take the articles I saved this week, write a 10-minute conversational script covering the key points of each, ready for text-to-speech.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/reading-list-turned-into-a-podcast.html"
  },
  {
   "id": "monitor-investing-news-and-produce-morning-macro-summaries",
   "title": "Monitor investing news and produce morning macro summaries",
   "summary": "Linked example of monitor investing news and produce morning macro summaries. User configured Meta Muse agent Alfred with ongoing tasks to monitor and summarize X posts on specific investing topics and watchlist names, transcribe and summarize podcasts, scan news and regulatory developments for those names, and compile a general morning macro and business news summary.",
   "agent": "Meta Muse",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/SouthernValue95/status/2098391130765246613",
   "prompt": "Assistant mode: monitor investing news and produce morning macro summaries. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/monitor-investing-news-and-produce-morning-macro-summaries.html"
  },
  {
   "id": "manage-customer-enquiries-from-telegram",
   "title": "Manage customer enquiries from Telegram",
   "summary": "According to the write-up, n8n workflow captures enquiries from email or forms, assigns a unique ID, auto-creates a ClickUp task, and pushes an instant Telegram notification. Owner replies naturally in Telegram; OpenAI then reads the reply and chooses to send a Gmail response, update status in Google Sheets, or close the thread. Enquiries stay in one place with no lost context or tool switching, letting the owner reply faster while staying in full control instead of full automation.",
   "agent": "n8n",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/Toma_web3/status/2071843246427492806",
   "prompt": "Assistant mode: manage customer enquiries from Telegram. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/manage-customer-enquiries-from-telegram.html"
  },
  {
   "id": "make-music-videos-portfolio-sites-and-sorting-scripts",
   "title": "Make music videos, portfolio sites and sorting scripts",
   "summary": "As described in the source, over one week the creator gave Grok Bot successive tasks: generate a complete music video from song through final edited clip; research Grok Imagine limits then output a reusable cinematic prompt engine; write full portfolio site code in HTML/CSS/JS with filters, mobile layout and live preview; write scripts that pulled 158 AI images/videos from X posts and auto-sorted them; plus code to cut/trim/stitch video clips and convert a theatre résumé into a career page. Someone directed the work and maintained separate agents for film, prompts and chief-of-staff roles. All requested outputs were delivered and replaced the studio, developer and assistant previously needed.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/Z_XSophie/status/2108240361197826430",
   "prompt": "Creative task: make music videos, portfolio sites and sorting scripts. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/make-music-videos-portfolio-sites-and-sorting-scripts.html"
  },
  {
   "id": "invoice-extraction-with-validation-checkpoints",
   "title": "Invoice extraction with validation checkpoints",
   "summary": "An n8n workflow enforces a schema on AI-extracted invoice data, checks formats and nulls, and waits for human sign-off before finalizing.",
   "agent": "n8n",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://jyoti03.medium.com/your-invoice-bot-is-lying-to-you-heres-how-to-make-it-stop-1f7a6f09c952",
   "prompt": "Extract invoice fields into this JSON schema. If any required field is missing or totals don't add up, return an error object explaining why instead of guessing.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/invoice-extraction-with-validation-checkpoints.html"
  },
  {
   "id": "generate-lego-cad-models-in-ldraw",
   "title": "Generate LEGO CAD models in LDraw",
   "summary": "Documented elsewhere: the creation of LDraw source files, which are then converted into LEGO CAD models, is facilitated by ChatGPT and Claude agents through a toolset and docs, and is packaged as a dockerized web application.",
   "agent": "ChatGPT",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49937917",
   "prompt": "Shopping task: generate LEGO CAD models in LDraw. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/generate-lego-cad-models-in-ldraw.html"
  },
  {
   "id": "fix-app-bugs-harden-security-and-ship-fixes-autonomously",
   "title": "Fix app bugs, harden security and ship fixes autonomously",
   "summary": "Omar’s health app got a Hermes babysitter during work hours: triage priority bugs, fence personal browser logins away from bots, sketch a tax-refund promo banner, promote patches only after green health checks.",
   "agent": "Hermes Agent",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/omx0r/status/2109648184565764501",
   "prompt": "On this health app, fix the priority bugs, harden browser/session isolation for automation, and prepare production fixes with health checks. Pause before any production push.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/fix-app-bugs-harden-security-and-ship-fixes-autonomously.html"
  },
  {
   "id": "create-a-presentation-on-ai-career-and-salary-trends",
   "title": "Create a presentation on AI career and salary trends",
   "summary": "Observed in the wild: in agent mode, the tester gave ChatGPT Agent a detailed prompt for a 10-15 slide PowerPoint on career and salary growth in generative AI. The prompt listed an 11-slide outline, with charts, icons and minimal text. The deck came out very basic in content and design, and its tables were hard to read.",
   "agent": "ChatGPT Agent",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://www.analyticsvidhya.com/blog/2025/07/chatgpt-agent/",
   "prompt": "Creative task: create a presentation on AI career and salary trends. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/create-a-presentation-on-ai-career-and-salary-trends.html"
  },
  {
   "id": "book-a-kid-x27-s-dental-checkup-by-phone-agent",
   "title": "Book a kid's dental checkup by phone agent",
   "summary": "Observed in the wild: a group of eight AI phone-call agents were tested to book a dental checkup on weekdays and following 3pm, powered by artificial intelligence.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1wuroa2/best_ai_phone_agent_round_2_i_put_6_of_them_on",
   "prompt": "Help me book a kid's dental checkup by phone agent. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 52,
   "page": "https://agentsightings.com/case/book-a-kid-x27-s-dental-checkup-by-phone-agent.html"
  },
  {
   "id": "ten-agents-on-a-mac-mini-zero-revenue",
   "title": "Ten agents on a Mac mini ran 200 sessions a day and earned $0",
   "summary": "A developer says his OpenClaw setup of ten agents, covering email triage, freelance proposals, bounty hunting, content and Discord bots, ran 200+ sessions a day on a Mac mini. After five days revenue was $0. He says the agents worked fine; they were just selling nothing. Self-reported.",
   "agent": "OpenClaw (10 agents)",
   "topic": "Automation",
   "source": "Blog (DEV Community)",
   "url": "https://dev.to/quantbit/i-run-10-ai-agents-247-on-a-mac-mini-heres-what-actually-happened-4did",
   "prompt": "Check my two Gmail accounts every 20 minutes, categorise the new mail and pass urgent items to me. Do not reply to anything yourself.",
   "added": "2026-10-29",
   "interest": 51,
   "page": "https://agentsightings.com/case/ten-agents-on-a-mac-mini-zero-revenue.html"
  },
  {
   "id": "bunnings-agent-saves-half-a-million-admin-hours",
   "title": "A hardware chain's internal agent saved 500,000 hours",
   "summary": "Google says an internal agent at Bunnings, part of Wesfarmers, saved staff half a million hours of administrative work. Shopping agents that help customers find products and build carts raised conversion up to 3x for Kmart and Officeworks. Vendor-reported.",
   "agent": "Wesfarmers agents (Gemini Enterprise)",
   "topic": "Productivity",
   "source": "News (Google Cloud)",
   "url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
   "prompt": "Take over the repetitive admin in my store job: read the request, fill in the forms and update the systems, and flag anything that needs a manager.",
   "added": "2026-10-29",
   "interest": 51,
   "page": "https://agentsightings.com/case/bunnings-agent-saves-half-a-million-admin-hours.html"
  },
  {
   "id": "prince-william-county-will-311-assistant",
   "title": "A county's 311 assistant handled 40,000 conversations in year one",
   "summary": "Civic Marketplace reports Prince William County, Virginia, put Citibot's assistant 'Will' on its 311 line in May 2025. In year one Will handled nearly 40,000 of about 90,000 resident interactions, mostly tax and real-estate questions. Vendor-influenced case study, October 2026.",
   "agent": "Will (Citibot)",
   "topic": "Customer Support",
   "source": "News (Civic Marketplace)",
   "url": "https://www.civicmarketplace.com/news/how-prince-william-county-handled-nearly-40-000-resident-conversations-with-ai-in-its-first-year",
   "prompt": "Answer residents' tax and real-estate questions day or night, and transfer to a live agent when the question is outside what you can answer.",
   "added": "2026-10-29",
   "interest": 51,
   "page": "https://agentsightings.com/case/prince-william-county-will-311-assistant.html"
  },
  {
   "id": "engine-travel-agent-eva-handles-cancellations",
   "title": "A business-travel platform's agent cancels bookings for people",
   "summary": "Engine's agent Eva handles hotel, flight and car-rental changes by chat and voice. Salesforce says it resolves 50% of chat inquiries, cuts handle time 15% and raises CSAT 16%. Engine's team handles 800,000+ requests a year. Vendor-published, June 2026.",
   "agent": "Eva (Agentforce)",
   "topic": "Travel & Booking",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/engine/agentic-service/",
   "prompt": "A traveller asks to cancel or change a booking. Look it up, make the change if the policy allows, and route complex group cases to a rep with context.",
   "added": "2026-10-29",
   "interest": 51,
   "page": "https://agentsightings.com/case/engine-travel-agent-eva-handles-cancellations.html"
  },
  {
   "id": "resolve-a-declined-flight-payment-from-email",
   "title": "Resolve a declined flight payment from email",
   "summary": "Linked example of resolve a declined flight payment from email. The agent spotted the declined flight payment email, connected it to a separate notice about a new debit card, told Someone exactly how to activate the card on the bank site, confirmed the payment succeeded, and added the booking to the calendar covering PNR, seat numbers, and check-in time.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/parrysingh/status/2104519011602137209",
   "prompt": "Go through my mail for this goal: resolve a declined flight payment from email. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 51,
   "page": "https://agentsightings.com/case/resolve-a-declined-flight-payment-from-email.html"
  },
  {
   "id": "research-purchases-for-delivered-price-and-working-coupons",
   "title": "Research purchases for delivered price and working coupons",
   "summary": "Documented elsewhere: the poster runs this self-hosted on a home server with a Claude subscription. Claude Code orchestrates parallel subagents to check review sites and YouTube, read Reddit, analyze video transcripts and frames, find listings, and verify checkout; before showing the report, an audit step sends the agent back if key checks like Amazon or owner reviews are missing. It condenses the shopping research the poster used to do by hand into one report focused on delivered price, valid codes, merchant fit for Finland, and owner feedback.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SideProject/comments/1wukkk1/built_an_ai_agent_that_researches_my_purchases",
   "prompt": "Research brief: research purchases for delivered price and working coupons. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 51,
   "page": "https://agentsightings.com/case/research-purchases-for-delivered-price-and-working-coupons.html"
  },
  {
   "id": "flag-churn-and-expansion-signals-and-draft-support-replies",
   "title": "Flag churn and expansion signals and draft support replies",
   "summary": "As described in the source, healthie assembled a Zapier agent that checks Salesforce, HubSpot, Vitally and Help Scout every Monday for early churn or expansion signals. It posts a prioritized Slack summary for support and product leads to act on. Erewhon says 70% of sketches go out unchanged, saving about 1,500 labor hours a year across 10 stores.",
   "agent": "Zapier Agents",
   "topic": "Automation",
   "source": "Blog",
   "url": "https://zapier.com/blog/ai-agent-use-cases/",
   "prompt": "Automation design: flag churn and expansion signals and draft support replies. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 51,
   "page": "https://agentsightings.com/case/flag-churn-and-expansion-signals-and-draft-support-replies.html"
  },
  {
   "id": "fight-consumer-claims-and-recover-refunds",
   "title": "Fight consumer claims and recover refunds",
   "summary": "Real-world thread: the builder describes each problem by voice-to-text. Claude gathers context and returns a dossier covering the contract, the company's marketing, owner reports and the relevant regulator, plus a multi-step plan. Ten fights, no losses after seven months, with over $12,000 recovered.",
   "agent": "Claude",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://www.sudomoin.com/p/consumer-claims",
   "prompt": "GTM task: fight consumer claims and recover refunds. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 51,
   "page": "https://agentsightings.com/case/fight-consumer-claims-and-recover-refunds.html"
  },
  {
   "id": "chase-unquoted-leads-and-book-callbacks",
   "title": "Chase unquoted leads and book callbacks",
   "summary": "Case study takeaway: a commenter described this as a service-business pattern they see people run: the agent watches the inbox or CRM, checks for leads that have gone 24 hours without a quote, and follows up to book callbacks. The same comment frames it as a recurring outreach/admin workflow rather than a one-off chat task, with the person choosing and driving the setup. The commenter says this is the biggest-ROI use case they see because service leads go cold fast.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1wy0v7z/reasons_to_buy",
   "prompt": "Go through my mail for this goal: chase unquoted leads and book callbacks. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 51,
   "page": "https://agentsightings.com/case/chase-unquoted-leads-and-book-callbacks.html"
  },
  {
   "id": "auto-buy-items-when-the-price-hits-a-target",
   "title": "Auto-buy items when the price hits a target",
   "summary": "Community report: in the Amazon app or website, you tap the cursive A icon to open Alexa for Shopping. On a product page you tell it a target price, such as a percentage off. The article describes the features but doesn't report a hands-on test or savings figures.",
   "agent": "Alexa+",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://www.techlicious.com/blog/alexa-for-shopping-price-tracking-auto-buy/",
   "prompt": "Shopping task: auto-buy items when the price hits a target. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 51,
   "page": "https://agentsightings.com/case/auto-buy-items-when-the-price-hits-a-target.html"
  },
  {
   "id": "asurion-copilot-studio-touchless-invoices",
   "title": "Touchless invoice processing rises from under 20% to over 70%",
   "summary": "Microsoft's customer story says Asurion used Copilot Studio to get invoice automation back on track, lifting touchless invoice processing from under 20% to over 70%. Vendor-published.",
   "agent": "Copilot Studio agents",
   "topic": "Finance",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/27456-asurion-insurance-services-microsoft-copilot-studio",
   "prompt": "Read each supplier invoice, match it to the purchase order, and approve it automatically when it matches. Send anything that does not match to a person with the reason.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/asurion-copilot-studio-touchless-invoices.html"
  },
  {
   "id": "hitachi-field-technicians-build-200-agents",
   "title": "Field technicians build 200+ agents, including a photo check of every wire",
   "summary": "Google says Hitachi gave frontline power-and-utilities workers agent-building tools, and technicians built agents for 200+ use cases. One checks photos of every tool and wire before and after a job. Hitachi cites productivity up to 30%. Vendor-published.",
   "agent": "Hitachi HMAX agents (Gemini Enterprise)",
   "topic": "Business & Startups",
   "source": "Blog (Google Cloud)",
   "url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
   "prompt": "Check the before and after photos for this job: is every tool and wire accounted for, and does anything look wrong?",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/hitachi-field-technicians-build-200-agents.html"
  },
  {
   "id": "southwest-agentforce-45-percent-inquiries",
   "title": "An airline's agent resolves 45% of eligible inquiries",
   "summary": "Salesforce says Southwest Airlines' Agentforce deployment has handled 2 million+ customer interactions with a 45% case resolution rate, and projects $6 million a year in savings. Its page also claims a 900% rise in customer satisfaction, with no baseline given. Vendor-published.",
   "agent": "Agentforce (Agent Script)",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/southwest-airlines/agentic-self-service/",
   "prompt": "Answer customers' travel questions, resolve what the rules let you, and pass the rest to a human agent with a summary.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/southwest-agentforce-45-percent-inquiries.html"
  },
  {
   "id": "ktern-agents-speed-up-sap-projects",
   "title": "Agents find the exceptions in SAP migration projects",
   "summary": "KTern.AI builds agents that analyse SAP systems. It reports they cut overall SAP project timelines by 45%, discovery and assessment time by 60-70%, and surface 90% of finance and sales exceptions on their own. These are the company's own measurements.",
   "agent": "KTern.AI agents (Amazon Bedrock AgentCore)",
   "topic": "Business & Startups",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/how-ktern-ai-built-agentic-ai-for-sap-on-amazon-bedrock-agentcore/",
   "prompt": "Scan our ERP data for finance and sales exceptions, group them by cause, and tell me which ones need a person. List what you checked so I can verify it.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/ktern-agents-speed-up-sap-projects.html"
  },
  {
   "id": "three-day-agent-saves-startup-four-hours-a-week",
   "title": "A three-day agent replaced a startup's manual triage chore",
   "summary": "A consultant built an agent in about three days for a growing startup. It reads incoming items, classifies them, extracts the key details, takes the routine action and logs it. By the client's estimate it cuts about 4.5 hours of weekly work to 0.5 hours of checking, with a person kept for odd cases.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Blog (Substack)",
   "url": "https://learneasyai.substack.com/p/i-buit-an-agent-in-3-days-saves-5",
   "prompt": "Take this repetitive six-step process: read what comes in, classify it, extract the details, decide the next action, do it in the system, record it. Build an agent for the routine cases and flag anything unusual to me for a check.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/three-day-agent-saves-startup-four-hours-a-week.html"
  },
  {
   "id": "santee-cooper-agent-grid-scenarios",
   "title": "A power utility builds financial scenarios in hours, not weeks",
   "summary": "Google says Santee Cooper's planning agent should build scenarios in hours or days that used to take a small team weeks or months, feeding a $10 billion grid-expansion budget. Another use is grid load forecasting. A scenario-time cut of up to 75% appears in Google's summary. Vendor-published.",
   "agent": "Gemini Enterprise Agent Platform",
   "topic": "Finance",
   "source": "Press release (Google Cloud)",
   "url": "https://www.googlecloudpresscorner.com/2026-09-03-Santee-Cooper-Partners-with-Google-Cloud-to-Help-Improve-Grid-Reliability-Through-Better-Planning",
   "prompt": "Take this spreadsheet model, run these three what-if scenarios, and give me the results side by side with every assumption listed.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/santee-cooper-agent-grid-scenarios.html"
  },
  {
   "id": "nexo-crypto-support-agentforce-91400-chats",
   "title": "A crypto platform's agent resolved 91,400 chats",
   "summary": "Salesforce says Nexo, with 7 million customers in 200+ jurisdictions, uses Agentforce to resolve 62% of cases, handled 91,400 chats and saved 2,600+ hours. Autonomous resolution rose 70%. Vendor-published, from July 2025.",
   "agent": "Agentforce",
   "topic": "Customer Support",
   "source": "Case study (Salesforce)",
   "url": "https://www.salesforce.com/customer-stories/nexo/",
   "prompt": "Answer questions about accounts, withdrawals and fees around the clock from our help articles, and pass anything involving a dispute to a person.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/nexo-crypto-support-agentforce-91400-chats.html"
  },
  {
   "id": "suffolk-county-cora-voice-agent-consumer-affairs",
   "title": "A county's voice agent resolves up to half of consumer-affairs calls",
   "summary": "Google says Suffolk County, NY's CORA voice agent has an autonomous resolution rate of 45-52% in Consumer Affairs, handled 874 calls in its first two weeks at the Comptroller's office, and cut incoming call volume there by 30%. It replaced 25+ phone-menu branches. Vendor-published.",
   "agent": "CORA (Google Public Sector)",
   "topic": "Customer Support",
   "source": "Case study (Google Cloud)",
   "url": "https://cloud.google.com/customers/suffolk-county-new-york",
   "prompt": "Answer calls to the county's consumer-affairs line in any language, resolve what you can, and hand the rest to staff with a summary.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/suffolk-county-cora-voice-agent-consumer-affairs.html"
  },
  {
   "id": "banner-health-voice-agents-patient-scheduling",
   "title": "24/7 phone agents schedule primary-care visits",
   "summary": "Banner Health selected ElevenLabs voice agents to answer patient calls around the clock with no hold queue, schedule, reschedule and cancel appointments directly in its electronic medical record, and warm-transfer to staff with call context when a human is needed.",
   "agent": "ElevenAgents (ElevenLabs)",
   "topic": "Travel & Booking",
   "source": "Vendor blog (ElevenLabs)",
   "url": "https://elevenlabs.io/blog/banner-health",
   "prompt": "Answer calls from patients, book, move or cancel appointments in the scheduling system, and transfer to a staff member with a summary when the caller needs a person.",
   "added": "2026-10-29",
   "interest": 50,
   "page": "https://agentsightings.com/case/banner-health-voice-agents-patient-scheduling.html"
  },
  {
   "id": "source-recruiting-candidates-on-linkedin-and-build-valuation-models",
   "title": "Source recruiting candidates on LinkedIn and build valuation models",
   "summary": "Stopwatch demo: ~17 minutes covering (a) LinkedIn → spreadsheet of Bangalore YouTube-manager prospects and (b) a separate Excel valuation sheet built from the firm’s financials—both driven by ChatGPT Agent.",
   "agent": "ChatGPT Agent",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://www.leadwithai.co/article/chatgpt-agent-50-use-cases-for-business-leaders",
   "prompt": "Creative task: source recruiting candidates on LinkedIn and build valuation models. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/source-recruiting-candidates-on-linkedin-and-build-valuation-models.html"
  },
  {
   "id": "search-rentals-and-negotiate",
   "title": "Search rentals and negotiate",
   "summary": "Documented elsewhere: openClaw shows up in a write-up about search rentals and negotiate, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://openclawcases.com/cases/openclaw-rental",
   "prompt": "Help me search rentals and negotiate. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/search-rentals-and-negotiate.html"
  },
  {
   "id": "run-scheduled-ai-deep-research-on-stock-watchlists",
   "title": "Run scheduled AI deep research on stock watchlists",
   "summary": "One practitioner notes that in a comment, the person said they wanted daily and weekly cron jobs to trigger a script that runs AI deep research across their watchlists, plus a few tools to pull quick baseline information on any stock. They were buying “skills” from a Substack and planned to adapt them for local AI on a MacBook, but described the setup as still a work in progress rather than a finished system. They did not report results, performance numbers, or a completed workflow.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ValueInvesting/comments/1vi4mcl/anyone_started_using_ai_agents_for_fundamental",
   "prompt": "Research brief: run scheduled AI deep research on stock watchlists. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/run-scheduled-ai-deep-research-on-stock-watchlists.html"
  },
  {
   "id": "run-bots-for-property-sheets-and-client-email-replies",
   "title": "Run bots for property sheets and client email replies",
   "summary": "Linked example of run bots for property sheets and client email replies. One commenter said they had several Grok Bot agents running at once: a real-estate bot that updates a Google Drive sheet with properties matching their requirements, an email bot trained on their tone to reply to clients about projects, and another bot building a day-job \"Command Center\" to keep them organized.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1vzg2wd/any_one_tried_grok_bot_is_it_garbage_or_just_me",
   "prompt": "Go through my mail for this goal: run bots for property sheets and client email replies. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/run-bots-for-property-sheets-and-client-email-replies.html"
  },
  {
   "id": "run-a-one-person-studio-x27-s-office-work",
   "title": "Run a one-person studio's office work",
   "summary": "As described in the source, grok Bot shows up in a write-up about run a one-person studio's office work. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/ardakaraduman/status/2109139740234743888",
   "prompt": "Research brief: run a one-person studio's office work. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/run-a-one-person-studio-x27-s-office-work.html"
  },
  {
   "id": "record-meetings-and-suggest-answers-live-from-uploaded-files",
   "title": "Record meetings and suggest answers live from uploaded files",
   "summary": "Observed in the wild: someone adds PDF, text or markdown files to Parrot's knowledge base. Parrot records system audio and the mic with no bot joining, and transcribes with Whisper locally. The site quotes about $0.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49910328",
   "prompt": "Help me record meetings and suggest answers live from uploaded files. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/record-meetings-and-suggest-answers-live-from-uploaded-files.html"
  },
  {
   "id": "pre-stage-household-grocery-carts-from-instacart-history",
   "title": "Pre-stage household grocery carts from Instacart history",
   "summary": "As described in the source, a nightly script imports each day's Instacart orders into a local SQLite database powered by Instacart's own item IDs, via a community CLI. A SQL analyzer takes the median days between purchases for each item and marks items due, soon, or not due. Moving from an LLM clicking through the browser to canonical IDs took accuracy from eleven of fourteen items wrong on a typical run to 100%.",
   "agent": "OpenClaw",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://standardcompute.com/blog/that-40-heads-of-garlic-openclaw-post-is-funny-until-you-realize-what-actually-b",
   "prompt": "Shopping task: pre-stage household grocery carts from Instacart history. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/pre-stage-household-grocery-carts-from-instacart-history.html"
  },
  {
   "id": "negotiate-a-tv-price-with-a-retailer-x27-s-chat",
   "title": "Negotiate a TV price with a retailer's chat",
   "summary": "Observed in the wild: chatGPT Agent shows up in a write-up about negotiate a TV price with a retailer's chat, with figures like 939. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1mjtrni/i_saved_200_on_a_new_oled_tv_using_chatgpt_agent",
   "prompt": "Help me negotiate a TV price with a retailer's chat. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/negotiate-a-tv-price-with-a-retailer-x27-s-chat.html"
  },
  {
   "id": "map-and-control-a-whole-smart-home",
   "title": "Map and control a whole smart home",
   "summary": "Community report: karpathy prompted the agent in chat with \"Can you find my Sonos?\". The agent scanned the home IP network, logged into the unprotected Sonos, web-searched and reverse-engineered its API endpoints, then repeated the process on lights, HVAC, shades, pool, spa, and security systems. The agent produced a unified control center for the full house and now executes high-level commands across all devices.",
   "agent": "Custom agent",
   "topic": "Home & Family",
   "source": "X",
   "url": "https://x.com/MilkRoadAI/status/2036820969067524357",
   "prompt": "Home/family task: map and control a whole smart home. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/map-and-control-a-whole-smart-home.html"
  },
  {
   "id": "grade-papers-and-build-lessons",
   "title": "Grade papers and build lessons",
   "summary": "Classroom workload that used to eat evenings—lesson outlines, paper stacks, slide decks, grade analytics—now gets a big assist from Claude Cowork, according to one teacher’s write-up.",
   "agent": "Claude Cowork",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1uox9uu/thank_you_anthropic_as_a_teacher_claude_cowork",
   "prompt": "Research brief: grade papers and build lessons. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/grade-papers-and-build-lessons.html"
  },
  {
   "id": "find-cheapest-nonstop-flights-on-google-flights",
   "title": "Find cheapest nonstop flights on Google Flights",
   "summary": "Real-world thread: chatGPT Agent shows up in a write-up about find cheapest nonstop flights on Google Flights, with figures like 4, 10, 737. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m8k3am",
   "prompt": "Help me find cheapest nonstop flights on Google Flights. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/find-cheapest-nonstop-flights-on-google-flights.html"
  },
  {
   "id": "fill-out-job-applications-in-a-browser",
   "title": "Fill out job applications in a browser",
   "summary": "A published walkthrough shows claude Cowork desktop app gives Claude its own Chrome browser plus three user docs (resume, cover letter, Q&A on common questions). User pastes one prompt listing job boards, salary/location rules, skip conditions, and Sheets logging. It applies to 50 jobs inside 30 minutes while Someone does other work; source shows no further results or failure rates.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/gudanglifehack/status/2035979587352691167",
   "prompt": "Help me fill out job applications in a browser. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/fill-out-job-applications-in-a-browser.html"
  },
  {
   "id": "email-daily-deal-briefs-from-zillow-listings-to-real-estate-agents",
   "title": "Email daily deal briefs from Zillow listings to real estate agents",
   "summary": "From the linked report, claude shows up in a write-up about email daily deal briefs from Zillow listings to real estate agents, with figures like 47. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/0xAI42exe/status/2067231788632584658",
   "prompt": "Go through my mail for this goal: email daily deal briefs from Zillow listings to real estate agents. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/email-daily-deal-briefs-from-zillow-listings-to-real-estate-agents.html"
  },
  {
   "id": "diagnose-a-low-energy-health-issue",
   "title": "Diagnose a low-energy health issue",
   "summary": "A published walkthrough shows as described in the source: noah stood up an agent that recommended which blood tests to take and which medical professionals to see, and it reveals that it was the answer four doctors failed to provide.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/nmtalkss/status/2106130130993111220",
   "prompt": "Help me diagnose a low-energy health issue. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/diagnose-a-low-energy-health-issue.html"
  },
  {
   "id": "build-a-pokemon-style-dog-collecting-app",
   "title": "Build a Pokemon-style dog collecting app",
   "summary": "As described in the source, someone instructed Replit Agent to build the full app. The agent scraped American Kennel Club regional population data to set dog rarity values for catches. App was completed as a dog-themed catch-em-all experience.",
   "agent": "Replit Agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/DanielPrevoznik/status/2088991596771459241",
   "prompt": "Help me build a Pokemon-style dog collecting app. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 50,
   "page": "https://agentsightings.com/case/build-a-pokemon-style-dog-collecting-app.html"
  },
  {
   "id": "lynchburg-dell-agent-47000-student-messages",
   "title": "A university's agent answers students at 9pm: 47,000 messages",
   "summary": "The University of Lynchburg launched Dell, a no-code Druid AI agent, after approving it in April and going live in October. Druid says it has handled 47,000+ interactions among about 2,800 students and passed the old chatbot's total usage within months. Vendor-published.",
   "agent": "Dell (Druid AI agent)",
   "topic": "Productivity",
   "source": "Blog (Druid AI)",
   "url": "https://www.druidai.com/blog/47000-student-conversations-how-the-university-of-lynchburg-scaled-support-with-ai-agent",
   "prompt": "Answer students' financial aid, registrar and IT questions at any hour from the university's pages, and point them to the right office when you do not know.",
   "added": "2026-10-29",
   "interest": 49,
   "page": "https://agentsightings.com/case/lynchburg-dell-agent-47000-student-messages.html"
  },
  {
   "id": "pldt-home-store-genie-resolution-time",
   "title": "A telecom's store agent cuts resolution time by up to 98%",
   "summary": "Amdocs says its Store Genie agent, built on Bedrock AgentCore, helps PLDT Home in the Philippines cut customer resolution time by up to 98% across its sales and service centres. It was already running at sister company Smart. Vendor press release.",
   "agent": "Store Genie (Amdocs aOS, Bedrock AgentCore)",
   "topic": "Customer Support",
   "source": "Press release (AWS / Amdocs)",
   "url": "https://press.aboutamazon.com/aws/2026/6/amdocs-expands-store-genie-to-pldt-home-bringing-agentic-ai-to-broadband-customer-service-operations",
   "prompt": "A customer walks into the store with a broadband problem. Look up their account and orders, find the cause, and give the staff member the next step.",
   "added": "2026-10-29",
   "interest": 49,
   "page": "https://agentsightings.com/case/pldt-home-store-genie-resolution-time.html"
  },
  {
   "id": "socar-turkiye-copilot-agents-7500-hours",
   "title": "A fuel company's agents do IBAN lookups, PO creation and onboarding",
   "summary": "Microsoft says SOCAR Türkiye's Copilot Studio agents cover finance, legal, HR, procurement and IT: payment lookups, approvals, SAP purchase-order creation, compliance guidance, tickets and onboarding. It reports 7,500+ hours saved a year. Vendor-published.",
   "agent": "Copilot Studio agents",
   "topic": "Business & Startups",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/26754-socar-turkiye-microsoft-copilot-studio",
   "prompt": "Handle routine back-office requests: look up a supplier's payment details, raise a purchase order from an approved request, and log a ticket for anything else.",
   "added": "2026-10-29",
   "interest": 49,
   "page": "https://agentsightings.com/case/socar-turkiye-copilot-agents-7500-hours.html"
  },
  {
   "id": "druva-druai-resolves-cyber-recovery-issues",
   "title": "A backup firm's agent resolves up to 63% of issues on its own",
   "summary": "AWS says Druva's DruAI agent, in use for 3,000+ customers, solves up to 63% of issues without human intervention and speeds up resolution by up to 58%. Vendor-reported.",
   "agent": "DruAI (Druva)",
   "topic": "DevOps & Infra",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-agentcore-and-claude-transforming-business-with-agentic-ai/",
   "prompt": "A customer reports a backup problem. Diagnose it from their logs and settings, fix it if you are allowed to, and otherwise hand the case to a person with what you found.",
   "added": "2026-10-29",
   "interest": 49,
   "page": "https://agentsightings.com/case/druva-druai-resolves-cyber-recovery-issues.html"
  },
  {
   "id": "turn-gmail-action-items-into-an-ai-to-do-list",
   "title": "Turn Gmail action items into an AI to-do list",
   "summary": "From the linked report, the poster assembled a Chrome extension that reads Gmail and uses AI to turn emails into a clean, prioritized to-do list, aimed at buried action items in threads, invites, and follow-ups. The setup is deliberately low-friction: install it and it works, with no signup, no manual copying, and no way for the maker to contact users because there are no logins. After 18 days, the extension had 47 organic users and later the maker mentioned a first sale of $2.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/productivity/comments/1rvcc9v/i_built_a_chrome_extension_that_turns_my_messy",
   "prompt": "Go through my mail for this goal: turn Gmail action items into an AI to-do list. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/turn-gmail-action-items-into-an-ai-to-do-list.html"
  },
  {
   "id": "summarize-the-day-x27-s-new-emails-and-manage-meetings",
   "title": "Summarize the day's new emails and manage meetings",
   "summary": "From the linked report, requested an AI assistant to summarize emails from the previous day and received a response within 30 seconds. The system also manages drafts, calendars, action items, and meeting summaries.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/artificial/comments/1tygu1j/what_is_the_most_useful_thing_youre_using_ai_for",
   "prompt": "Assistant mode: summarize the day's new emails and manage meetings. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/summarize-the-day-x27-s-new-emails-and-manage-meetings.html"
  },
  {
   "id": "screen-candidates-and-summarize-resumes",
   "title": "Screen candidates and summarize resumes",
   "summary": "Observed in the wild: the use of AI by commenters is exemplified by its ability to summarize resumes, screen candidates in large quantities, avoid boolean questions, and provide an agent that can screen candidates at any time.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/recruiting/comments/1r44jm7/how_much_of_a_role_does_ai_really_play_in_the",
   "prompt": "Help me screen candidates and summarize resumes. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/screen-candidates-and-summarize-resumes.html"
  },
  {
   "id": "run-bookkeeping-and-receipt-matching-in-quickbooks",
   "title": "Run bookkeeping and receipt matching in QuickBooks",
   "summary": "As described in the source, someone connected an AI agent to QuickBooks, SharePoint, and email. It attaches receipts sent by email to the matching line item in QB, forwards invoices into the checks-to-print queue, adds new vendors, and sends W-9 requests. Bookkeeping and ongoing accountant work were replaced; Someone states the bot performed better than the prior accountant at these tasks.",
   "agent": "Custom agent",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1uh3ka8/automate_monthend_receipt_reconciliation_skill",
   "prompt": "Finance task: run bookkeeping and receipt matching in QuickBooks. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/run-bookkeeping-and-receipt-matching-in-quickbooks.html"
  },
  {
   "id": "run-a-company-autonomously",
   "title": "Run a company autonomously",
   "summary": "Observed in the wild: andon Labs assembled Pion to run its real businesses: vending machines, a retail store in SF, a cafe in Stockholm and radio stations. Persistent agents get email, phone, banking, browser and secure computing environments. Early agents made costly mistakes, and by late 2025 frontier models ran a vending machine profitably.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49700477",
   "prompt": "Go through my mail for this goal: run a company autonomously. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/run-a-company-autonomously.html"
  },
  {
   "id": "run-codex-tasks-on-your-home-assistant-config",
   "title": "Run Codex tasks on your Home Assistant config",
   "summary": "Community report: A Home Assistant add-on runs Codex CLI with read/write access to /config, and a custom integration (installable through HACS) exposes actions and entities. Tasks start from scripts, automations, or Assist and run non-interactively. The builder shows an Assist request that had Codex update a dashboard card, followed by a completion notification.",
   "agent": "Codex",
   "topic": "Personal Assistant",
   "source": "GitHub",
   "url": "https://github.com/itsreverence/ha-codex-assist",
   "prompt": "Assistant mode: run Codex tasks on your Home Assistant config. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/run-codex-tasks-on-your-home-assistant-config.html"
  },
  {
   "id": "organize-a-friday-night",
   "title": "Organize a Friday night",
   "summary": "From the linked report, custom agent shows up in a write-up about organize a Friday night. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/ethancole_ai/status/2099951040678465585",
   "prompt": "Assistant mode: organize a Friday night. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/organize-a-friday-night.html"
  },
  {
   "id": "morning-test-coverage-agent-on-merged-code",
   "title": "Morning test-coverage agent on merged code",
   "summary": "Every morning a Cursor automation reviews recently merged code, adds tests that match existing conventions, runs them, and opens a PR.",
   "agent": "Cursor",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "Review code merged in the last 24 hours. Identify missing test coverage, add tests following existing conventions without changing production behavior unless required, run the relevant targets, and open a PR.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/morning-test-coverage-agent-on-merged-code.html"
  },
  {
   "id": "manage-projects-by-chat-through-openclaw",
   "title": "Manage projects by chat through OpenClaw",
   "summary": "A writer uses OpenClaw on a dedicated machine to create projects, plan, drive Claude Code, test and deploy entirely over messaging apps from a phone.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://reorx.com/blog/openclaw-is-changing-my-life/",
   "prompt": "Create a new project for <idea>, draft a plan for my review, then after I approve, implement it by directing the coding agent. Message me progress and ask before deploys.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/manage-projects-by-chat-through-openclaw.html"
  },
  {
   "id": "have-claude-implement-most-of-a-day-job-x27-s-work",
   "title": "Have Claude implement most of a day job's work",
   "summary": "A published walkthrough shows provides Claude with work responsibilities and ensures that the majority of the functionality is executed, but the final 10% is still the most demanding.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1shngqm/i_automated_most_of_my_job",
   "prompt": "Help me have Claude implement most of a day job's work. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/have-claude-implement-most-of-a-day-job-x27-s-work.html"
  },
  {
   "id": "get-a-second-opinion-on-an-mri",
   "title": "Get a second opinion on an MRI",
   "summary": "Documented elsewhere: the builder gave Claude Code (Opus 4.8, xhigh) a roughly 266 MB DICOM export of a shoulder MRI, let it install packages, and told it only the symptom (2–3 weeks of right shoulder pain). It planned and read the scans for about an hour. Opus located an intact tendon, while the doctor described a Grade III partial tear.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48708941",
   "prompt": "Help me get a second opinion on an MRI. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/get-a-second-opinion-on-an-mri.html"
  },
  {
   "id": "flag-risks-in-a-50-page-contract-before-signing",
   "title": "Flag risks in a 50-page contract before signing",
   "summary": "Case study takeaway: the poster uploaded a 50-page contract PDF to Claude and used a prompt asking it to read the whole document, explain in plain English what they were agreeing to, list obligations and risks, say what a lawyer would flag as unusual or negotiable, and then name the three points to push back on hardest. Claude surfaced buried clauses, and the poster says a person should still bring a lawyer in for anything serious. It flagged three issues the poster says they would have missed, covering an auto-renewal near the end and a liability term they say was not standard.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/PromptEngineering/comments/1u4f58a/i_uploaded_a_50_page_contract_and_asked_claude",
   "prompt": "Help me flag risks in a 50-page contract before signing. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/flag-risks-in-a-50-page-contract-before-signing.html"
  },
  {
   "id": "file-receipts-from-a-folder",
   "title": "File receipts from a folder",
   "summary": "Case study takeaway: user ran Grok Bot against a receipts folder. The agent filed 14 receipts on its own, held 2 because they needed a card number, and saved a payment method it located without Someone writing it. Returned 14 filed receipts, 2 held for card details, and one saved payment method.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Argona0x/status/2090181890565181801",
   "prompt": "Help me file receipts from a folder. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/file-receipts-from-a-folder.html"
  },
  {
   "id": "create-bills-in-quickbooks-from-supplier-invoices",
   "title": "Create bills in QuickBooks from supplier invoices",
   "summary": "Linked example of create bills in QuickBooks from supplier invoices. A commenter described a hands-free setup where supplier invoices are sent to a dedicated inbox, then an AI agent parses each invoice and pushes the details straight into accounting software like QuickBooks, covering taxes and net terms.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49617903",
   "prompt": "Go through my mail for this goal: create bills in QuickBooks from supplier invoices. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/create-bills-in-quickbooks-from-supplier-invoices.html"
  },
  {
   "id": "buy-a-sell-out-hyrox-event-ticket",
   "title": "Buy a sell-out Hyrox event ticket",
   "summary": "From the linked report, someone directed their Grok Bot to book a Hyrox ticket. The agent handled the purchase on its own and secured the spot. The ticket was obtained before sell-out.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/smartmoneykai/status/2108615410321158638",
   "prompt": "Help me buy a sell-out Hyrox event ticket. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/buy-a-sell-out-hyrox-event-ticket.html"
  },
  {
   "id": "build-a-spending-insights-app-from-bank-transactions",
   "title": "Build a spending-insights app from bank transactions",
   "summary": "Observed in the wild: you set up a Replit account and a Plaid dashboard account, then copy your client ID and secret into Replit, which stores them in a vault. Next you add Plaid as a connector and link your bank, or use sandbox data first. The post reports no results from a finished app.",
   "agent": "Replit Agent",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://plaid.com/blog/build-personalized-finance-apps-replit-plaid/",
   "prompt": "Finance task: build a spending-insights app from bank transactions. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/build-a-spending-insights-app-from-bank-transactions.html"
  },
  {
   "id": "build-a-newsletter-performance-dashboard-without-coding",
   "title": "Build a newsletter performance dashboard without coding",
   "summary": "Community report: the builder, who says they don't write code, described in plain English a dashboard for tracking newsletter performance across their whole network. Claude Code, run from the terminal, assembled a working tool from that text description. The builder says the dashboard existed about 30 minutes after they described it, replacing work they Leaned on to pay contractors $2,000-5,000 for.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://futureinsidernews.substack.com/p/claude-code-changed-everything-heres",
   "prompt": "Go through my mail for this goal: build a newsletter performance dashboard without coding. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/build-a-newsletter-performance-dashboard-without-coding.html"
  },
  {
   "id": "automate-freelance-admin",
   "title": "Automate freelance admin",
   "summary": "From the linked report, openClaw shows up in a write-up about automate freelance admin, with figures like 9, 20. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://dev.to/farooqshabbir/i-built-a-247-freelance-operations-agent-with-openclaw-it-runs-my-business-while-i-sleep-4lde",
   "prompt": "Assistant mode: automate freelance admin. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/automate-freelance-admin.html"
  },
  {
   "id": "automate-event-social-posting-across-platforms",
   "title": "Automate event social posting across platforms",
   "summary": "A published walkthrough shows the poster used Claude Code to connect their social channels for promoting the AI x Marketing Summit in San Francisco on May 28–29 and drive registrations. From the thread, the setup let Claude handle cross-platform posting, and the person judged the results by platform rather than describing a fully hands-off workflow; comments mostly discuss alternative stacks rather than extra implementation detail. It worked well on Twitter and Reddit, which the poster called super smooth, and only somewhat on LinkedIn.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ai_x_marketing/comments/1tedcuf/trying_to_automate_social_posting_for_an_event",
   "prompt": "Help me automate event social posting across platforms. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/automate-event-social-posting-across-platforms.html"
  },
  {
   "id": "auto-fill-job-applications",
   "title": "Auto-fill job applications",
   "summary": "Documented elsewhere: as described in the source: the process of editing a CV and cover letter for job applications is simplified by an agent who reviews the user's work.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48753323",
   "prompt": "Help me auto-fill job applications. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/auto-fill-job-applications.html"
  },
  {
   "id": "apply-to-jobs-automatically-and-schedule-calls",
   "title": "Apply to jobs automatically and schedule calls",
   "summary": "Documented elsewhere: muse was given instructions to handle job applications end to end. It updated resumes and cover letters then submitted automatic applications to over 60 positions on its own. Three companies responded and one advanced to interview rounds.",
   "agent": "Meta Muse",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/blackhousecap/status/2108616839160836606",
   "prompt": "Help me apply to jobs automatically and schedule calls. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/apply-to-jobs-automatically-and-schedule-calls.html"
  },
  {
   "id": "add-todo-items-to-todoist-from-telegram",
   "title": "Add todo items to Todoist from Telegram",
   "summary": "From the linked report, the commenter uses an orchestration platform, either n8n or windmill.dev, connected to a Telegram bot. An LLM covers the natural language side, so a message sent to the bot gets turned into a todo item and added to their Todoist account. The commenter says they do this themselves and that it covers the use case of adding home todos from elsewhere.",
   "agent": "n8n",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47728484",
   "prompt": "Assistant mode: add todo items to Todoist from Telegram. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 49,
   "page": "https://agentsightings.com/case/add-todo-items-to-todoist-from-telegram.html"
  },
  {
   "id": "vestas-jira-agent-routes-support-tickets",
   "title": "Route wind-turbine software tickets to the right team",
   "summary": "Vestas says a Jira Ticket Agent built on Dataiku and Snowflake reads new tickets and posts a recommended component or team as a comment, with engineers reviewing before action. The target is under one hour for assignment, against a 72-hour SLA. Those are targets and projections, not measured results.",
   "agent": "Jira Ticket Agent (Dataiku, Snowflake)",
   "topic": "Software Development",
   "source": "News (Hanging Context)",
   "url": "https://hangingcontext.com/stories/2026-10-28/enterprise-agents-measured-work/",
   "prompt": "Read each new support ticket and suggest which component or team should own it. Post the suggestion as a comment with your reasoning, but do not change any fields. A person decides.",
   "added": "2026-10-29",
   "interest": 48,
   "page": "https://agentsightings.com/case/vestas-jira-agent-routes-support-tickets.html"
  },
  {
   "id": "bluesight-prism-agents-hospital-compliance",
   "title": "Hospital pharmacy software gets an agent that reasons across six products",
   "summary": "Bluesight built Prism, an AgentCore assistant spanning hospital and pharmacy compliance products. AWS says it went from a three-day sprint to general availability in under nine months, and its first assistant is used by 20 health systems. A multi-product version is due later in 2026. Vendor-published.",
   "agent": "Prism Assistant (Amazon Bedrock AgentCore)",
   "topic": "Business & Startups",
   "source": "Blog (AWS)",
   "url": "https://aws.amazon.com/blogs/machine-learning/building-an-agentic-ai-solution-at-bluesight-with-amazon-bedrock/",
   "prompt": "Pull data from several compliance systems, find the unusual items, and explain them in plain English with links back to the source records.",
   "added": "2026-10-29",
   "interest": 48,
   "page": "https://agentsightings.com/case/bluesight-prism-agents-hospital-compliance.html"
  },
  {
   "id": "regal-rexnord-rrx-gpt-answers-employee-questions",
   "title": "An industrial firm's policy bot answers 2,000 questions a month",
   "summary": "Microsoft says Regal Rexnord's RRX GPT, a Copilot Studio agent in Teams, answers employee questions on policy and internal knowledge from SharePoint, Microsoft 365 and ServiceNow. It handles 2,000+ inquiries a month and saves an estimated 2,400 hours a year. Vendor-published.",
   "agent": "RRX GPT (Copilot Studio)",
   "topic": "Productivity",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/26197-regal-rexnord-microsoft-copilot-studio",
   "prompt": "Answer employees' questions about company policy using our SharePoint and ServiceNow articles. Cite the page you used and say so when you do not know.",
   "added": "2026-10-29",
   "interest": 48,
   "page": "https://agentsightings.com/case/regal-rexnord-rrx-gpt-answers-employee-questions.html"
  },
  {
   "id": "speed-up-support-ticket-research-and-triage",
   "title": "Speed up support ticket research and triage",
   "summary": "According to the write-up, the source is a listing page with only short blurbs, so it has little on mechanics. Corey Smith, a Senior Technical Support Engineer at ClickUp, assembled a system for the research that comes before a rep replies to each ticket. ClickUp's research time per ticket fell from 15 minutes to 4, about 917+ hours a month across 5,000+ tickets.",
   "agent": "Zapier Agents",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://zapier.com/customer-stories",
   "prompt": "Research brief: speed up support ticket research and triage. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/speed-up-support-ticket-research-and-triage.html"
  },
  {
   "id": "search-venue-sites-for-concerts-to-attend",
   "title": "Search venue sites for concerts to attend",
   "summary": "Community report: the builder asked Operator, which runs ChatGPT in its own virtual machine with a browser, to check Berlin venue websites for classical concerts. Operator had to set dates and see which concerts were sold out. The first answer came back after 15 minutes, but names and dates alone didn't let The builder find the concerts.",
   "agent": "Operator",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://antoine.fi/trying-openai-operator",
   "prompt": "Go through my mail for this goal: search venue sites for concerts to attend. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/search-venue-sites-for-concerts-to-attend.html"
  },
  {
   "id": "reply-to-outreach-responses-from-email-and-linkedin-inboxes",
   "title": "Reply to outreach responses from email and LinkedIn inboxes",
   "summary": "As described in the source, one commenter said they switched from Claude Code to Claude Cowork and connected it to their Gmail or Apollo-sent mailbox through MCP, plus enrichment tools covering Apollo, Clay, Zero Bounce, LinkedIn, and LinkedIn Jobpost. Claude Cowork scanned email every 10 minutes and LinkedIn inboxes through Phantombuster every 60 minutes, then replied to outreach responses or expressions of interest automatically. The thread gives no hard numbers for this specific setup, only that the commenter uses it to answer interested replies from email and LinkedIn.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/coldemail/comments/1vbwh4o/anyone_actually_using_claude_code_agents_in_their",
   "prompt": "Go through my mail for this goal: reply to outreach responses from email and LinkedIn inboxes. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/reply-to-outreach-responses-from-email-and-linkedin-inboxes.html"
  },
  {
   "id": "rebuild-a-platform-and-add-rag-features-in-one-day",
   "title": "Rebuild a platform and add RAG features in one day",
   "summary": "Free Replit Agent day → full-stack rebuild + seven live RAG features + a flashy demo reel. Rough scoreboard: 46 commits, 14,000+ LOC, ~12 hours, three shippable pieces.",
   "agent": "Replit Agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/LakshOnline/status/2050977394861990332",
   "prompt": "Creative task: rebuild a platform and add RAG features in one day. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/rebuild-a-platform-and-add-rag-features-in-one-day.html"
  },
  {
   "id": "orchestrate-video-generation-from-the-command-line",
   "title": "Orchestrate video generation from the command line",
   "summary": "Case study takeaway: someone ran their custom agent dimul from the command line. It connected to Grok via CLI under a SuperGrok subscription and Leaned on it to orchestrate Imagine clip generation from prompts. A 215-second video resulted.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/NURM_Dima/status/2107878966510158087",
   "prompt": "Creative task: orchestrate video generation from the command line. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/orchestrate-video-generation-from-the-command-line.html"
  },
  {
   "id": "generate-ads-from-a-website-url",
   "title": "Generate ads from a website URL",
   "summary": "According to the write-up, user gave the Arcads agent one website URL plus one sentence. The agent read the site to learn the brand, pulled existing assets, generated ad ideas on its own, then produced static ads and AI UGC videos without further input. Output quality was surprisingly good.",
   "agent": "Custom agent",
   "topic": "Design & Creative",
   "source": "X",
   "url": "https://x.com/Just_sharon7/status/2084628330943582342",
   "prompt": "Design task: generate ads from a website URL. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/generate-ads-from-a-website-url.html"
  },
  {
   "id": "fill-a-supermarket-cart-from-regular-purchases-and-sales",
   "title": "Fill a supermarket cart from regular purchases and sales",
   "summary": "Documented elsewhere: the developer wrote silpo-agent-cli, a terminal CLI and Claude Code skill that talks to Silpo's official MCP server. You give Claude a plain-language instruction, such as reordering usual groceries, swapping in a sale brand and staying under 1500 UAH. The source gives no measured results.",
   "agent": "Claude Code",
   "topic": "Shopping",
   "source": "Blog",
   "url": "https://dev.to/mit9/real-world-agentic-e-commerce-rebuilding-my-grocery-cart-with-claude-and-mcp-3b82",
   "prompt": "Shopping task: fill a supermarket cart from regular purchases and sales. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/fill-a-supermarket-cart-from-regular-purchases-and-sales.html"
  },
  {
   "id": "do-taxes-by-selecting-the-right-invoices",
   "title": "Do taxes by selecting the right invoices",
   "summary": "Linked example of do taxes by selecting the right invoices. The poster used ChatGPT’s Agent tool to help with quarterly Xero reconciliation for a business, not fully file taxes end to end.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m2v4t6/gpt_agent_is_doing_my_taxes",
   "prompt": "Help me do taxes by selecting the right invoices. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/do-taxes-by-selecting-the-right-invoices.html"
  },
  {
   "id": "clear-local-disk-space-with-a-guided-plan",
   "title": "Clear local disk space with a guided plan",
   "summary": "Lenny asks Claude Code how to free storage, discusses options, then has it run the safer cleanup steps.",
   "agent": "Claude Code",
   "topic": "Personal Files",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "How can I clear storage on this computer? List the largest reclaimable items with sizes, risk, and the exact commands. Wait for my OK before deleting anything.",
   "added": "2026-10-15",
   "interest": 48,
   "page": "https://agentsightings.com/case/clear-local-disk-space-with-a-guided-plan.html"
  },
  {
   "id": "sentara-chart-agent-for-discharge-planning",
   "title": "Ask a patient's chart in plain English for discharge planning",
   "summary": "Sentara Health, a 12-hospital system, deployed a Regard-built AI tool so case managers can query a chart in plain language and see what changed in the last 24 hours, replacing hours of manual chart review. Sentara estimates it could save nearly 90,000 hours. That is an estimate, not a measured result.",
   "agent": "Regard AI case management",
   "topic": "Research & Data",
   "source": "News (Regard)",
   "url": "https://regard.com/press-media/sentara-health-extends-ai-clinical-reasoning-to-case-managers-providing-a-complete-clinical-picture-for-accelerated-care-transitions",
   "prompt": "Summarise what changed in this patient's chart in the last 24 hours and list history, screenings and status changes a case manager needs for discharge planning. Cite where each item came from.",
   "added": "2026-10-29",
   "interest": 47,
   "page": "https://agentsightings.com/case/sentara-chart-agent-for-discharge-planning.html"
  },
  {
   "id": "warehouse-exchange-agents-match-unused-warehouse-space",
   "title": "Agents match renters to unused warehouse space, answering in 40 seconds",
   "summary": "Google's customer page says Warehouse Exchange made more than $25M in revenue from unused logistics space through teams of Gemini agents. The company's own post says its renter-facing agent, Robin, answers inquiries in about 40 seconds, and its site describes the $25M as transactions. Company-reported.",
   "agent": "Robin and other Gemini agents",
   "topic": "Sales & Marketing",
   "source": "Case study (Google Cloud)",
   "url": "https://cloud.google.com/customers",
   "prompt": "A renter asks about warehouse space. Answer within a minute, match them to available space, and hand pricing questions to a person.",
   "added": "2026-10-29",
   "interest": 47,
   "page": "https://agentsightings.com/case/warehouse-exchange-agents-match-unused-warehouse-space.html"
  },
  {
   "id": "banco-hipotecario-dynabot-case-creation-80-percent",
   "title": "A bank's bot opens the support case for you, 80% faster",
   "summary": "Microsoft says Banco Hipotecario's DynaBot, built in Copilot Studio with Dynamics 365, captures a customer's issue and creates the incident in Dataverse, cutting average case-creation time by 80% against manual handling. Vendor-published.",
   "agent": "DynaBot (Copilot Studio)",
   "topic": "Customer Support",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/26969-banco-hipotecario-github-copilot",
   "prompt": "Ask the customer what went wrong, fill in the case form from their answers, and create the incident. Ask again if something required is missing.",
   "added": "2026-10-29",
   "interest": 47,
   "page": "https://agentsightings.com/case/banco-hipotecario-dynabot-case-creation-80-percent.html"
  },
  {
   "id": "ogier-harvey-legal-research-and-drafting",
   "title": "A 500-lawyer offshore firm rolls out Harvey firmwide",
   "summary": "Law firm Ogier has deployed Harvey across its 13 offices and 500-strong legal team after a firmwide pilot, for document analysis, legal research, drafting and knowledge retrieval, under its existing governance framework.",
   "agent": "Harvey",
   "topic": "Research & Data",
   "source": "News (Ogier)",
   "url": "https://www.ogier.com/news-and-insights/news/ogier-expands-its-ai-enabled-legal-service-offering-with-the-adoption-of-harvey/",
   "prompt": "Read this contract bundle and summarise the key terms. Find the clauses that differ from our standard wording and draft a first mark-up for a lawyer to review.",
   "added": "2026-10-29",
   "interest": 47,
   "page": "https://agentsightings.com/case/ogier-harvey-legal-research-and-drafting.html"
  },
  {
   "id": "research-crypto-assets-and-execute-swaps-and-perps-trades",
   "title": "Research crypto assets and execute swaps and perps trades",
   "summary": "Real-world thread: the agent received full access to Someone's portfolio data covering balances, allocations, open positions, entry prices, liquidation levels and running PnL. It performed deep asset research covering price action, volume, funding, fundamentals and catalysts, conducted portfolio analysis, discovered tokens by momentum or other filters, prepared swaps with cross-chain routing and bridging, and executed perp positions covering opening, closing and limit orders on Hyperliquid. No specific results or performance numbers are stated in the source.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/Kiojame_/status/2105253059777482798",
   "prompt": "Research brief: research crypto assets and execute swaps and perps trades. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/research-crypto-assets-and-execute-swaps-and-perps-trades.html"
  },
  {
   "id": "program-a-sub-2k-personal-robot",
   "title": "Program a sub-$2k personal robot",
   "summary": "A published walkthrough shows the source is a launch post for MARS, an assembled, calibrated robot with a Jetson Orin Nano, a 5DoF arm with wrist camera, an RGBD camera, 2D LiDAR and speakers. You write \"behaviors\" in the BASIC SDK, which runs the robot like an AI agent with VLAs. The post is a launch announcement and reports no results or measurements.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=45504127",
   "prompt": "Help me program a sub-$2k personal robot. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/program-a-sub-2k-personal-robot.html"
  },
  {
   "id": "prepare-and-file-overdue-german-vat-statements",
   "title": "Prepare and file overdue German VAT statements",
   "summary": "Community report: gave Claude Cowork a project folder holding invoice and receipt PDFs and asked it to prepare accounts for each quarter. Logged into Elster, then Leaned on the Claude Chrome extension to locate the forms, fill them with the prepared data, and submit the Q3 2025 VAT statements. The overdue Q3 2025 statements were filed.",
   "agent": "Claude Cowork",
   "topic": "Finance",
   "source": "X",
   "url": "https://x.com/EastlondonDev/status/2020826097890242582",
   "prompt": "Finance task: prepare and file overdue German VAT statements. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/prepare-and-file-overdue-german-vat-statements.html"
  },
  {
   "id": "pdf-schedules-turned-into-calendar-files",
   "title": "PDF schedules turned into calendar files",
   "summary": "A script has an LLM convert quarterly PDF schedules into ICS files, then cross-checks the ICS against the original PDF.",
   "agent": "Claude",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=44405067",
   "prompt": "Convert this PDF schedule into an .ics file. Then compare your file to the PDF and list any events that don't match.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/pdf-schedules-turned-into-calendar-files.html"
  },
  {
   "id": "log-meals-and-medical-metrics-into-a-health-spreadsheet",
   "title": "Log meals and medical metrics into a health spreadsheet",
   "summary": "Linked example of log meals and medical metrics into a health spreadsheet. The poster made a Claude Cowork “Health Related” folder on a desktop and asked Claude to create both a spreadsheet and a skill for tracking diet goals and medical metrics.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCowork/comments/1sz5yqb/healthdiet_tracking_skill",
   "prompt": "Help me log meals and medical metrics into a health spreadsheet. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/log-meals-and-medical-metrics-into-a-health-spreadsheet.html"
  },
  {
   "id": "let-ai-models-trade-stocks-autonomously-with-1k-each",
   "title": "Let AI models trade stocks autonomously with $1k each",
   "summary": "Linked example of let AI models trade stocks autonomously with $1k each. The poster set up four paper-trading portfolios with $1,000 each on Alpaca and ran the same weekday pre-market prompt against ChatGPT, Gemini, Claude, and Perplexity with Deep Research enabled.",
   "agent": "Gemini",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1rq33za/chatgpt_vs_gemini_vs_claude_vs_perplexity_i_gave",
   "prompt": "Help me let AI models trade stocks autonomously with $1k each. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/let-ai-models-trade-stocks-autonomously-with-1k-each.html"
  },
  {
   "id": "give-an-ai-its-own-inbox-to-act-as-personal-assistant",
   "title": "Give an AI its own inbox to act as personal assistant",
   "summary": "From the linked report, the poster connected two mail accounts to ChatGPT: a dedicated Outlook inbox for the AI and their personal Gmail separately. In chat, they could tell it things like emailing a reminder next Friday at 6 pm, and ChatGPT drafted the message from the Outlook account to Gmail and queued it with delayed delivery; the person just issued the request and checked the result. It worked in their test: a reminder requested for one hour later was booked in Outlook and turned the AI inbox into a simple reminder system.",
   "agent": "ChatGPT",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1wqcb99/i_gave_my_ai_its_own_email_inbox",
   "prompt": "Go through my mail for this goal: give an AI its own inbox to act as personal assistant. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/give-an-ai-its-own-inbox-to-act-as-personal-assistant.html"
  },
  {
   "id": "draft-daily-niche-news-posts-for-four-social-platforms",
   "title": "Draft daily niche-news posts for four social platforms",
   "summary": "A published walkthrough shows put together from one prompt in 15 minutes. The agent scans daily niche news, selects the most relevant story, composes tailored posts for LinkedIn, X, Instagram and Facebook, produces platform-cropped images, afterward emails the four sketches at 7am each morning. No performance numbers or results are stated.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/coreyganim/status/2037199397122175248",
   "prompt": "Creative task: draft daily niche-news posts for four social platforms. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/draft-daily-niche-news-posts-for-four-social-platforms.html"
  },
  {
   "id": "debug-low-level-cryptography-code",
   "title": "Debug low-level cryptography code",
   "summary": "A published walkthrough shows filippo Valsorda ran Claude Code (Opus 4.1, no system prompt) on his new Go ML-DSA implementation. His prompt named the test command and code directory, described the failing symptom, and asked it to think hard. It located all three bugs in one shot without help.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=45784179",
   "prompt": "Help me debug low-level cryptography code. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/debug-low-level-cryptography-code.html"
  },
  {
   "id": "compile-homework-turn-ins-into-one-report",
   "title": "Compile homework turn-ins into one report",
   "summary": "Community report: someone described a teacher who used no-code automations to stop manually checking Google Classroom, email, and paper submissions for homework. powered by Office 365 tools plus platforms named as Power Automate, BhindiAI, Zapier, and Make, the setup scanned those sources and produced one unified turn-in report; the human role described was choosing and wiring the no-code workflow. The stated result was that a task that used to take 45 minutes was reduced to seconds.",
   "agent": "Custom agent",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1nohjoy/teacher_using_ai_agents_for_school_work",
   "prompt": "Assistant mode: compile homework turn-ins into one report. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/compile-homework-turn-ins-into-one-report.html"
  },
  {
   "id": "capture-leads-on-whatsapp-and-save-them-to-sheets",
   "title": "Capture leads on WhatsApp and save them to Sheets",
   "summary": "From the linked report, the n8n workflow starts with a WhatsApp trigger that feeds incoming messages to an AI agent. The agent asks the lead for name and email during the chat, then writes the data directly to Google Sheets. The demo shows the lead captured and saved automatically to Sheets.",
   "agent": "n8n",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/Anshuman_7_7_7_/status/2109942223420973344",
   "prompt": "Go through my mail for this goal: capture leads on WhatsApp and save them to Sheets. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/capture-leads-on-whatsapp-and-save-them-to-sheets.html"
  },
  {
   "id": "bulk-unsubscribe-from-gmail-newsletters",
   "title": "Bulk-unsubscribe from Gmail newsletters",
   "summary": "From the linked report, in a Cowork project with the Gmail connector enabled, Ben Patterson asked Claude Sonnet to review his \"archivable\" label and list newsletters safe to unsubscribe from, without acting yet. Claude then quizzed him on each one in multiple-choice form with its own recommendation. Claude unsubscribed from 21 newsletters in about five minutes, one for each he chose to cut.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.pcworld.com/article/3185314/i-was-buried-in-email-newsletters-heres-how-claude-rescued-me.html",
   "prompt": "Go through my mail for this goal: bulk-unsubscribe from Gmail newsletters. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/bulk-unsubscribe-from-gmail-newsletters.html"
  },
  {
   "id": "build-a-thyroid-episode-warning-model-from-wearable-data",
   "title": "Build a thyroid episode warning model from wearable data",
   "summary": "One practitioner notes that the poster exported 9.5 years of Apple Watch data through Apple Health plus Whoop data, labeled periods that matched hyperthyroid phases, and gave Claude Code the dataset to build a detector. In comments, they said Claude tried many models for over an hour, landed on XGBoost, engineered 53 features before keeping three resting-heart-rate features, and the poster also held out the most recent episode to manually test the result. They said the backtest on the held-out episode would have flagged trouble in early August before labs confirmed it at the end of the month.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1qi8x0r/i_gave_claude_code_95_years_of_health_data_to",
   "prompt": "Help me build a thyroid episode warning model from wearable data. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/build-a-thyroid-episode-warning-model-from-wearable-data.html"
  },
  {
   "id": "build-alexa-routines-to-regulate-a-daily-schedule",
   "title": "Build Alexa routines to regulate a daily schedule",
   "summary": "Real-world thread: the writer assembled Alexa Routines across the day on Echo speakers and other smart devices, after getting tips from a psychotherapist. At 11AM and 3PM Alexa asks about mood and water intake and reminds her about lunch. She says the routines helped her regulate her day and improved her mental health and productivity.",
   "agent": "Alexa+",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.techradar.com/home/smart-home/i-tried-using-smart-home-devices-to-help-fix-my-irregular-routine-and-it-changed-my-life",
   "prompt": "Help me build Alexa routines to regulate a daily schedule. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/build-alexa-routines-to-regulate-a-daily-schedule.html"
  },
  {
   "id": "auto-post-scheduled-images-to-instagram-and-pinterest",
   "title": "Auto-post scheduled images to Instagram and Pinterest",
   "summary": "Case study takeaway: A Make.com scenario watches a Google Sheet that the client populates with caption, image URL from Drive, and date. On the booked date it pulls the image and posts natively to Instagram (carousel or reel) plus Pinterest (rich pin). After 30 days the brand reached 100% posting consistency (vs ~60% before), +34% Instagram reach, and +1.",
   "agent": "Make",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/lcraftlab/status/2107894270669582479",
   "prompt": "Creative task: auto-post scheduled images to Instagram and Pinterest. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/auto-post-scheduled-images-to-instagram-and-pinterest.html"
  },
  {
   "id": "add-home-todo-items-by-telegram-message",
   "title": "Add home TODO items by Telegram message",
   "summary": "Observed in the wild: the builder ran OpenClaw and wrote a custom API that adds items to a TODO list kept only on their home PC, with no phone syncing. From the office they sent a Telegram message, and the agent called that API to add the item. It worked well enough that The builder misses it.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47724016",
   "prompt": "Assistant mode: add home TODO items by Telegram message. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 47,
   "page": "https://agentsightings.com/case/add-home-todo-items-by-telegram-message.html"
  },
  {
   "id": "briggs-stratton-assist-supply-chain-agents",
   "title": "A small-engine maker's customer agent grows into the supply chain",
   "summary": "Google says Briggs & Stratton's customer-service agent, Briggs & Stratton Assist, led to supply-chain agents that use SAP data in BigQuery to handle volume change controls, exception messages and safety stock in seconds. Staff have built 25+ low-code agents. Vendor-published, no figures beyond that.",
   "agent": "Briggs & Stratton Assist (Gemini Enterprise)",
   "topic": "Business & Startups",
   "source": "Blog (Google Cloud)",
   "url": "https://cloud.google.com/transform/agentic-factory-manufacturing-new-age-of-autonomy-industrial-ai",
   "prompt": "Review the SAP exception messages for today, decide which are routine, resolve those, and list the rest for a planner.",
   "added": "2026-10-29",
   "interest": 46,
   "page": "https://agentsightings.com/case/briggs-stratton-assist-supply-chain-agents.html"
  },
  {
   "id": "bob-open-source-agent-170-tools-built-in-3-days",
   "title": "A local agent with 170 tools, built in three days",
   "summary": "A developer built Bob, an open-source agent that runs on your own machine with 170+ tools covering calendar, expenses, invoices, Gmail, eBay and Etsy listings, Home Assistant, phone calls and browser automation. All tools are first-party code rather than a public skill marketplace.",
   "agent": "Bob",
   "topic": "Personal Assistant",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46943263",
   "prompt": "Run on my laptop only. Handle my calendar, expenses, invoices and email, but only with tools you ship yourself, and ask me before any action that goes outside the house.",
   "added": "2026-10-29",
   "interest": 46,
   "page": "https://agentsightings.com/case/bob-open-source-agent-170-tools-built-in-3-days.html"
  },
  {
   "id": "turn-product-photos-into-video-ads",
   "title": "Turn product photos into video ads",
   "summary": "According to the write-up, custom agent shows up in a write-up about turn product photos into video ads. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1rwghv5/are_people_seriously_having_ai%20automatically_doing_their_business",
   "prompt": "Creative task: turn product photos into video ads. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/turn-product-photos-into-video-ads.html"
  },
  {
   "id": "turn-a-keyword-into-approved-instagram-content",
   "title": "Turn a keyword into approved Instagram content",
   "summary": "Linked example of turn a keyword into approved Instagram content. The setup was a no-code Make.com workflow powered by OpenAI, Gmail, and Google Sheets.",
   "agent": "Make",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/maketemplates/comments/1s4g5wc/i_built_an_instagram_content_pipeline_on_makecom",
   "prompt": "Creative task: turn a keyword into approved Instagram content. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/turn-a-keyword-into-approved-instagram-content.html"
  },
  {
   "id": "triage-client-messages-into-tickets-and-replies",
   "title": "Triage client messages into tickets and replies",
   "summary": "A published walkthrough shows A Slack watcher on one client thread triggers a Devin session on each new message. Devin investigates the issue on its own, creates a Linear ticket, and sketches a response. The setup automates triage, ticketing, and drafting for customer-service messages, routing all client contact through the human for approval.",
   "agent": "Devin",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/russelljkaplan/status/2057932036892197337",
   "prompt": "Help me triage client messages into tickets and replies. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/triage-client-messages-into-tickets-and-replies.html"
  },
  {
   "id": "take-orders-for-a-sushi-chain",
   "title": "Take orders for a sushi chain",
   "summary": "One practitioner notes that constructed a Claude representative who accepts requests for a sushi chain with seven locations, utilizing information on the menu, ingredients, delivery location, book, and promotional materials.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1ubktm6/i_put_a_claude_agent_in_charge_of_taking_orders",
   "prompt": "Help me take orders for a sushi chain. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/take-orders-for-a-sushi-chain.html"
  },
  {
   "id": "search-linkedin-and-submit-tailored-job-applications",
   "title": "Search LinkedIn and submit tailored job applications",
   "summary": "As described in the source, chrome runs with remote debugging on, and Claude Code connects to it through Selenium scripts it writes itself. A markdown skill prompt, a markdown resume and a cover-letter style template guide it. A search run yields 50-130 postings in minutes, and each application takes 60-90 seconds.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1sd2f37/i_built_an_ai_job_search_system_with_claude_code",
   "prompt": "Help me search LinkedIn and submit tailored job applications. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/search-linkedin-and-submit-tailored-job-applications.html"
  },
  {
   "id": "schedule-meetings-tidy-drive-and-log-reward-balances",
   "title": "Schedule meetings, tidy Drive and log reward balances",
   "summary": "Real-world thread: the writer installed the Claude in Chrome extension and granted it broad access to their browsing. For meetings, Claude read Google Calendar and drafted an email proposing open slots. Drive went from close to 900 loose files to six top folders with subfolders, and nearly 50 duplicates were removed.",
   "agent": "Claude",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://www.techradar.com/ai-platforms-assistants/claude/i-tried-the-new-claude-in-chrome-extension-and-it-delivered-convenience-with-a-side-of-digital-paranoia",
   "prompt": "Assistant mode: schedule meetings, tidy Drive and log reward balances. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/schedule-meetings-tidy-drive-and-log-reward-balances.html"
  },
  {
   "id": "schedule-60-short-videos-in-youtube-studio",
   "title": "Schedule 60 short videos in YouTube Studio",
   "summary": "Real-world thread: opus 5.5 generated 60 short videos for the Kepler IPTV app and suggested scheduling them in YouTube Studio. Someone prompted it to handle the task itself. All 60 shorts were booked.",
   "agent": "Claude",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/dailyvibecoding/status/2107901668763304254",
   "prompt": "Creative task: schedule 60 short videos in YouTube Studio. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/schedule-60-short-videos-in-youtube-studio.html"
  },
  {
   "id": "run-life-admin-and-to-dos-from-claude-code",
   "title": "Run life admin and to-dos from Claude Code",
   "summary": "Real-world thread: hilary Gridley uses Claude Code as a simple life-admin system she calls an anti-system system. An iPhone back-tap shortcut captures to-dos without switching apps. The source is a podcast page with a topic list and no measured results.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/how-to-turn-claude-code-into-your",
   "prompt": "Creative task: run life admin and to-dos from Claude Code. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/run-life-admin-and-to-dos-from-claude-code.html"
  },
  {
   "id": "run-content-creation",
   "title": "Run content creation",
   "summary": "One practitioner notes that grok Bot shows up in a write-up about run content creation, with figures like 44. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://stealwhatworks.com/blogs/news/grok-bot-content-creation",
   "prompt": "Creative task: run content creation. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/run-content-creation.html"
  },
  {
   "id": "reconcile-monthly-business-transactions",
   "title": "Reconcile monthly business transactions",
   "summary": "From the linked report, rather than pasting a CSV into chat, the author had Claude Code build a re-runnable pipeline.",
   "agent": "Claude Code",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://dev.to/kenimo49/i-let-claude-code-run-a-month-of-my-business-books-it-reconciled-200-transactions-and-513d",
   "prompt": "Finance task: reconcile monthly business transactions. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/reconcile-monthly-business-transactions.html"
  },
  {
   "id": "produce-cinematic-video",
   "title": "Produce cinematic video",
   "summary": "A published walkthrough shows utilizing three bespoke agents, Grok Imagine was established as a cinematic studio, providing prompt direction, audio scripting, and consistent video content with consistent characters.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/grok/comments/1spehdf/full_guide_i_built_a_complete_cinematic_studio",
   "prompt": "Creative task: produce cinematic video. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/produce-cinematic-video.html"
  },
  {
   "id": "produce-a-short-form-video-from-research-to-captions",
   "title": "Produce a short-form video from research to captions",
   "summary": "One practitioner notes that the builder prompted Hermes Agent step by step. It searched the web and compared sources, then wrote a research document and recommended a topic. The result is a rendered vertical video for Shorts, Reels or TikTok.",
   "agent": "Hermes Agent",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://dev.to/vivek_shetye/build-an-end-to-end-ai-video-production-pipeline-with-hermes-agent-and-remotion-183i",
   "prompt": "Creative task: produce a short-form video from research to captions. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/produce-a-short-form-video-from-research-to-captions.html"
  },
  {
   "id": "produce-a-full-video",
   "title": "Produce a full video",
   "summary": "Case study takeaway: grok Bot shows up in a write-up about produce a full video. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/dvorahfr/status/2050711692204490901",
   "prompt": "Creative task: produce a full video. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/produce-a-full-video.html"
  },
  {
   "id": "produce-a-finished-music-release-from-one-chat",
   "title": "Produce a finished music release from one chat",
   "summary": "Linked example of produce a finished music release from one chat. Claude Code, connected to the AetherWave Studio connector, wrote the lyrics in the chat.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://media.aetherwavestudio.com/campaigns/use-cases/01-one-chat-away/index.html",
   "prompt": "Creative task: produce a finished music release from one chat. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/produce-a-finished-music-release-from-one-chat.html"
  },
  {
   "id": "organize-deal-flow",
   "title": "Organize deal flow",
   "summary": "Observed in the wild: grok Bot shows up in a write-up about organize deal flow. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/JamesonCamp/article/2091970409612898369",
   "prompt": "GTM task: organize deal flow. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/organize-deal-flow.html"
  },
  {
   "id": "order-lunch-through-an-ai-agent",
   "title": "Order lunch through an AI agent",
   "summary": "Real-world thread: abhishek gave an AI agent his CVV and OTP so it could place a lunch order. The agent completed the purchase and payment independently. The food was delivered without any intervention.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/_logik/status/2098381294367633552",
   "prompt": "Help me order lunch through an AI agent. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/order-lunch-through-an-ai-agent.html"
  },
  {
   "id": "migrate-toolchain-architecture-and-site-design",
   "title": "Migrate toolchain, architecture and site design",
   "summary": "According to the write-up, claude Code shows up in a write-up about migrate toolchain, architecture and site design. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://sergioazocar.com/en/blog/how-to-migrate-with-claude-code-and-not-die-trying/",
   "prompt": "Creative task: migrate toolchain, architecture and site design. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/migrate-toolchain-architecture-and-site-design.html"
  },
  {
   "id": "hunt-jobs-tailor-resumes-and-apply-automatically",
   "title": "Hunt jobs, tailor resumes and apply automatically",
   "summary": "As described in the source, user described desired job type, schedule and pay to the Resume bot at x.ai. The bot searched listings, ranked them by fit, drafted tailored resumes for top matches, and tracked applicant status. Secured 3 interviews and 1 job offer with start date the following week that aligned with existing schedule and finances.",
   "agent": "Devin",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/ItsMeTrueG/status/2107417874738639053",
   "prompt": "Help me hunt jobs, tailor resumes and apply automatically. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/hunt-jobs-tailor-resumes-and-apply-automatically.html"
  },
  {
   "id": "find-web-design-prospects",
   "title": "Find web design prospects",
   "summary": "As described in the source, chatGPT shows up in a write-up about find web design prospects. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/passive_income/comments/1rmsb2o/i_set_up_6_ai_agents_that_sell_web_services_24_7",
   "prompt": "Creative task: find web design prospects. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/find-web-design-prospects.html"
  },
  {
   "id": "evaluate-hiring-candidates-against-competency-levels",
   "title": "Evaluate hiring candidates against competency levels",
   "summary": "From the linked report, they assembled a basic hiring agent in Copilot Pro and loaded it with their company’s competency definitions and levels through SharePoint as the knowledge base. The agent was used to analyze job descriptions, recommend which competencies were critical for a role, and generate behavioral and situational interview questions plus a scoring rubric; the human set it up in natural language and tested it a few times to make sure it worked correctly. The commenter says it was very simple, required no coding, and saved them lots of time.",
   "agent": "GitHub Copilot",
   "topic": "Business & Startups",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/humanresources/comments/1pol44d/ai_agent_use_cases_in_hr_na",
   "prompt": "Biz task: evaluate hiring candidates against competency levels. Break into steps, show intermediate output, and pause before irreversible or public actions.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/evaluate-hiring-candidates-against-competency-levels.html"
  },
  {
   "id": "download-a-youtube-video-for-offline-review",
   "title": "Download a YouTube video for offline review",
   "summary": "A simple Claude Code request downloads a YouTube video locally when the user needs offline access.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Download this YouTube video to ~/Downloads in the best available quality under 1080p and confirm the saved filename.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/download-a-youtube-video-for-offline-review.html"
  },
  {
   "id": "download-google-doc-images-in-high-res",
   "title": "Download Google Doc images in high res",
   "summary": "Claude Code pulls every embedded image from a Google Doc at high resolution for reuse in other materials.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Download all embedded images from this Google Doc at the highest available resolution into ./assets and name them from nearby heading text.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/download-google-doc-images-in-high-res.html"
  },
  {
   "id": "create-schedule-and-post-content-across-platforms-automatically",
   "title": "Create, schedule and post content across platforms automatically",
   "summary": "According to the write-up, n8n shows up in a write-up about create, schedule and post content across platforms automatically, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8nbusinessautomation/comments/1qsmnly/built_an_ai_agent_system_that_handles_content",
   "prompt": "Creative task: create, schedule and post content across platforms automatically. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/create-schedule-and-post-content-across-platforms-automatically.html"
  },
  {
   "id": "contact-job-candidates-for-initial-hr-screening",
   "title": "Contact job candidates for initial HR screening",
   "summary": "Linked example of contact job candidates for initial HR screening. The post describes a company powered by an automated recruiter that first emailed from a generic recruiting address, then placed a phone call that immediately launched into a scripted screening once the candidate answered.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/recruitinghell/comments/1mq851o/just_got_my_first_ai_hr_screening_call_oh_dear",
   "prompt": "Help me contact job candidates for initial HR screening. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/contact-job-candidates-for-initial-hr-screening.html"
  },
  {
   "id": "compare-coca-cola-prices-across-belgian-supermarkets",
   "title": "Compare Coca-Cola prices across Belgian supermarkets",
   "summary": "Real-world thread: peter van Hees asked Grok Bot in plain language to compare Coca-Cola prices at the major Belgian supermarkets, without naming shops or sizes. The bot worked from the official store sites rather than comparison blogs. It described 1.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.petervanhees.com/grok-bot-knows-when-to-stop/",
   "prompt": "Help me compare Coca-Cola prices across Belgian supermarkets. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/compare-coca-cola-prices-across-belgian-supermarkets.html"
  },
  {
   "id": "buy-a-microwave-from-costco",
   "title": "Buy a microwave from Costco",
   "summary": "According to the write-up, someone's Meta Muse agent bought a microwave from Costco. The purchase was completed.",
   "agent": "Meta Muse",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/lissimore1/status/2107660512112607428",
   "prompt": "Help me buy a microwave from Costco. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/buy-a-microwave-from-costco.html"
  },
  {
   "id": "build-your-own-video-editing-agents-on-a-canvas",
   "title": "Build your own video-editing agents on a canvas",
   "summary": "Observed in the wild: the page describes Mosaic as a canvas for agentic video editing. You run edits on autopilot and make several variants from the same raw footage to A/B test. The page gives no results, numbers, or limitations.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=45980760",
   "prompt": "Creative task: build your own video-editing agents on a canvas. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/build-your-own-video-editing-agents-on-a-canvas.html"
  },
  {
   "id": "build-a-dramatic-digital-envelope-app-for-asking-friends-for-money",
   "title": "Build a dramatic digital envelope app for asking friends for money",
   "summary": "Buildathon gag app: fake chat bubbles, drum rolls, live progress bars—then the real ask for cash. Stack was Replit Agent with Design Mode driving the UI theatrics.",
   "agent": "Replit Agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/AbdullahEr0glu/status/2086113172847513862",
   "prompt": "Creative task: build a dramatic digital envelope app for asking friends for money. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/build-a-dramatic-digital-envelope-app-for-asking-friends-for-money.html"
  },
  {
   "id": "block-spam-and-deliver-morning-briefings",
   "title": "Block spam and deliver morning briefings",
   "summary": "Community report: someone gave their Grok bot access to all email accounts, LinkedIn, calendar, and an ongoing business launch. The agent blocks and removes spam from email and LinkedIn, blocks spam calendar invites, and produces a daily morning update covering critical items plus any unfinished personal or business to-dos from the prior day. No results, metrics, or limitations are stated.",
   "agent": "Grok Bot",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/buddy7/status/2107824169849139380",
   "prompt": "Assistant mode: block spam and deliver morning briefings. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/block-spam-and-deliver-morning-briefings.html"
  },
  {
   "id": "batch-convert-and-translate-audio-files",
   "title": "Batch-convert and translate audio files",
   "summary": "A creator uses Claude Code to change sample rates, rename, and translate Portuguese audio files to English in bulk.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "For every audio file in this folder, convert to 44.1 kHz WAV, rename consistently, and produce an English transcript for any Portuguese file. Report what you changed.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/batch-convert-and-translate-audio-files.html"
  },
  {
   "id": "apply-for-jobs-and-turn-notes-into-coding-prompts",
   "title": "Apply for jobs and turn notes into coding prompts",
   "summary": "Case study takeaway: one commenter uses Grok Bot through the Cursor CLI: they link it to their developer docs, send it shorthand notes while away from their desk, and it rewrites those notes into a full prompt before having Cursor start coding. The same commenter says a separate small team of bots covers job hunting by searching for jobs and applying to them after being given their resume, a cover letter, and an FAQ sheet about them. The thread presents both as active personal automations, but gives no numbers on applications, interviews, hires, code output, or time saved.",
   "agent": "Grok Bot",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1wiuwmw/whats_the_most_useful_grokbot_automation_youve",
   "prompt": "Notes task: apply for jobs and turn notes into coding prompts. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/apply-for-jobs-and-turn-notes-into-coding-prompts.html"
  },
  {
   "id": "answer-home-service-customer-calls-and-text-the-owner",
   "title": "Answer home-service customer calls and text the owner",
   "summary": "Documented elsewhere: the voice agent answers calls for home-service businesses. It records the caller's name and number, the request in the customer's own words, and a promised update time if the owner cannot respond immediately, then sends that summary via text. The post describes the flow closing the gap between unanswered calls and owner callbacks but gives no results or numbers.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/TheChattyAi/status/2104995656331055203",
   "prompt": "Help me answer home-service customer calls and text the owner. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 46,
   "page": "https://agentsightings.com/case/answer-home-service-customer-calls-and-text-the-owner.html"
  },
  {
   "id": "vertafore-agents-enter-and-check-insurance-policies",
   "title": "Agents key in insurance policies and check what was issued",
   "summary": "Vertafore released two agents for independent insurance agencies: one moves policy information from documents into the management system, and one checks that the issued policy matches what was intended.",
   "agent": "Velocity Policy Intake and Policy Check Agents",
   "topic": "Business & Startups",
   "source": "News (Vertafore)",
   "url": "https://www.vertafore.com/resources/press-releases/vertafore-automates-policy-processing-two-new-velocity-ai-agents",
   "prompt": "Read this policy PDF and enter the details into our agency system. Then compare the issued policy with the quote we asked for and list every difference in coverage, limits or price.",
   "added": "2026-10-29",
   "interest": 45,
   "page": "https://agentsightings.com/case/vertafore-agents-enter-and-check-insurance-policies.html"
  },
  {
   "id": "sam-hartley-agents-article-writing-minutes",
   "title": "A developer says his agents cut an article from hours to 15 minutes",
   "summary": "Sam Hartley says agents running on a Mac mini 24/7 automated his dev workflow, and that writing an article went from 2-3 hours to 10-15 minutes. Self-reported, with no measurement method given.",
   "agent": "Agents on a Mac mini",
   "topic": "Software Development",
   "source": "Blog (DEV Community)",
   "url": "https://dev.to/samhartley_dev/i-automated-my-entire-dev-workflow-with-ai-agents-running-247-on-a-mac-mini-1ikc",
   "prompt": "Draft an article from these notes and my previous posts in my voice. Mark any claim you could not check.",
   "added": "2026-10-29",
   "interest": 45,
   "page": "https://agentsightings.com/case/sam-hartley-agents-article-writing-minutes.html"
  },
  {
   "id": "turn-ideas-into-finished-videos",
   "title": "Turn ideas into finished videos",
   "summary": "According to the write-up, user assembled custom agent \"dimul\" that connects Grok via CLI powered by SuperGrok subscription. Grok acted as orchestrator for Imagine to generate clips from its prompts. Produced finished videos shown in the post and follow-up example.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/XFreeze/status/2107878966510158087",
   "prompt": "Creative task: turn ideas into finished videos. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/turn-ideas-into-finished-videos.html"
  },
  {
   "id": "run-roadmap-support-inbox-calendar-and-todos",
   "title": "Run roadmap, support inbox, calendar, and todos",
   "summary": "As described in the source, the agent runs on a DigitalOcean VPS with its own forwarded email inbox, delegate X account via custom browser proxy, Linear, Obsidian for todos/PKM, and PDF extraction skill. It creates Linear cards for issues, prioritizes and flags support inbox items, schedules calendar events from images/PDFs, checks X notifications and DMs, and tracks sponsorships. After 10-14 hours of setup the agent already pays back massive dividends across roadmap, support, calendar, and todos.",
   "agent": "OpenClaw",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/euboid/status/2019094563369910692",
   "prompt": "Go through my mail for this goal: run roadmap, support inbox, calendar, and todos. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/run-roadmap-support-inbox-calendar-and-todos.html"
  },
  {
   "id": "reclaim-forgotten-crypto-airdrops-and-deposits",
   "title": "Reclaim forgotten crypto airdrops and deposits",
   "summary": "A published walkthrough shows someone gave Claude and GPT agents access to their full blockchain transaction history. The agents scanned every past transaction and attempted to recompose each one to check whether any deposit lacked a matching withdrawal. This recovered no longer remembered airdrops and deposits worth 10x the annual cost of the AI subscriptions.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/pscltllrd/status/2109733789240811787",
   "prompt": "Help me reclaim forgotten crypto airdrops and deposits. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/reclaim-forgotten-crypto-airdrops-and-deposits.html"
  },
  {
   "id": "prepare-plan-execute-review-for-non-code-work",
   "title": "Prepare, plan, execute, review for non-code work",
   "summary": "A designer runs brand-system work through Claude Code using plan mode, then a review prompt that lists issues only so they can choose fixes.",
   "agent": "Claude Code",
   "topic": "Design & Creative",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1n4t349/my_prepre_plan_execute_review_process_for_using/",
   "prompt": "Review the output against the brief and the attached references. List issues only, numbered, without suggesting fixes. I'll pick which ones to address.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/prepare-plan-execute-review-for-non-code-work.html"
  },
  {
   "id": "monitor-reddit-draft-replies-and-schedule-posts",
   "title": "Monitor Reddit, draft replies and schedule posts",
   "summary": "Observed in the wild: the commenter says they assembled a personal Claude agent as a social media manager powered by the Agent SDK plus a Claude Code subscription. In the thread and later follow-up, they say it keeps watch on Reddit, sketches replies, and schedules posts, powered by old posts for tone and account stats plus current-vs-previous-period comparisons for performance; sketches still wait for their approval before scheduling. They report it works within their Claude subscription if they route cheaper drafting to Haiku and final output to Sonnet.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rmhayt/claudeai_as_social_media_manager",
   "prompt": "Help me monitor Reddit, draft replies and schedule posts. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/monitor-reddit-draft-replies-and-schedule-posts.html"
  },
  {
   "id": "let-ai-agents-search-a-store-build-carts-and-prep-checkout",
   "title": "Let AI agents search a store, build carts, and prep checkout",
   "summary": "Documented elsewhere: the post is a pitch rather than a walkthrough: The builder says they can make ecommerce sites “agentic commerce ready” so ChatGPT users can discover products through Vendable. The concrete mechanics come from a seller’s comment, which says their store exposes its own MCP server so Claude or ChatGPT agents can search the catalog, create a custom tee, build a cart, and start checkout, with payment completed on the store site. The thread gives no performance numbers or rollout results.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_In_ECommerce/comments/1vcx8s9/i_made_chatgpt_find_and_buy_products_from_real",
   "prompt": "Shopping task: let AI agents search a store, build carts, and prep checkout. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/let-ai-agents-search-a-store-build-carts-and-prep-checkout.html"
  },
  {
   "id": "index-a-newsletter-archive-and-repurpose-writing",
   "title": "Index a newsletter archive and repurpose writing",
   "summary": "Real-world thread: the guide says Casey Newton of Platformer indexed 818 newsletter editions into a searchable archive. Darren Rowse audited about 1,000 notes in an Obsidian vault, which Claude Code reads as plain markdown folders. Newton's index took under 30 minutes.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://ccforeveryone.com/guides/claude-code-for-writers",
   "prompt": "Go through my mail for this goal: index a newsletter archive and repurpose writing. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/index-a-newsletter-archive-and-repurpose-writing.html"
  },
  {
   "id": "hand-everyday-errands-to-a-local-agent",
   "title": "Hand everyday errands to a local agent",
   "summary": "Community report: the commenter runs a local agent they trust not to share conversations, and uses it like a search engine in place of a browser. They hand it errands: plan flights, a hotel and sights for a conference talk, convert downloaded FLAC music to OGG for a player, find DRM-free epubs for owned books, and make a tweakable OpenSCAD holder from a photo beside a tape measure. The comment only lists use cases.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49980512",
   "prompt": "Go through my mail for this goal: hand everyday errands to a local agent. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/hand-everyday-errands-to-a-local-agent.html"
  },
  {
   "id": "create-a-subtitled-video-from-one-prompt",
   "title": "Create a subtitled video from one prompt",
   "summary": "From the linked report, user entered a simple prompt into Agent Mode in Grok Imagine. The agent took the prompt and a single image, then generated matching scenes from different angles, assembled a full visual storyline, and added subtitles to produce the complete video on its own. It produced a 36-second video in five minutes.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/XFreeze/status/2078833307207483764",
   "prompt": "Creative task: create a subtitled video from one prompt. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/create-a-subtitled-video-from-one-prompt.html"
  },
  {
   "id": "compile-an-annotated-bibliography-and-draft-a-paper",
   "title": "Compile an annotated bibliography and draft a paper",
   "summary": "ChatGPT Agent covered compile an annotated bibliography and draft a paper. The builder gave ChatGPT Agent a single /agent prompt asking for peer-reviewed papers from the past five years on Big Data and Analytics in decision-making, focused on Education and Healthcare.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://datawithsultan.medium.com/i-put-chatgpt-agent-to-the-test-heres-how-it-helped-me-write-an-academic-paper-984327423d95",
   "prompt": "Research brief: compile an annotated bibliography and draft a paper. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/compile-an-annotated-bibliography-and-draft-a-paper.html"
  },
  {
   "id": "build-e-ink-fridge-shopping-list-firmware",
   "title": "Build e-ink fridge shopping list firmware",
   "summary": "According to the write-up, the repo holds about 2,400 lines of C++ firmware for the M5Stack PaperMono, an ESP32-S3 with an e-paper touchscreen. A small FastAPI and SQLite server sits between the device and a phone web UI. It works as a fridge shopping list grouped by aisle, with offline use and queued edits.",
   "agent": "Claude Code",
   "topic": "Shopping",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49875801",
   "prompt": "Shopping task: build e-ink fridge shopping list firmware. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/build-e-ink-fridge-shopping-list-firmware.html"
  },
  {
   "id": "answer-dental-clinic-calls-and-manage-appointments",
   "title": "Answer dental clinic calls and manage appointments",
   "summary": "One practitioner notes that the builder describes a six-part setup: an always-on voice agent covers booking, rescheduling, cancellations, slot checks, and routine questions, while urgent calls trigger a Telegram alert to the doctor. Calls are then analyzed with GPT-4 and logged into Supabase CRM, and bookings sync with Calcom; the automations run in self-hosted n8n with ElevenLabs/Vapi for voice. The stated outcome was fewer missed calls, a measurable drop in no-shows, and automatic reactivation of patients idle for 90+ days, but the post gives no hard metrics.",
   "agent": "n8n",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Automate/comments/1m5juzf",
   "prompt": "Help me answer dental clinic calls and manage appointments. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 45,
   "page": "https://agentsightings.com/case/answer-dental-clinic-calls-and-manage-appointments.html"
  },
  {
   "id": "vanderlande-copilot-studio-1500-hours-monthly",
   "title": "Agents replace spreadsheet reporting and chasing suppliers",
   "summary": "Microsoft says Vanderlande, a logistics automation company, uses Power Platform and Copilot Studio to replace spreadsheet-based operational reporting and supplier communications, saving over 1,500 manual hours a month. Vendor-published, with few details beyond that.",
   "agent": "Copilot Studio / Power Platform",
   "topic": "Automation",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/27170-vanderlande-microsoft-power-platform",
   "prompt": "Build the weekly operations report from the source systems instead of our spreadsheet, and email suppliers whose figures are missing.",
   "added": "2026-10-29",
   "interest": 44,
   "page": "https://agentsightings.com/case/vanderlande-copilot-studio-1500-hours-monthly.html"
  },
  {
   "id": "sompo-10000-custom-agents-document-search",
   "title": "10,000+ custom agents for a Japanese insurer's staff",
   "summary": "Google says SOMPO, a Japanese insurer, built more than 10,000 custom agents across its 34,000 employees for tasks like document search and summarisation. Its engineers also cut development time for new models from a week to a day. Vendor-reported.",
   "agent": "Gemini Enterprise agents",
   "topic": "Knowledge & Notes",
   "source": "News (Google Cloud)",
   "url": "https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026",
   "prompt": "Search our internal documents for what I describe, summarise the matching passages and cite the file for each one.",
   "added": "2026-10-29",
   "interest": 44,
   "page": "https://agentsightings.com/case/sompo-10000-custom-agents-document-search.html"
  },
  {
   "id": "watch-item-prices-and-email-on-drops",
   "title": "Watch item prices and email on drops",
   "summary": "Linked example of watch item prices and email on drops. The commenter set up a recurring ChatGPT task powered by the schedule tool rather than separate watches for each item.",
   "agent": "ChatGPT Agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1wqm0j0/what_are_the_crazy_mind_blowing_ways_of_using",
   "prompt": "Go through my mail for this goal: watch item prices and email on drops. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/watch-item-prices-and-email-on-drops.html"
  },
  {
   "id": "triage-inbox-draft-replies-and-manage-calendar-from-a-desktop-agent",
   "title": "Triage inbox, draft replies and manage calendar from a desktop agent",
   "summary": "A published walkthrough shows microsoft Scout shows up in a write-up about triage inbox, draft replies and manage calendar from a desktop agent, with figures like 365. The linked post is the primary source for steps and outcomes.",
   "agent": "Microsoft Scout",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://intelligink.com/blog/microsoft-scout-desktop-agent-review/",
   "prompt": "Go through my mail for this goal: triage inbox, draft replies and manage calendar from a desktop agent. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/triage-inbox-draft-replies-and-manage-calendar-from-a-desktop-agent.html"
  },
  {
   "id": "synthesize-customer-calls-into-product-evidence",
   "title": "Synthesize customer calls into product evidence",
   "summary": "A product person pipes Fireflies transcripts through Claude Code with Linear and Notion MCP to tally which assumptions customer calls support or kill.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Read this week's customer call transcripts. For each open assumption in assumptions.md, add supporting or contradicting quotes with links, and update the confidence score.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/synthesize-customer-calls-into-product-evidence.html"
  },
  {
   "id": "summarize-emails-and-draft-replies",
   "title": "Summarize emails and draft replies",
   "summary": "Case study takeaway: gemini shows up in a write-up about summarize emails and draft replies. The linked post is the primary source for steps and outcomes.",
   "agent": "Gemini",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.tomsguide.com/ai/i-turned-gemini-gems-into-automated-google-workspace-agents-heres-how",
   "prompt": "Go through my mail for this goal: summarize emails and draft replies. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/summarize-emails-and-draft-replies.html"
  },
  {
   "id": "summarize-email-and-prep-briefings-from-your-phone",
   "title": "Summarize email and prep briefings from your phone",
   "summary": "Linked example of summarize email and prep briefings from your phone. Over a month, the writer used Dispatch, which links the Claude phone app to Claude Desktop and runs tasks in Cowork on her computer.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://fortune.com/2026/04/28/claude-dispatch-feature-capabilities-service/",
   "prompt": "Go through my mail for this goal: summarize email and prep briefings from your phone. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/summarize-email-and-prep-briefings-from-your-phone.html"
  },
  {
   "id": "sort-2-200-downloads-files-into-folders",
   "title": "Sort 2,200 Downloads files into folders",
   "summary": "A published walkthrough shows in under 20 minutes, Coworkers managed to summarize a Downloads folder of 2,200 loose files by type, mostly without any destructorial intervention.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1up77np/ai_as_coworkers_tools_not_just_coding_agents",
   "prompt": "Help me sort 2,200 Downloads files into folders. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/sort-2-200-downloads-files-into-folders.html"
  },
  {
   "id": "run-personal-life-admin",
   "title": "Run personal life admin",
   "summary": "Real-world thread: claude Code shows up in a write-up about run personal life admin. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/thmsmlr/status/2001166615744786890",
   "prompt": "Go through my mail for this goal: run personal life admin. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/run-personal-life-admin.html"
  },
  {
   "id": "run-a-consulting-business",
   "title": "Run a consulting business",
   "summary": "Documented elsewhere: manus shows up in a write-up about run a consulting business. The linked post is the primary source for steps and outcomes.",
   "agent": "Manus",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ManusOfficial/comments/1ru1esg/i_use_manus_as_my_personal_chief_of_staff_to_get",
   "prompt": "Go through my mail for this goal: run a consulting business. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/run-a-consulting-business.html"
  },
  {
   "id": "run-a-24-7-agent-for-email-forms-and-calendar-tasks",
   "title": "Run a 24/7 agent for email, forms and calendar tasks",
   "summary": "Community report: one Discord someone (@monty_13277) keeps Claude for day-to-day chat and operates Hermes around the clock on a mini PC for email, web browsing, form filling, calendar updates and cron jobs. Another user (@kouff) told Hermes to manage tasks in Obsidian and cross-check them against Apple Calendar. The source gives no numbers or measured results.",
   "agent": "Hermes Agent",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://hermes-agent.nousresearch.com/docs/user-stories",
   "prompt": "Go through my mail for this goal: run a 24/7 agent for email, forms and calendar tasks. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/run-a-24-7-agent-for-email-forms-and-calendar-tasks.html"
  },
  {
   "id": "prep-meetings-and-check-inbox-against-calendar",
   "title": "Prep meetings and check inbox against calendar",
   "summary": "According to the write-up, meta Muse shows up in a write-up about prep meetings and check inbox against calendar. The linked post is the primary source for steps and outcomes.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.aiagentslibrary.com/blog/meta-muse-prompts/",
   "prompt": "Go through my mail for this goal: prep meetings and check inbox against calendar. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/prep-meetings-and-check-inbox-against-calendar.html"
  },
  {
   "id": "maintain-a-profitable-stripe-backed-app",
   "title": "Maintain a profitable Stripe-backed app",
   "summary": "One practitioner notes that replit Agent shows up in a write-up about maintain a profitable Stripe-backed app, with figures like 20. The linked post is the primary source for steps and outcomes.",
   "agent": "Replit Agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1sfkflq/anyone_here_built_a_productionready_app_using",
   "prompt": "GTM task: maintain a profitable Stripe-backed app. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/maintain-a-profitable-stripe-backed-app.html"
  },
  {
   "id": "list-a-quest-3-and-negotiate",
   "title": "List a Quest 3 and negotiate",
   "summary": "Real-world thread: meta Muse shows up in a write-up about list a Quest 3 and negotiate, with figures like 3. The linked post is the primary source for steps and outcomes.",
   "agent": "Meta Muse",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/mattdeitke/status/2102468747642720700",
   "prompt": "Notes task: list a Quest 3 and negotiate. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/list-a-quest-3-and-negotiate.html"
  },
  {
   "id": "inbox-triage-on-a-schedule",
   "title": "Inbox triage on a schedule",
   "summary": "A background agent re-reads incoming mail every few minutes, labels spam, rescues misfiled messages and stars threads awaiting a reply.",
   "agent": "Clor",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48375347",
   "prompt": "Every 10 minutes, review new emails in my inbox. Label obvious spam, move legitimate mail out of spam, and star any thread where I owe a reply. Report what you changed.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/inbox-triage-on-a-schedule.html"
  },
  {
   "id": "improve-screenshot-quality-for-a-newsletter",
   "title": "Improve screenshot quality for a newsletter",
   "summary": "Screenshots for a post are sharpened and cleaned up by Claude Code before publishing.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Improve the image quality of this screenshot: reduce noise, sharpen text, and export a web-ready PNG under 500 KB without cropping important UI.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/improve-screenshot-quality-for-a-newsletter.html"
  },
  {
   "id": "handle-routine-email-tasks-and-follow-up-reminders",
   "title": "Handle routine email tasks and follow-up reminders",
   "summary": "One practitioner notes that constructed an n8n personal assistant that performs roughly 70% of daily email tasks and alerts others when a task is in progress, following addressing initial glitches.",
   "agent": "n8n",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://dev.to/nova_gg/how-i-built-a-personal-assistant-ai-agent-with-n8n-in-2-hours-no-code-step-by-step-3i38",
   "prompt": "Go through my mail for this goal: handle routine email tasks and follow-up reminders. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/handle-routine-email-tasks-and-follow-up-reminders.html"
  },
  {
   "id": "handle-incoming-emails",
   "title": "Handle incoming emails",
   "summary": "Case study takeaway: grok Bot shows up in a write-up about handle incoming emails. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/XFreeze/status/2083786357390868891",
   "prompt": "Go through my mail for this goal: handle incoming emails. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/handle-incoming-emails.html"
  },
  {
   "id": "grow-an-email-list",
   "title": "Grow an email list",
   "summary": "One practitioner notes that hermes Agent shows up in a write-up about grow an email list, with figures like 1,000, 80. The linked post is the primary source for steps and outcomes.",
   "agent": "Hermes Agent",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://ryandoser.com/hermes-ai-agents/",
   "prompt": "Go through my mail for this goal: grow an email list. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/grow-an-email-list.html"
  },
  {
   "id": "give-a-teacher-a-claude-code-workspace-for-lessons-and-reports",
   "title": "Give a teacher a Claude Code workspace for lessons and reports",
   "summary": "Community report: claude Code shows up in a write-up about give a teacher a Claude Code workspace for lessons and reports. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Home & Family",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1s5m6a2/built_this_for_teachers",
   "prompt": "Home/family task: give a teacher a Claude Code workspace for lessons and reports. Keep changes reversible and ask before messages, payments, or deletes.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/give-a-teacher-a-claude-code-workspace-for-lessons-and-reports.html"
  },
  {
   "id": "find-and-email-businesses-that-need-ai",
   "title": "Find and email businesses that need AI",
   "summary": "A published walkthrough shows five Grok bots each got their own inbox via AgentMail. Bot 1 pulled 468 local businesses and filtered out 69% powered by one rule on gaps it could close in a week. 27 replies arrived in the bots' inboxes; 23 requested calls.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/ArchiveExplorer/status/2093346938494005477",
   "prompt": "Go through my mail for this goal: find and email businesses that need AI. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/find-and-email-businesses-that-need-ai.html"
  },
  {
   "id": "file-roofing-permits-and-inspection-forms-online",
   "title": "File roofing permits and inspection forms online",
   "summary": "As described in the source, price Foulger, a roofing contractor, gave Grok Bot his office paperwork over two days. It filled out and pulled an online roofing permit. All three tasks were done within two days.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://grokbotpulse.com/en/categories/email",
   "prompt": "Go through my mail for this goal: file roofing permits and inspection forms online. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/file-roofing-permits-and-inspection-forms-online.html"
  },
  {
   "id": "email-daily-clickup-digests-to-stakeholders",
   "title": "Email daily ClickUp digests to stakeholders",
   "summary": "Linked example of email daily ClickUp digests to stakeholders. They set up ClickUp with a master action-item list plus related project and contact lists, rolling stakeholder emails into each task and auto-adding the right contacts when a project is chosen.",
   "agent": "Make",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Make/comments/1u6amg9/clickup_x_makecom_automation_my_boss_wants_me_to",
   "prompt": "Go through my mail for this goal: email daily ClickUp digests to stakeholders. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/email-daily-clickup-digests-to-stakeholders.html"
  },
  {
   "id": "control-gmail-calendar-and-errands-from-a-chat-agent",
   "title": "Control Gmail, calendar and errands from a chat agent",
   "summary": "According to the write-up, openClaw runs on your own machine, with a cloud model behind it, and you message it through Telegram or WhatsApp. Users give it Gmail and Calendar access. The source is a roundup of user testimonials, with no measured results.",
   "agent": "OpenClaw",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://thenuancedperspective.substack.com/p/how-are-people-using-openclaw",
   "prompt": "Go through my mail for this goal: control Gmail, calendar and errands from a chat agent. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/control-gmail-calendar-and-errands-from-a-chat-agent.html"
  },
  {
   "id": "check-prospect-fit-and-write-cold-email-openers",
   "title": "Check prospect fit and write cold email openers",
   "summary": "Linked example of check prospect fit and write cold email openers. The poster starts with a broad company list exported from Apollo, then runs it through an AI segmentation tool they assembled with Claude Code.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/coldemail/comments/1tbzhw3/claude_code_for_cold_email",
   "prompt": "Go through my mail for this goal: check prospect fit and write cold email openers. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/check-prospect-fit-and-write-cold-email-openers.html"
  },
  {
   "id": "build-an-ai-to-do-list-web-app-in-9-hours",
   "title": "Build an AI to-do list web app in 9 hours",
   "summary": "One sitting with Replit’s Agent V2 stretched from first prompt to a working AI to-do web app—call it a long workday, roughly nine hours on the clock.",
   "agent": "Replit Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1js1825",
   "prompt": "Help me build an AI to-do list web app in 9 hours. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/build-an-ai-to-do-list-web-app-in-9-hours.html"
  },
  {
   "id": "build-a-profitable-online-business-from-scratch",
   "title": "Build a profitable online business from scratch",
   "summary": "Real-world thread: utilized content marketing on Reddit, Twitter, and LinkedIn to establish an online business that attracted over 400 visitors, and gave $50 to an OpenClaw/Claude agent for a 60-day timeframe.",
   "agent": "OpenClaw",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SideProject/comments/1r1z42o/i_gave_an_ai_agent_60_days_and_50_bucks_to_build",
   "prompt": "GTM task: build a profitable online business from scratch. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/build-a-profitable-online-business-from-scratch.html"
  },
  {
   "id": "automate-follow-ups-and-inventory-tracking-for-a-food-distributor",
   "title": "Automate follow-ups and inventory tracking for a food distributor",
   "summary": "Real-world thread: custom agent shows up in a write-up about automate follow-ups and inventory tracking for a food distributor. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1t6dt98/i_built_boring_ai_agents_for_a_food_distributor",
   "prompt": "Go through my mail for this goal: automate follow-ups and inventory tracking for a food distributor. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/automate-follow-ups-and-inventory-tracking-for-a-food-distributor.html"
  },
  {
   "id": "automate-email-replies-form-data-extraction-and-lead-notes",
   "title": "Automate email replies, form data extraction, and lead notes",
   "summary": "A published walkthrough shows and wiring small AI workflows around lead handling, with tried Lindy and Gumloop for drafting email replies.",
   "agent": "Lindy",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Application/comments/1qp9jm1/lindy_ai_vs_gumloop_comparison_after_actually_trying_them",
   "prompt": "Go through my mail for this goal: automate email replies, form data extraction, and lead notes. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/automate-email-replies-form-data-extraction-and-lead-notes.html"
  },
  {
   "id": "audit-and-rebuild-a-store-x27-s-email-automations",
   "title": "Audit and rebuild a store's email automations",
   "summary": "Observed in the wild: meta Muse shows up in a write-up about audit and rebuild a store's email automations. The linked post is the primary source for steps and outcomes.",
   "agent": "Meta Muse",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/JulianGoldieSEO/status/2105534392781373708",
   "prompt": "Go through my mail for this goal: audit and rebuild a store's email automations. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 44,
   "page": "https://agentsightings.com/case/audit-and-rebuild-a-store-x27-s-email-automations.html"
  },
  {
   "id": "triage-email-by-voice-while-walking",
   "title": "Triage email by voice while walking",
   "summary": "As described in the source, used ChatGPT mobile app with Codex remote on a paid subscription, after granting aggressive permissions and setting up a sticky note on Mac. While walking with EarPods, spoke voice commands to have it scan and triage emails (skip newsletters, handle client/contract/invoice/bug ones), archive/delete/star items, draft replies, and log action items to the note; extended to Notion/Slack and goal reviews for day/week/month. Returned from walk exercised and happy, then reviewed the sticky-note to-do list and email sketches at desk.",
   "agent": "Codex",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/alliekmiller/status/2089035175216685374",
   "prompt": "Go through my mail for this goal: triage email by voice while walking. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/triage-email-by-voice-while-walking.html"
  },
  {
   "id": "teach-an-agent-to-trade",
   "title": "Teach an agent to trade",
   "summary": "According to the write-up, and conduct log trades to Google Sheets, with @mufasadubai commissioned Manus to write a Python script that utilized ProjectX APIs to generate real-time charts and establish Telegram signals.",
   "agent": "Manus",
   "topic": "Personal Assistant",
   "source": "X",
   "url": "https://x.com/mufasadubai/status/2026221794277904433",
   "prompt": "Assistant mode: teach an agent to trade. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/teach-an-agent-to-trade.html"
  },
  {
   "id": "schedule-social-posts-through-an-ai-agent",
   "title": "Schedule social posts through an AI agent",
   "summary": "Case study takeaway: in the Reddit thread, one commenter says their setup uses Obsei to ingest social content and getlate.dev to post it across social networks, excluding Reddit. They add that Socialclaw can connect that posting stack to Claude or another AI agent so the agent can schedule posts, but the thread does not spell out the workflow beyond those tool links. The source gives no metrics or performance results.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1s34qto/building_ai_agents_for_social_media_marketing",
   "prompt": "Creative task: schedule social posts through an AI agent. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/schedule-social-posts-through-an-ai-agent.html"
  },
  {
   "id": "run-production-deploys-and-port-scans-from-telegram",
   "title": "Run production deploys and port scans from Telegram",
   "summary": "Case study takeaway: openClaw shows up in a write-up about run production deploys and port scans from Telegram. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://simonroses.com/2026/02/my-experience-using-openclaw-a-security-professionals-journey/",
   "prompt": "Assistant mode: run production deploys and port scans from Telegram. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/run-production-deploys-and-port-scans-from-telegram.html"
  },
  {
   "id": "run-nightly-briefing-pipelines",
   "title": "Run nightly briefing pipelines",
   "summary": "One practitioner notes that claude Code shows up in a write-up about run nightly briefing pipelines. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Personal Assistant",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1s2j53l/claude_code_cloud_scheduled_tasks_one_feature",
   "prompt": "Assistant mode: run nightly briefing pipelines. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/run-nightly-briefing-pipelines.html"
  },
  {
   "id": "run-browser-agents-for-small-admin-tasks",
   "title": "Run browser agents for small admin tasks",
   "summary": "One practitioner notes that anna Mills tried agentic browsers (Perplexity Comet, ChatGPT Atlas, Claude in Chrome, Claude Cowork) on small admin tasks. In these tools a chat window sits beside the page, and the AI can see the page, click and fill forms with Someone's permission. She located the tools clunky, slow and untrustworthy, but still useful for these tasks.",
   "agent": "Perplexity Comet",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://annamills.substack.com/p/agentic-ai-considerations-for-educators",
   "prompt": "Creative task: run browser agents for small admin tasks. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/run-browser-agents-for-small-admin-tasks.html"
  },
  {
   "id": "run-a-creative-workflow-from-brainstorm-to-video",
   "title": "Run a creative workflow from brainstorm to video",
   "summary": "One practitioner notes that grok's Agent Mode was Leaned on on a single canvas. It performed brainstorming, writing, image generation, editing, and video production in sequence. A 74-second demo covered the full workflow and put together half a typical creative app stack look redundant.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/MarioNawfal/status/2050391070123716884",
   "prompt": "Creative task: run a creative workflow from brainstorm to video. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/run-a-creative-workflow-from-brainstorm-to-video.html"
  },
  {
   "id": "research-on-x-summarize-youtube-transcripts-build-decks",
   "title": "Research on X, summarize YouTube transcripts, build decks",
   "summary": "Observed in the wild: user subscribed to Comet as the AI agent browser and gave it access to X and YouTube. The agent browsed X posts for research results on its own, and on YouTube it extracted video transcripts then generated summaries. Comet delivered reliable X research and YouTube summaries that Someone incorporated into decks.",
   "agent": "Perplexity Comet",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/Tape_Vector/status/2106316591465910357",
   "prompt": "Creative task: research on X, summarize YouTube transcripts, build decks. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/research-on-x-summarize-youtube-transcripts-build-decks.html"
  },
  {
   "id": "rename-and-sort-100-images-into-folders",
   "title": "Rename and sort 100+ images into folders",
   "summary": "ChatGPT Agent covered rename and sort 100+ images into folders. A commenter said they uploaded a zip file containing more than 100 images, asked ChatGPT Agent to rename each file from what it saw in the image, then create subfolders and move the images into categories.",
   "agent": "ChatGPT Agent",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1m8k3xh/agent_mode_just_released_on_plus",
   "prompt": "Creative task: rename and sort 100+ images into folders. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/rename-and-sort-100-images-into-folders.html"
  },
  {
   "id": "plan-article-topics-and-draft-web-articles",
   "title": "Plan article topics and draft web articles",
   "summary": "According to the write-up, on a Windows PC, the writer points Cowork at a folder of past articles. It reads them and saves a topic table that flags overlap with earlier pieces and suggests how to differentiate. The writer now produces 40 to 50 pieces a month, commissioned articles and notes combined, and no longer copy-pastes between chat and files.",
   "agent": "Claude Cowork",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://note.com/0724_1224/n/n81bad42b4fbd?hl=en",
   "prompt": "Notes task: plan article topics and draft web articles. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/plan-article-topics-and-draft-web-articles.html"
  },
  {
   "id": "plan-a-birthday-day",
   "title": "Plan a birthday day",
   "summary": "From the linked report, the AI assistant of a co-founder orchestrated his birthday, collecting stories from colleagues, arranging the day, reserving dinner, and placing a cake.",
   "agent": "Custom agent",
   "topic": "Business & Startups",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/FoundersBuildwithAI/comments/1vl4ckr/i_created_an_ai_agent_in_about_1_minute_and_let",
   "prompt": "Biz task: plan a birthday day. Break into steps, show intermediate output, and pause before irreversible or public actions.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/plan-a-birthday-day.html"
  },
  {
   "id": "manage-rentals-and-arrange-apartment-tours",
   "title": "Manage rentals and arrange apartment tours",
   "summary": "A published walkthrough shows claude in Chrome is utilized by a property manager's team for daily management tasks, along with a database that enables an AI to respond to inquiries and monitor schedules, aiding in arranging apartment visits.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.almanacnews.com/technology/2026/10/06/how-ai-is-changing-the-apartment-search-is-it-working/",
   "prompt": "Help me manage rentals and arrange apartment tours. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/manage-rentals-and-arrange-apartment-tours.html"
  },
  {
   "id": "make-a-dinner-reservation",
   "title": "Make a dinner reservation",
   "summary": "A published walkthrough shows marily documented her initial attempt to book a dinner reservation powered by Operator, finding it to be slower than anticipated due to the absence of searching and detail entry.",
   "agent": "Operator",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://marily.substack.com/p/got-access-to-openais-operator-and",
   "prompt": "Notes task: make a dinner reservation. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/make-a-dinner-reservation.html"
  },
  {
   "id": "let-claude-buy-a-game-autonomously",
   "title": "Let Claude buy a game autonomously",
   "summary": "Community report: the poster says they assembled payment infrastructure with a few custom MCPs so Claude could shop for a game. In the demo, Claude searched 10,000+ games, located Civilization III Complete for $0.99, authorized payment through x402 with a human confirmation step, settled funds through an MCP server with a digital wallet, and then received the license key. The run took 60 seconds total with 0 clicks, broken out as 10 seconds for search, 5 for authorization, 30 for settlement, and 15 for fulfillment.",
   "agent": "Claude",
   "topic": "Shopping",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1owqgrp/i_just_bought_a_game_in_60_seconds_by_telling",
   "prompt": "Shopping task: let Claude buy a game autonomously. Collect order/policy details, prepare exact steps, and pause before submit or payment.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/let-claude-buy-a-game-autonomously.html"
  },
  {
   "id": "generate-investment-analysis-from-live-market-data",
   "title": "Generate investment analysis from live market data",
   "summary": "As described in the source, the poster used Claude Code CLI to build a personal wealth advisor that pulled live market data, macro indicators and news into one workflow, then fed it to Claude with a CFA-style system prompt. In comments, they say a memory system let it recall and revisit earlier calls from weeks prior, which improved analysis quality; the person assembled the setup and used Claude’s output rather than handing decisions over fully. They say the goal was analysis beyond generic stock tips, but the thread excerpt gives no performance numbers or verified investing results.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1sdtgof/used_claude_code_to_build_myself_a_personal",
   "prompt": "Help me generate investment analysis from live market data. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/generate-investment-analysis-from-live-market-data.html"
  },
  {
   "id": "generate-a-hiring-pack-from-existing-notion-docs",
   "title": "Generate a hiring pack from existing Notion docs",
   "summary": "A PM at Clay gave Claude Code internal hiring Notion pages and peer JDs; it produced a job description, hiring plan, interview plan and rubric for a new role.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Using the hiring materials in this folder and similar external job descriptions, draft a JD, hiring plan, interview plan and scoring rubric for this new role. Stay in plan mode until I approve.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/generate-a-hiring-pack-from-existing-notion-docs.html"
  },
  {
   "id": "find-buyer-intent-leads-and-research-them-for-outreach",
   "title": "Find buyer-intent leads and research them for outreach",
   "summary": "As described in the source, a commenter said they run two bots: the first scans the web with multiple sub-agents for more than 30 buyer-intent signals, such as companies hiring many of the same roles or rumors of an announcement. The second bot reviews those signals, scores each one as a good or bad opportunity, and locates relevant people to contact; the human then uses the resulting weekly lead list for manual email outreach rather than AI-written messages. They said this is working for them because prospects are more receptive when buyer intent is already present and the outreach email is written by a person.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1x22xmr/grok_bot_for_lead_generation",
   "prompt": "GTM task: find buyer-intent leads and research them for outreach. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/find-buyer-intent-leads-and-research-them-for-outreach.html"
  },
  {
   "id": "find-an-nyc-apartment",
   "title": "Find an NYC apartment",
   "summary": "Community report: zapier Agents shows up in a write-up about find an NYC apartment, with figures like $20. The linked post is the primary source for steps and outcomes.",
   "agent": "Zapier Agents",
   "topic": "Automation",
   "source": "X",
   "url": "https://x.com/ckor/status/2051310336980947305",
   "prompt": "Automation design: find an NYC apartment. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/find-an-nyc-apartment.html"
  },
  {
   "id": "design-ui-pages-and-explainer-videos-by-voice",
   "title": "Design UI pages and explainer videos by voice",
   "summary": "Documented elsewhere: user spoke instructions and supplied design references to Grok Bot to build UI/UX pages for an agentic financial desk. The bot generated the pages, stood up images, animated the designs on its own, and produced explainer videos. All requested pages, animations, and videos were completed.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/0x3Matt/status/2108797829430812912",
   "prompt": "Creative task: design UI pages and explainer videos by voice. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/design-ui-pages-and-explainer-videos-by-voice.html"
  },
  {
   "id": "claim-unclaimed-property-using-chatgpt-guidance",
   "title": "Claim unclaimed property using ChatGPT guidance",
   "summary": "As described in the source, in a comment on the removed Reddit thread, someone said they got a mailed notice from a company offering to file an unclaimed-property claim for a 10% fee. They photographed the letter, sent it to ChatGPT, and ChatGPT told them to check Florida’s treasury site and file the claim themselves; they then submitted the claim without powered by the company. They said the claim worked: the fiancé received the money by deposit, it took only a few weeks, and they did not have to jump through any hoops.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/unclaimedproperty/comments/1t41wnj/removed",
   "prompt": "Help me claim unclaimed property using ChatGPT guidance. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/claim-unclaimed-property-using-chatgpt-guidance.html"
  },
  {
   "id": "build-small-custom-agents-such-as-one-in-a-music-player",
   "title": "Build small custom agents such as one in a music player",
   "summary": "Real-world thread: the builder says they write lots of small agents instead of another coding agent, one of them inside their music player. They argue a useful agent's core takes very little code, so these are trivial as code artifacts. The source states no results or numbers.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49698832",
   "prompt": "Creative task: build small custom agents such as one in a music player. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/build-small-custom-agents-such-as-one-in-a-music-player.html"
  },
  {
   "id": "build-a-game-level-from-a-design-document",
   "title": "Build a game level from a design document",
   "summary": "Real-world thread: the poster used Grok 4.5 to build level 2 of an existing Three.js game from a very detailed game design document, powered by their own game-dev harness as the setup around the model. The source only says the level was AI-assembled and then manually cleaned up and polished afterward by the poster. The result was a playable new level for the existing game, but the poster said it still lacked in several areas and they were polishing it.",
   "agent": "Grok Bot",
   "topic": "Media & Content",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/aigamedev/comments/1usy7ui/built_this_game_using_grok_45_threejs_and_my_game",
   "prompt": "Creative task: build a game level from a design document. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/build-a-game-level-from-a-design-document.html"
  },
  {
   "id": "auto-generate-a-searchable-help-video-library-for-an-app",
   "title": "Auto-generate a searchable help video library for an app",
   "summary": "Linked example of auto-generate a searchable help video library for an app. The builder copied this exact prompt into a Replit agent: add ?demo=1 mode with seeded fake data and no DB writes; define each video in code as scenes specifying route, narration script, and scripted cursor movements; have Playwright record the app while TTS narrates; use ffmpeg to add captions and stitch scenes; output MP4 + transcript + ...",
   "agent": "Replit Agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/NIftek/status/2075249459655512265",
   "prompt": "Creative task: auto-generate a searchable help video library for an app. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/auto-generate-a-searchable-help-video-library-for-an-app.html"
  },
  {
   "id": "auto-apply-to-jobs-on-linkedin",
   "title": "Auto-apply to jobs on LinkedIn",
   "summary": "As described in the source, chatGPT Agent shows up in a write-up about auto-apply to jobs on LinkedIn, with figures like 40, 2. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ArtificialInteligence/comments/1s9zio8/does_chatgpt_agent_mode_do_a_good_job_to_auto",
   "prompt": "Notes task: auto-apply to jobs on LinkedIn. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/auto-apply-to-jobs-on-linkedin.html"
  },
  {
   "id": "apply-to-jobs",
   "title": "Apply to jobs",
   "summary": "A published walkthrough shows custom agent shows up in a write-up about apply to jobs. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48753295",
   "prompt": "Notes task: apply to jobs. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/apply-to-jobs.html"
  },
  {
   "id": "analyze-your-chess-games-into-commented-videos",
   "title": "Analyze your chess games into commented videos",
   "summary": "Observed in the wild: the builder recorded themselves thinking aloud during a 15+10 rapid game, then gave Claude Code the Lichess link and the mp3. Claude transcribed it locally with whisper.cpp and matched sentences to moves powered by PGN clock times. It produced an annotated PGN, an HTML board and a 6:49 narrated video with subtitles for The builder's real game.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49857528",
   "prompt": "Creative task: analyze your chess games into commented videos. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/analyze-your-chess-games-into-commented-videos.html"
  },
  {
   "id": "add-61-kid-friendly-mini-games-to-an-app",
   "title": "Add 61 kid-friendly mini-games to an app",
   "summary": "One practitioner notes that users pointed Replit Agent at their existing 40-game mini-games app and told it to reach 101 total games, with one rule only: no repeated games. The agent had full autonomy to choose, build, and integrate the rest. The vault reached 101 games.",
   "agent": "Replit Agent",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/Andrew_Blumson/status/2099243389963194640",
   "prompt": "Notes task: add 61 kid-friendly mini-games to an app. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 43,
   "page": "https://agentsightings.com/case/add-61-kid-friendly-mini-games-to-an-app.html"
  },
  {
   "id": "optifly-opty-airline-schedule-planning-agent",
   "title": "Airline planners ask an agent to explain and change schedules",
   "summary": "Optifly says airlines already use its Opty agent in production to analyse and explain their schedules, make changes and work with its optimiser. From January 2027 schedulers will describe planning challenges in plain words and have agents generate and compare alternative schedules. Vendor announcement.",
   "agent": "Opty (Optifly)",
   "topic": "Travel & Booking",
   "source": "News (Optifly)",
   "url": "https://optifly.com/news/optifly-launches-opty",
   "prompt": "Explain why this airline schedule has weak connections on Tuesdays, suggest changes, and compare the trade-offs between two alternatives.",
   "added": "2026-10-29",
   "interest": 42,
   "page": "https://agentsightings.com/case/optifly-opty-airline-schedule-planning-agent.html"
  },
  {
   "id": "mobilezone-copilot-studio-agents-sales-orders",
   "title": "Agents save 5-10 minutes on every sales order",
   "summary": "Microsoft says mobilezone's Copilot Studio agents save 5-10 minutes per sales order and proactively resolve one to two store issues a day. Vendor-published.",
   "agent": "Copilot Studio agents",
   "topic": "Sales & Marketing",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/27511-mobilezone-microsoft-copilot-studio",
   "prompt": "Handle the paperwork on each new sales order, and each morning check our stores for problems and fix the ones you can.",
   "added": "2026-10-29",
   "interest": 42,
   "page": "https://agentsightings.com/case/mobilezone-copilot-studio-agents-sales-orders.html"
  },
  {
   "id": "lifetime-products-copilot-studio-order-touches",
   "title": "A manufacturer's agent removes 50-80 manual order touches a day",
   "summary": "Microsoft says Lifetime Products used Copilot Studio to eliminate 50-80 daily retail-order touches, saving 2-3 hours a day. Vendor-published; only headline figures are public.",
   "agent": "Copilot Studio agent",
   "topic": "Business & Startups",
   "source": "Case study (Microsoft)",
   "url": "https://www.microsoft.com/en/customers/story/27087-lifetime-products-microsoft-copilot-studio",
   "prompt": "Check each incoming retail order for the details we need, fix what you can, and only send a person the ones that do not match.",
   "added": "2026-10-29",
   "interest": 42,
   "page": "https://agentsightings.com/case/lifetime-products-copilot-studio-order-touches.html"
  },
  {
   "id": "triage-1-847-unread-emails-into-folders",
   "title": "Triage 1,847 unread emails into folders",
   "summary": "Observed in the wild: Custom agent was pointed at triage 1,847 unread emails into folders. from the linked report: within 40 minutes, an AI agent read email topics, organized them into 6 folders, and produced 12 one-line draft replies.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/Sonofpeace0001/status/2092487261401010459",
   "prompt": "Go through my mail for this goal: triage 1,847 unread emails into folders. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 42,
   "page": "https://agentsightings.com/case/triage-1-847-unread-emails-into-folders.html"
  },
  {
   "id": "synthesize-interview-transcripts-and-pdfs-for-research",
   "title": "Synthesize interview transcripts and PDFs for research",
   "summary": "Observed in the wild: transcripts go in one folder, one file per interview with consistent names, plus a CLAUDE.md holding research questions, a codebook and a rule to quote only verbatim text with file names. Claude codes them in batches of 5 to 8, or via parallel subagents, and writes themes.md and evidence tables. The guide says one typed question replaces an hour of Ctrl+F, and Else van der Berg runs 50-transcript studies this way.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://ccforeveryone.com/guides/claude-code-for-researchers",
   "prompt": "Research brief: synthesize interview transcripts and PDFs for research. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 42,
   "page": "https://agentsightings.com/case/synthesize-interview-transcripts-and-pdfs-for-research.html"
  },
  {
   "id": "support-call-summaries",
   "title": "Support call summaries",
   "summary": "A team runs speech recognition plus a small LLM to transcribe and summarize support calls, beating the agents' own notes.",
   "agent": "Custom agent",
   "topic": "Customer Support",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=44405067",
   "prompt": "Summarize this support call transcript: customer issue, steps tried, resolution, follow-ups owed, and customer sentiment.",
   "added": "2026-10-15",
   "interest": 42,
   "page": "https://agentsightings.com/case/support-call-summaries.html"
  },
  {
   "id": "run-a-daily-chief-of-staff-agent-across-inbox-calendar-and-crm",
   "title": "Run a daily chief-of-staff agent across inbox, calendar and CRM",
   "summary": "A published walkthrough shows from the linked report: the surgeon manages roughly twenty Cowork tasks that involve managing email, organizing meetings, identifying dead deals, and preserving sketches in a dashboard.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/claude/comments/1twhsm9/best_everyday_life_use_cases_for_claude_cowork",
   "prompt": "Go through my mail for this goal: run a daily chief-of-staff agent across inbox, calendar and CRM. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 42,
   "page": "https://agentsightings.com/case/run-a-daily-chief-of-staff-agent-across-inbox-calendar-and-crm.html"
  },
  {
   "id": "cleaning-up-194-gb-of-photos",
   "title": "Cleaning up 194 GB of photos",
   "summary": "A user had an LLM help consolidate and de-duplicate a large photo archive, shrinking it to about a third of its size.",
   "agent": "Claude Code",
   "topic": "Personal Files",
   "source": "Blog",
   "url": "https://clement.n8r.ch/en/articles/photo-migration-llm/",
   "prompt": "Working on a copy, find duplicate photos by hash, rename files as YYYY-MM-DD_HHMMSS using EXIF data, and sort them into year folders. Report counts before and after.",
   "added": "2026-10-15",
   "interest": 42,
   "page": "https://agentsightings.com/case/cleaning-up-194-gb-of-photos.html"
  },
  {
   "id": "call-the-dmv-to-schedule-an-appointment",
   "title": "Call the DMV to schedule an appointment",
   "summary": "According to the write-up, olivia Moore established a voice agent who contacted the DMV and booked an appointment in under 30 minutes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/omooretweets/status/1823380097531093232",
   "prompt": "Help me call the DMV to schedule an appointment. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 42,
   "page": "https://agentsightings.com/case/call-the-dmv-to-schedule-an-appointment.html"
  },
  {
   "id": "turn-email-invoices-into-a-live-financial-dashboard",
   "title": "Turn email invoices into a live financial dashboard",
   "summary": "One practitioner notes that the workflow watches Gmail for incoming invoice emails, then uses AI to classify each one as an expense or income and extract fields covering supplier, amount, date, description, and category. It writes the structured data into Google Sheets, where the sheet itself powers a live dashboard showing totals, net profit, and category breakdowns; the Reddit post does not describe a manual review step. The stated result is a real-time finance dashboard tracking total income, total expenses, net profit, and spending by category.",
   "agent": "n8n",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1o27rk5/i_built_an_ai_accountant_in_n8n_that_turns_email",
   "prompt": "Go through my mail for this goal: turn email invoices into a live financial dashboard. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/turn-email-invoices-into-a-live-financial-dashboard.html"
  },
  {
   "id": "turn-notion-prds-into-scoped-linear-tickets",
   "title": "Turn Notion PRDs into scoped Linear tickets",
   "summary": "According to the write-up, this gallery page only lists short blurbs, not walkthroughs. For the PRD entry, Devin is pointed at a Notion PRD and creates scoped Linear tickets, one per work package. The page states no results, numbers, or limitations.",
   "agent": "Devin",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://docs.devin.ai/use-cases/gallery",
   "prompt": "Notes task: turn Notion PRDs into scoped Linear tickets. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/turn-notion-prds-into-scoped-linear-tickets.html"
  },
  {
   "id": "test-a-browser-agent-on-web-games",
   "title": "Test a browser agent on web games",
   "summary": "As described in the source, on each game's page in the ChatGPT Atlas browser, the researchers turned on Agent Mode (Preview) and gave one zero-shot instruction: play until you get stuck. They ran 10 trials per game and did not intervene. Sudoku took an average of 2m28s with 100% accuracy, against a human baseline of 10-12 minutes.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://arxiv.org/html/2510.26298v1",
   "prompt": "Help me test a browser agent on web games. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/test-a-browser-agent-on-web-games.html"
  },
  {
   "id": "run-a-private-personal-admin-agent-in-cursor",
   "title": "Run a private personal-admin agent in Cursor",
   "summary": "Observed in the wild: the builder runs brainstorming and personal admin in the same Cursor interface but switches constraints on purpose. For personal work he uses a private notes area and rules that stop personal details from being copied into public or commercial sketches. The source gives no measured results.",
   "agent": "Cursor",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://petralian.com/posts/cursor-customize-brainstorm-and-personal-agent-2026",
   "prompt": "Go through my mail for this goal: run a private personal-admin agent in Cursor. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/run-a-private-personal-admin-agent-in-cursor.html"
  },
  {
   "id": "run-a-live-kalshi-prediction-market-account-autonomously",
   "title": "Run a live Kalshi prediction-market account autonomously",
   "summary": "From the linked report, mirabilis, an autonomous AI agent powered by Claude Code, was given access to a live $20 Kalshi prediction-market account. It traded on its own for nine days, participating in the Liquidity Incentive Program while a human provided supervision throughout. The agent collected $20.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/NewsTongueX/status/2109411919748575518",
   "prompt": "Help me run a live Kalshi prediction-market account autonomously. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/run-a-live-kalshi-prediction-market-account-autonomously.html"
  },
  {
   "id": "review-meta-ad-performance-inside-ads-manager",
   "title": "Review Meta ad performance inside Ads Manager",
   "summary": "A published walkthrough shows skaleit gave Manus, the agent assembled into Meta Ads Manager, access to a live client ecommerce account whose performance had dropped since January 2026. They asked it to compare the last 30 days with January, explain why ROAS fell, and propose a recovery plan. Manus described January ROAS of 14.",
   "agent": "Manus",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://skaleit.agency/blog/manus-ai-meta-ads-honest-review/",
   "prompt": "Help me review Meta ad performance inside Ads Manager. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/review-meta-ad-performance-inside-ads-manager.html"
  },
  {
   "id": "research-the-web-visually-by-browsing-pages",
   "title": "Research the web visually by browsing pages",
   "summary": "From the linked report, the commenter says they used ChatGPT Agent Mode as a visual browser rather than a text crawler. It spun up a virtual computer, opened sites in a browser, looked at the rendered pages the way a person would, and clicked through live buttons and interfaces; they used it to search many online stores for a specific item and check whether each would ship to their country. They say this worked much better on dynamic interfaces than text-based crawling and let the agent visit hundreds of stores to find what they considered the best option.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1usngow/is_openai_silently_removing_agent_mode",
   "prompt": "Research brief: research the web visually by browsing pages. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/research-the-web-visually-by-browsing-pages.html"
  },
  {
   "id": "pull-wallet-data-from-dashboards-into-a-spreadsheet",
   "title": "Pull wallet data from dashboards into a spreadsheet",
   "summary": "A published walkthrough shows claude computer use was enabled with no custom setup. One prompt asked it to pull wallet data from dashboards, check them, and clean a summary in a spreadsheet. Task finished in 19 minutes with visible screenshots and actions for oversight; cost stayed tiny even while exploring.",
   "agent": "Claude computer use",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Rifat_EE/status/2068626729875181660",
   "prompt": "Help me pull wallet data from dashboards into a spreadsheet. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/pull-wallet-data-from-dashboards-into-a-spreadsheet.html"
  },
  {
   "id": "pull-invoices-from-gmail-and-organize-them",
   "title": "Pull invoices from Gmail and organize them",
   "summary": "According to the write-up, the poster assembled Invoice Agent for teams that still handle vendor invoices through email. It connects to Gmail or Outlook, detects incoming invoice emails, uses AI to extract the key details, and puts them into a searchable place; at the time of the post, the human’s role was still validating the workflow, pricing, and positioning through feedback. The post gives no usage or accuracy numbers yet; it presents the product as a side project in validation.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SideProject/comments/1sdncbk/built_a_side_project_that_pulls_invoices_from",
   "prompt": "Go through my mail for this goal: pull invoices from Gmail and organize them. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/pull-invoices-from-gmail-and-organize-them.html"
  },
  {
   "id": "port-a-chrome-extension-to-safari",
   "title": "Port a Chrome extension to Safari",
   "summary": "Observed in the wild: the builder already had Session Buddy installed in Chrome, so they told ChatGPT it could pull the source code from there. They then asked it to port the extension to Safari and install it. After about 20 minutes, Session Buddy was running in Safari.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50078690",
   "prompt": "Help me port a Chrome extension to Safari. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/port-a-chrome-extension-to-safari.html"
  },
  {
   "id": "plan-vacations-handle-visas-and-make-bookings",
   "title": "Plan vacations, handle visas and make bookings",
   "summary": "According to the write-up, the poster ran AI that filtered hundreds of dating outreach messages, letting dozens of live chats reach direct contact after chosen ones passed the filters. Separately the AI planned a vacation, handled the visa, and completed all bookings. Someone states it is going great and loves the setup, with no other results or shortfalls described.",
   "agent": "Make",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/nicebit_org/status/2106049786742768046",
   "prompt": "GTM task: plan vacations, handle visas and make bookings. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/plan-vacations-handle-visas-and-make-bookings.html"
  },
  {
   "id": "generate-parametric-cad-models-from-a-prompt",
   "title": "Generate parametric CAD models from a prompt",
   "summary": "Case study takeaway: you describe a part in plain English, optionally with reference images. The app uses Claude to write OpenSCAD code, which runs in the browser through WebAssembly and shows a live Three.js preview. The README's benchmark gallery shows single prompts producing parametric models, from knurled knobs and threaded bolts to a V8 engine with 22 dimension controls.",
   "agent": "Claude",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48572553",
   "prompt": "Creative task: generate parametric CAD models from a prompt. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/generate-parametric-cad-models-from-a-prompt.html"
  },
  {
   "id": "find-unclaimed-money-across-old-addresses",
   "title": "Find unclaimed money across old addresses",
   "summary": "As described in the source, the agent received Someone's list of old addresses. It searched unclaimed-money databases across them, locating an extra $500 plus US amounts from prior residence there. The agent described the $500 discovery and finished filling forms plus ID prep.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/S_Padival/status/2109389004311560338",
   "prompt": "Help me find unclaimed money across old addresses. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/find-unclaimed-money-across-old-addresses.html"
  },
  {
   "id": "find-outperforming-youtube-videos-and-newsletter-advertisers",
   "title": "Find outperforming YouTube videos and newsletter advertisers",
   "summary": "Observed in the wild: the builder ran two research tasks in Manus at the same time. One had it browse YouTube for AI-tool videos from the last month that beat their channel size on views. Manus returned a trends summary and a spreadsheet of advertisers with ad copy and contact details.",
   "agent": "Manus",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.nocode.mba/articles/manus-ai-agents-automation",
   "prompt": "Go through my mail for this goal: find outperforming YouTube videos and newsletter advertisers. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/find-outperforming-youtube-videos-and-newsletter-advertisers.html"
  },
  {
   "id": "fill-out-pdf-forms-like-a-w-9-from-saved-personal-details",
   "title": "Fill out PDF forms like a W-9 from saved personal details",
   "summary": "One practitioner notes that dave Swift put a blank W-9 in a folder and had Claude Code list the form fields. He then gave it his details, and it wrote a short script that filled the fields and saved a new PDF. Three W-9s came out in about thirty seconds in a batch run.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/jerryjliu0/status/2008954366762602527",
   "prompt": "Notes task: fill out PDF forms like a W-9 from saved personal details. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/fill-out-pdf-forms-like-a-w-9-from-saved-personal-details.html"
  },
  {
   "id": "draft-email-campaigns-in-klaviyo",
   "title": "Draft email campaigns in Klaviyo",
   "summary": "One practitioner notes that the Grok Bot received brand voice instructions plus access to Dropbox product photos and Figma. It wrote full campaigns in that voice, pulled the photos, stood up hero artwork, and inserted an editable draft directly into Klaviyo. The setup replaced Someone's prior email agency.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/willroman/status/2109012028526403619",
   "prompt": "Go through my mail for this goal: draft email campaigns in Klaviyo. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/draft-email-campaigns-in-klaviyo.html"
  },
  {
   "id": "draft-a-dissertation-lit-review-from-uploaded-sources",
   "title": "Draft a dissertation lit review from uploaded sources",
   "summary": "Community report: after ChatGPT invented citations, the writer switched to NotebookLM so the model could cite only papers they had uploaded. They renamed files before uploading and gave one standing instruction for citation format, which changed NotebookLM’s bare numbered references into The builder-year citations; they then used a broader workflow to vet sources, organize them, get synthesis, and manually audit citations before export. It cut down a lot of manual cross-checking and helped them draft dissertation literature review chapters without made-up references.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/WritingWithAI/comments/1uzecdt/i_managed_to_use_ai_safely_to_help_me_write_my",
   "prompt": "Help me draft a dissertation lit review from uploaded sources. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/draft-a-dissertation-lit-review-from-uploaded-sources.html"
  },
  {
   "id": "collect-shower-glass-supplier-quotes-by-email",
   "title": "Collect shower glass supplier quotes by email",
   "summary": "A published walkthrough shows someone asked OpenClaw to research local shower glass suppliers afterward contact them. The agent sent emails on its own, gathered replies, and managed the back-and-forth conversation inside a single chat thread. It produced supplier quotes through email.",
   "agent": "OpenClaw",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/freeatnet/status/2105403769437516055",
   "prompt": "Go through my mail for this goal: collect shower glass supplier quotes by email. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/collect-shower-glass-supplier-quotes-by-email.html"
  },
  {
   "id": "code-from-your-phone",
   "title": "Code from your phone",
   "summary": "A published walkthrough shows the builder runs Claude Code on a Vultr cloud VM reachable only through Tailscale, powered by Termius with mosh from an iPhone. Each git worktree gets its own tmux window and agent. He runs six agents in parallel on separate features from his phone, handling tasks of 10-20 minutes between notifications.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46491486",
   "prompt": "Help me code from your phone. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/code-from-your-phone.html"
  },
  {
   "id": "clean-inbox-and-draft-social-posts-in-your-voice",
   "title": "Clean inbox and draft social posts in your voice",
   "summary": "According to the write-up, the Grok bot was given access to Someone's computer and apps. It logged into sites, clicked through them independently, cleaned and analyzed the inbox, afterward sketched X and LinkedIn posts in Someone's voice. The bots managed and improved daily ops with ongoing coordination.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/galileowilson/status/2090498508373799112",
   "prompt": "Go through my mail for this goal: clean inbox and draft social posts in your voice. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/clean-inbox-and-draft-social-posts-in-your-voice.html"
  },
  {
   "id": "chase-unpaid-invoices-and-paperwork-by-email-for-small-firms",
   "title": "Chase unpaid invoices and paperwork by email for small firms",
   "summary": "Linked example of chase unpaid invoices and paperwork by email for small firms. The poster describes an email-based collections and paperwork agent for small firms that keeps watch on unpaid invoices and missing admin, keeps chasing by email until it gets a yes, a no, or a payment, and hands anything needing judgment back to the owner.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1vgeo9g/built_an_agent_that_chases_invoices_and_paperwork",
   "prompt": "Go through my mail for this goal: chase unpaid invoices and paperwork by email for small firms. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/chase-unpaid-invoices-and-paperwork-by-email-for-small-firms.html"
  },
  {
   "id": "build-and-launch-a-code-snippet-sharing-app",
   "title": "Build and launch a code-snippet sharing app",
   "summary": "Case study takeaway: the poster says they assembled the Snippra app itself with Replit Agent while keeping the main website on WordPress. They describe feeding the agent a UI reference image from Behance via Figma, asking it to clone the design, then powered by it for UI copy, dummy snippets, feature planning, debugging, naming, and idea refinement while they steered prompts and tweaked the result. They launched Snippra live with 40+ languages, custom categories, shareable links, and a Pro snippet-to-social image feature.",
   "agent": "Replit Agent",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1kqzkhx/just_published_my_first_replit_agent_app",
   "prompt": "Notes task: build and launch a code-snippet sharing app. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/build-and-launch-a-code-snippet-sharing-app.html"
  },
  {
   "id": "build-an-invitation-card-website-purely-by-prompting",
   "title": "Build an invitation card website purely by prompting",
   "summary": "Case study takeaway: fabio Zapata gave Cursor a first prompt describing the page's goals, look and technical approach, and let it run console commands. The envelope animation went badly for 15–20 minutes of vague feedback. The site took 2–3 hours, against his estimate of 4–8 by hand.",
   "agent": "Cursor",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://dev.to/fazd/my-first-experience-with-cursor-4a5g",
   "prompt": "Help me build an invitation card website purely by prompting. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/build-an-invitation-card-website-purely-by-prompting.html"
  },
  {
   "id": "build-a-benchmark-image-collage-in-canva",
   "title": "Build a benchmark image collage in Canva",
   "summary": "Community report: the builder asked ChatGPT Agent to pull the benchmark images from OpenAI's announcement post and arrange them in Canva, with red rectangles around the Agent's columns. The builder solved a CAPTCHA and logged into Canva by hand, then redirected the agent when it tried to draw the charts itself. It took over 75 minutes and the collage was rough.",
   "agent": "ChatGPT Agent",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://www.datacamp.com/tutorial/chatgpt-agent",
   "prompt": "Creative task: build a benchmark image collage in Canva. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/build-a-benchmark-image-collage-in-canva.html"
  },
  {
   "id": "book-appointments-and-send-emails-and-messages",
   "title": "Book appointments and send emails and messages",
   "summary": "One practitioner notes that the poster said they used ChatGPT agent for two days to book appointments, send emails, and send messages to their boss. In the comments, they explained one concrete workflow: they had it find tutor contact details and write 10 personalized outreach emails in one prompt, and when it needed mail access it paused to ask them to sign into their email before continuing. They said the agent actually sent the emails and felt dramatically more useful than expected.",
   "agent": "ChatGPT Agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m49e7d",
   "prompt": "Go through my mail for this goal: book appointments and send emails and messages. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/book-appointments-and-send-emails-and-messages.html"
  },
  {
   "id": "audit-and-fix-a-home-assistant-setup",
   "title": "Audit and fix a Home Assistant setup",
   "summary": "Real-world thread: the writer ran a local MCP proxy on a Mac, authenticated with a Long-Lived Access Token, and linked it to Claude Desktop. Because the Home Assistant MCP can't write config, Claude drove the Chrome connector to paste YAML into the editors, and used JavaScript injection when typing garbled the text. The audit located triplicate TV entities, ghost duplicates, unassigned lights, and a fan registered as a plain switch.",
   "agent": "Claude",
   "topic": "Personal Assistant",
   "source": "Blog",
   "url": "https://www.androidauthority.com/using-claude-fix-home-assistant-smart-home-3674998/",
   "prompt": "Assistant mode: audit and fix a Home Assistant setup. Log every action, prefer reversible steps, and ask before emailing, paying, or deleting.",
   "added": "2026-10-15",
   "interest": 41,
   "page": "https://agentsightings.com/case/audit-and-fix-a-home-assistant-setup.html"
  },
  {
   "id": "stoel-rives-legora-firmwide-after-pilot",
   "title": "A US law firm adopts Legora firmwide after a 100-lawyer pilot",
   "summary": "Stoel Rives announced it is deploying the legal AI platform Legora across the whole firm, following a pilot with more than 100 lawyers.",
   "agent": "Legora",
   "topic": "Research & Data",
   "source": "News (Stoel Rives)",
   "url": "https://www.stoel.com/insights/pressreleases/stoel-rives-deepens-its-ai-strategy-by-deploying-legora-firm-wide",
   "prompt": "Review this document set for the issues listed in my brief, pull the relevant passages with page references, and draft a short summary memo for a partner to check.",
   "added": "2026-10-29",
   "interest": 40,
   "page": "https://agentsightings.com/case/stoel-rives-legora-firmwide-after-pilot.html"
  },
  {
   "id": "use-a-coding-agent-for-everyday-non-programming-chores",
   "title": "Use a coding agent for everyday non-programming chores",
   "summary": "Same Claude Code window, zero new code required: rotate/copy photos, classify exam PDF questions, SIGKILL a stuck process, price outdoor steel, warranty lookup, image merge, rename a string in six paths.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/peta2la/status/2107834753139331557",
   "prompt": "Help with this non-coding chore on my machine: explain what you will touch, do the reversible steps, and ask before deleting files or running destructive commands.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/use-a-coding-agent-for-everyday-non-programming-chores.html"
  },
  {
   "id": "transcript-cleanup-to-customer-style-rules",
   "title": "Transcript cleanup to customer style rules",
   "summary": "A transcription business uses LLM agents to proofread human-edited transcripts and apply each customer's formatting rules.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=42431361",
   "prompt": "Proofread this transcript and apply these style rules: <rules>. Don't change wording beyond fixing errors; list every change you made.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/transcript-cleanup-to-customer-style-rules.html"
  },
  {
   "id": "text-every-world-cup-final-score-to-a-friends-x27-signal-group",
   "title": "Text every World Cup final score to a friends' Signal group",
   "summary": "Real-world thread: claude Code shows up in a write-up about text every World Cup final score to a friends' Signal group, with figures like 30, 2026, 3. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "GitHub",
   "url": "https://github.com/bronsonelliott/worldcup-scores-bot-public",
   "prompt": "Help me text every World Cup final score to a friends' Signal group. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/text-every-world-cup-final-score-to-a-friends-x27-signal-group.html"
  },
  {
   "id": "teach-students-with-assessments-custom-lessons-and-reassessment",
   "title": "Teach students with assessments, custom lessons and reassessment",
   "summary": "According to the write-up, custom agent shows up in a write-up about teach students with assessments, custom lessons and reassessment. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/HughSballz/status/2105020037736353861",
   "prompt": "Help me teach students with assessments, custom lessons and reassessment. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/teach-students-with-assessments-custom-lessons-and-reassessment.html"
  },
  {
   "id": "spec-features",
   "title": "Spec features",
   "summary": "Case study takeaway: chatGPT shows up in a write-up about spec features. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47033624",
   "prompt": "Help me spec features. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/spec-features.html"
  },
  {
   "id": "sell-your-unused-items-online",
   "title": "Sell your unused items online",
   "summary": "A published walkthrough shows chatGPT Agent shows up in a write-up about sell your unused items online, with figures like 10. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPTPromptGenius/comments/1pt74uj/chatgpt_agent_mode_can_sell_your_stuff_online_in",
   "prompt": "Help me sell your unused items online. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/sell-your-unused-items-online.html"
  },
  {
   "id": "search-facebook-marketplace-and-tabulate-results-in-sheets",
   "title": "Search Facebook Marketplace and tabulate results in Sheets",
   "summary": "Linked example of search Facebook Marketplace and tabulate results in Sheets. The poster tried to use the Claude Code Chrome extension to search Facebook Marketplace listings and enter the results into a Google Sheet.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1qelajn/claude_for_chrome_constantly_asking_for_permissions",
   "prompt": "Help me search Facebook Marketplace and tabulate results in Sheets. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/search-facebook-marketplace-and-tabulate-results-in-sheets.html"
  },
  {
   "id": "run-web-searches-and-browser-tasks",
   "title": "Run web searches and browser tasks",
   "summary": "Documented elsewhere: Operator was pointed at run web searches and browser tasks. an early-access user tested a browser and browsed the web, and described that the operator's web tasks were handled efficiently.",
   "agent": "Operator",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPTPro/comments/1i8jln3/i_am_among_the_first_people_to_gain_access_to",
   "prompt": "Help me run web searches and browser tasks. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-web-searches-and-browser-tasks.html"
  },
  {
   "id": "run-scheduled-specialist-bots",
   "title": "Run scheduled specialist bots",
   "summary": "According to the write-up, grok Bot shows up in a write-up about run scheduled specialist bots, with figures like 9, 00. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://sharbel.com/articles/grok-bot-review",
   "prompt": "Help me run scheduled specialist bots. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-scheduled-specialist-bots.html"
  },
  {
   "id": "run-overnight-memecoin-checks-and-alerts",
   "title": "Run overnight memecoin checks and alerts",
   "summary": "Linked example of run overnight memecoin checks and alerts. User recorded their 4-minute memecoin routine once (dexscreener, gmgn top traders tab, rugcheck, bubblemap, X pre-call buyers) and had the Grok Bot replicate it on its cloud computer.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Hrundel75/status/2108901142264487999",
   "prompt": "Help me run overnight memecoin checks and alerts. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-overnight-memecoin-checks-and-alerts.html"
  },
  {
   "id": "run-multi-step-tasks",
   "title": "Run multi-step tasks",
   "summary": "According to the write-up, gemini shows up in a write-up about run multi-step tasks, with figures like 7, 3, 4.6. The linked post is the primary source for steps and outcomes.",
   "agent": "Gemini",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/dani_avila7/status/2026702478801252697",
   "prompt": "Help me run multi-step tasks. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-multi-step-tasks.html"
  },
  {
   "id": "run-dental-office-queries-scheduling-and-reminders",
   "title": "Run dental office queries, scheduling and reminders",
   "summary": "Case study takeaway: custom agent shows up in a write-up about run dental office queries, scheduling and reminders. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1oucqcv/ai_automation_for_dental_office",
   "prompt": "Help me run dental office queries, scheduling and reminders. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-dental-office-queries-scheduling-and-reminders.html"
  },
  {
   "id": "run-an-online-store-x27-s-inventory-listings-and-photos",
   "title": "Run an online store's inventory, listings and photos",
   "summary": "A published walkthrough shows custom agent shows up in a write-up about run an online store's inventory, listings and photos. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/0xVikk/status/2107878376589447516",
   "prompt": "Help me run an online store's inventory, listings and photos. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-an-online-store-x27-s-inventory-listings-and-photos.html"
  },
  {
   "id": "run-an-11-agent-team-for-trading-prep",
   "title": "Run an 11-agent team for trading prep",
   "summary": "Linked example of run an 11-agent team for trading prep. The trader moved a daily swing-trading prep workflow onto Grok Bot and organized it as 11 separate agents with an org-chart structure, a shared memory layer, and an agentic verification step.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/grok/comments/1wz4f7p/running_an_11agent_team_on_grok_bot_for_my",
   "prompt": "Help me run an 11-agent team for trading prep. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-an-11-agent-team-for-trading-prep.html"
  },
  {
   "id": "run-a-commercial-real-estate-business",
   "title": "Run a commercial real estate business",
   "summary": "Observed in the wild: custom agent shows up in a write-up about run a commercial real estate business. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/DMAC_19/status/2105719791751487542",
   "prompt": "Help me run a commercial real estate business. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-a-commercial-real-estate-business.html"
  },
  {
   "id": "run-a-client-x27-s-facebook-ad-account",
   "title": "Run a client's Facebook ad account",
   "summary": "A published walkthrough shows custom agent shows up in a write-up about run a client's Facebook ad account. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/MattCMilne/status/2107186563977531742",
   "prompt": "Help me run a client's Facebook ad account. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/run-a-client-x27-s-facebook-ad-account.html"
  },
  {
   "id": "reverse-engineer-a-1997-pc-game-and-make-it-run-on-modern-hardware",
   "title": "Reverse-engineer a 1997 PC game and make it run on modern hardware",
   "summary": "Documented elsewhere: claude Code shows up in a write-up about reverse-engineer a 1997 PC game and make it run on modern hardware. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50080674",
   "prompt": "Help me reverse-engineer a 1997 PC game and make it run on modern hardware. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/reverse-engineer-a-1997-pc-game-and-make-it-run-on-modern-hardware.html"
  },
  {
   "id": "recompile-a-1991-assembly-pc-game",
   "title": "Recompile a 1991 assembly PC game",
   "summary": "One practitioner notes that claude Code shows up in a write-up about recompile a 1991 assembly PC game, with figures like 1991, 386. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47689685",
   "prompt": "Help me recompile a 1991 assembly PC game. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/recompile-a-1991-assembly-pc-game.html"
  },
  {
   "id": "qualify-scraped-leads-and-write-personalized-openers",
   "title": "Qualify scraped leads and write personalized openers",
   "summary": "Linked example of qualify scraped leads and write personalized openers. In the thread, the concrete use described is feeding Claude Code a prospect list plus recent company news so it can automate research-heavy prospecting.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/coldemail/comments/1r05w98/who_here_is_using_claude_code_for_outbound_how",
   "prompt": "Help me qualify scraped leads and write personalized openers. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/qualify-scraped-leads-and-write-personalized-openers.html"
  },
  {
   "id": "process-a-sales-order-from-verification-to-client-email-draft",
   "title": "Process a sales order from verification to client email draft",
   "summary": "Sales desk flow on Grok Bot: verify the commercial paperwork trio (proforma, PO, bank proof), patch the mis-typed deposit day, spit out commission + packing artifacts, and stash a client-facing note in Gmail’s draft folder.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/rassailor/status/2108221352641741264",
   "prompt": "Process this sales order end to end: verify PI against PO and payment proof, fix data issues you find, prepare commission and packing artifacts, and draft (do not send) the client email.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/process-a-sales-order-from-verification-to-client-email-draft.html"
  },
  {
   "id": "move-openclaw-skills-to-grok-bot",
   "title": "Move OpenClaw skills to Grok Bot",
   "summary": "Linked example of move OpenClaw skills to Grok Bot. The commenter said they replaced OpenClaw and other automations with Grok Bot for admin and personal-assistant work, not mainly development.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1vvx2fg/is_grok_bot_worth_it",
   "prompt": "Help me move OpenClaw skills to Grok Bot. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/move-openclaw-skills-to-grok-bot.html"
  },
  {
   "id": "monitor-tech-subreddits-and-competitors-draft-emails-and-articles",
   "title": "Monitor tech subreddits and competitors, draft emails and articles",
   "summary": "A published walkthrough shows claude Cowork shows up in a write-up about monitor tech subreddits and competitors, draft emails and articles. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rn9ojd/what_are_your_best_cowork_examples_use_cases",
   "prompt": "Help me monitor tech subreddits and competitors, draft emails and articles. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/monitor-tech-subreddits-and-competitors-draft-emails-and-articles.html"
  },
  {
   "id": "monitor-socials-and-flag-comments-dms-and-mentions-worth-answering",
   "title": "Monitor socials and flag comments, DMs and mentions worth answering",
   "summary": "As described in the source, custom agent shows up in a write-up about monitor socials and flag comments, DMs and mentions worth answering. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/vibegoat",
   "prompt": "Help me monitor socials and flag comments, DMs and mentions worth answering. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/monitor-socials-and-flag-comments-dms-and-mentions-worth-answering.html"
  },
  {
   "id": "monitor-public-documents-for-changes-affecting-teaching-materials",
   "title": "Monitor public documents for changes affecting teaching materials",
   "summary": "According to the write-up, dots shows up in a write-up about monitor public documents for changes affecting teaching materials. The linked post is the primary source for steps and outcomes.",
   "agent": "Dots",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://note.com/kazu_t/n/n7eb14540ba04?hl=en",
   "prompt": "Help me monitor public documents for changes affecting teaching materials. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/monitor-public-documents-for-changes-affecting-teaching-materials.html"
  },
  {
   "id": "monitor-resy-and-book-date-night-tables",
   "title": "Monitor Resy and book date night tables",
   "summary": "From the linked report, brian spun up a Hermes subagent inside his personal agent Hannah (which used Obsidian as the brain and Telegram as the interface) to monitor Resy for openings at a few favorite NYC restaurants and book tables for date nights. The subagent ran checks on its own with no rate limits configured. The subagent secured openings effectively until Resy suspended the account for hammering their system and violating TOS.",
   "agent": "Hermes Agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/bdistel/status/2089421994450170039",
   "prompt": "Help me monitor Resy and book date night tables. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/monitor-resy-and-book-date-night-tables.html"
  },
  {
   "id": "migrate-a-legacy-product-stack",
   "title": "Migrate a legacy product stack",
   "summary": "A published walkthrough shows claude Code shows up in a write-up about migrate a legacy product stack, with figures like 4, 2,, 15. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://dev.to/aws-builders/the-setup-is-the-strategy-how-i-orchestrated-a-product-migration-with-claude-code-b92",
   "prompt": "Help me migrate a legacy product stack. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/migrate-a-legacy-product-stack.html"
  },
  {
   "id": "manage-files-in-a-chosen-folder",
   "title": "Manage files in a chosen folder",
   "summary": "One practitioner notes that as described in the source: constructed an open-source desktop agent that functions within a designated folder, allowing users to read and write files and request changes before modifying them.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCowork/comments/1wuqa6c/i_built_an_open-source_cowork-style_folder_agent",
   "prompt": "Help me manage files in a chosen folder. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/manage-files-in-a-chosen-folder.html"
  },
  {
   "id": "manage-10-social-media-accounts",
   "title": "Manage 10 social media accounts",
   "summary": "Case study takeaway: openClaw shows up in a write-up about manage 10 social media accounts, with figures like 10. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://buildtolaunch.substack.com/p/openclaw-setup-guide",
   "prompt": "Help me manage 10 social media accounts. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/manage-10-social-media-accounts.html"
  },
  {
   "id": "jira-issues-from-slack-threads-at-rippling",
   "title": "Jira issues from Slack threads at Rippling",
   "summary": "Besides the dashboard cron, Rippling engineers trigger Cursor automations that turn Slack threads into Jira issues and Confluence summaries.",
   "agent": "Cursor",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "This Slack thread is a work request. Create a Jira issue with summary, acceptance criteria and links back to the thread. If discussion is long, also draft a Confluence summary page.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/jira-issues-from-slack-threads-at-rippling.html"
  },
  {
   "id": "give-an-agent-a-phone-number-to-make-and-answer-calls",
   "title": "Give an agent a phone number to make and answer calls",
   "summary": "A published walkthrough shows david Ondrej gave his Hermes Agent a phone number. The agent afterward put together calls, answered calls, and ran tasks on its own as he slept. The agent performed the calls and tasks without further input from him.",
   "agent": "Hermes Agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/DavidOndrej1/status/2071249713588768861",
   "prompt": "Help me give an agent a phone number to make and answer calls. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/give-an-agent-a-phone-number-to-make-and-answer-calls.html"
  },
  {
   "id": "find-target-businesses",
   "title": "Find target businesses",
   "summary": "Documented elsewhere: manus shows up in a write-up about find target businesses. The linked post is the primary source for steps and outcomes.",
   "agent": "Manus",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://dicloak.com/blog-detail/testing-manus-ai-targeting-10m-businesses--real-results-exposed",
   "prompt": "Help me find target businesses. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/find-target-businesses.html"
  },
  {
   "id": "find-customer-leads-on-reddit-and-twitter-automatically",
   "title": "Find customer leads on Reddit and Twitter automatically",
   "summary": "Linked example of find customer leads on Reddit and Twitter automatically. The poster describes a tool where you enter your website plus product keywords, then the agent scans Reddit and Twitter for matching mentions and conversations.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/newproducts/comments/1g5pnmu",
   "prompt": "Help me find customer leads on Reddit and Twitter automatically. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/find-customer-leads-on-reddit-and-twitter-automatically.html"
  },
  {
   "id": "find-buying-signals-on-reddit",
   "title": "Find buying signals on Reddit",
   "summary": "According to the write-up, grok Bot covered find buying signals on Reddit. mentions a Grok Bot that located buying signals on Reddit.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1vxjipg/grok_bot_review",
   "prompt": "Help me find buying signals on Reddit. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/find-buying-signals-on-reddit.html"
  },
  {
   "id": "file-a-trademark-application",
   "title": "File a trademark application",
   "summary": "Paperwork for a USPTO trademark filing was filled with Claude Cowork as the guide; the application later came back approved.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/brannonmc/status/2105282148857155882",
   "prompt": "Help me file a trademark application. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/file-a-trademark-application.html"
  },
  {
   "id": "edit-existing-video-clips",
   "title": "Edit existing video clips",
   "summary": "Community report: gemini shows up in a write-up about edit existing video clips, with figures like 36. The linked post is the primary source for steps and outcomes.",
   "agent": "Gemini",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://alici.ai/blog/gemini-omni-complete-guide",
   "prompt": "Creative task: edit existing video clips. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/edit-existing-video-clips.html"
  },
  {
   "id": "edit-and-migrate-substack-posts",
   "title": "Edit and migrate Substack posts",
   "summary": "Case study takeaway: claude shows up in a write-up about edit and migrate Substack posts. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://priscasolutionsai.substack.com/p/i-walked-away-while-claude-edited",
   "prompt": "Help me edit and migrate Substack posts. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/edit-and-migrate-substack-posts.html"
  },
  {
   "id": "draft-personalized-job-outreach-emails-from-linkedin-leads",
   "title": "Draft personalized job outreach emails from LinkedIn leads",
   "summary": "A published walkthrough shows pratyush uploaded an Excel of LinkedIn leads plus his resume to Claude Cowork. The agent drafted one personalized email per role, matching the JD to the resume, and placed each draft into Gmail ready to send. Job outreach that had taken hours was completed in minutes.",
   "agent": "Claude Cowork",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/Pratyush2241/status/2106348278975705117",
   "prompt": "GTM task: draft personalized job outreach emails from LinkedIn leads. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/draft-personalized-job-outreach-emails-from-linkedin-leads.html"
  },
  {
   "id": "draft-reddit-answers-to-reach-a-karma-goal",
   "title": "Draft Reddit answers to reach a karma goal",
   "summary": "Linked example of draft Reddit answers to reach a karma goal. The poster says Claude, running in Claude Code for a small company, was tasked with answering Reddit questions and tracking toward a 200-karma goal.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1x315wb/ai_agent_running_in_claude_code_day_15_i_hit_my",
   "prompt": "Help me draft Reddit answers to reach a karma goal. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/draft-reddit-answers-to-reach-a-karma-goal.html"
  },
  {
   "id": "discover-room-temperature-magnetic-semiconductor-candidates",
   "title": "Discover room-temperature magnetic semiconductor candidates",
   "summary": "According to the write-up, the post's headline stated that two candidate room-temperature magnetic semiconductors were identified by powered by oxide 5.5 agents.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49970667",
   "prompt": "Help me discover room-temperature magnetic semiconductor candidates. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/discover-room-temperature-magnetic-semiconductor-candidates.html"
  },
  {
   "id": "debloat-an-android-tv",
   "title": "Debloat an Android TV",
   "summary": "Real-world thread: as described in the source: the task of unplugging his Android TV was accomplished by Rakesh Prakash when he requested assistance from his Hermes agent.",
   "agent": "Hermes Agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/rakeshprakash09/status/2109827195472642538",
   "prompt": "Help me debloat an Android TV. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/debloat-an-android-tv.html"
  },
  {
   "id": "cover-support-calls-24-7",
   "title": "Cover support calls 24/7",
   "summary": "Case study takeaway: Custom agent was pointed at cover support calls 24/7. robin Delta designed a voice agent who is available to answer support calls 24/7, escalate tier-1 issues, and book appointments.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/heyrobinai/status/2070150130423230953",
   "prompt": "Help me cover support calls 24/7. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/cover-support-calls-24-7.html"
  },
  {
   "id": "control-your-mac-by-voice",
   "title": "Control your Mac by voice",
   "summary": "Case study takeaway: claude shows up in a write-up about control your Mac by voice. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50049588",
   "prompt": "Help me control your Mac by voice. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/control-your-mac-by-voice.html"
  },
  {
   "id": "collect-monthly-invoices-and-email-them-to-an-accountant",
   "title": "Collect monthly invoices and email them to an accountant",
   "summary": "Custom agent covered collect monthly invoices and email them to an accountant. The poster made a Chrome extension called Severino for recurring vendor bills like AWS, Vercel, GitHub, Google Cloud, Twilio, and Slack.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AccountingDepartment/comments/1vm8o9g/i_got_tired_of_manually_downloading_invoices_for",
   "prompt": "Go through my mail for this goal: collect monthly invoices and email them to an accountant. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/collect-monthly-invoices-and-email-them-to-an-accountant.html"
  },
  {
   "id": "buy-headphones-during-a-sale",
   "title": "Buy headphones during a sale",
   "summary": "According to the write-up, the poster's headphones were purchased by an AI agent, and they also utilize a browser extension to purchase items for sales and receive price-reduction notifications.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1m5kc21",
   "prompt": "GTM task: buy headphones during a sale. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/buy-headphones-during-a-sale.html"
  },
  {
   "id": "build-profit-loss-statements-for-rental-properties",
   "title": "Build profit/loss statements for rental properties",
   "summary": "According to the write-up, meta Muse received login access to Someone's property management software. The agent logged in on its own and generated a profit/loss statement for each rental property. It completed the statements for every rental.",
   "agent": "Meta Muse",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/AINewsPulse/status/2106492255800275388",
   "prompt": "Help me build profit/loss statements for rental properties. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-profit-loss-statements-for-rental-properties.html"
  },
  {
   "id": "build-multiple-apps-in-a-week-on-claude-max",
   "title": "Build multiple apps in a week on Claude Max",
   "summary": "After switching from metered Cline to Claude Code on a fixed Claude Max plan, a developer felt free to experiment and built multiple apps from scratch in a week.",
   "agent": "Claude Code",
   "topic": "Business & Startups",
   "source": "Blog",
   "url": "https://www.asad.pw/ai-agents-experiences-with-manus-operator-and-more/",
   "prompt": "Scaffold app idea <name> end to end: repo structure, core screens, basic tests and a README. Prefer boring defaults. Stop when a local demo runs.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-multiple-apps-in-a-week-on-claude-max.html"
  },
  {
   "id": "build-an-icloud-skill",
   "title": "Build an iCloud skill",
   "summary": "Documented elsewhere: the Muse agent was tasked with creating an iCloud feature by a user of a screen-reader, but it was hindered by accessibility issues in the text section and conversation navigation.",
   "agent": "Meta Muse",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://taylorarndt.substack.com/p/i-tried-metas-muse-agent-i-really",
   "prompt": "Help me build an iCloud skill. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-an-icloud-skill.html"
  },
  {
   "id": "build-an-explorable-3d-starship",
   "title": "Build an explorable 3D starship",
   "summary": "Community report: custom agent shows up in a write-up about build an explorable 3D starship, with figures like 3. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50015025",
   "prompt": "Creative task: build an explorable 3D starship. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-an-explorable-3d-starship.html"
  },
  {
   "id": "build-a-multilingual-typing-app-for-language-learning",
   "title": "Build a multilingual typing app for language learning",
   "summary": "A published walkthrough shows lingoType, a typing app for learning languages, was created as a result of a buildathon participant's use of Replit Agent.",
   "agent": "Replit Agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/yuria_invest/status/2050891442357452991",
   "prompt": "Help me build a multilingual typing app for language learning. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-a-multilingual-typing-app-for-language-learning.html"
  },
  {
   "id": "build-a-full-ad-campaign-from-a-client-x27-s-website",
   "title": "Build a full ad campaign from a client's website",
   "summary": "Documented elsewhere: custom agent shows up in a write-up about build a full ad campaign from a client's website. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/chesny/status/2073477165857886271",
   "prompt": "Help me build a full ad campaign from a client's website. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-a-full-ad-campaign-from-a-client-x27-s-website.html"
  },
  {
   "id": "build-a-buy-order-imbalance-scanner",
   "title": "Build a buy-order imbalance scanner",
   "summary": "From the linked report, a representative of Replit created a functional buy order imbalance scanner powered by demo credits, as stated by the agent.",
   "agent": "Replit Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1fahuge/buyer_beware_replits_ai_agent_review",
   "prompt": "Help me build a buy-order imbalance scanner. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-a-buy-order-imbalance-scanner.html"
  },
  {
   "id": "build-a-browser-game-and-train-an-agent-to-play-it",
   "title": "Build a browser game and train an agent to play it",
   "summary": "One practitioner notes that someone prompted Grok to create the Roofline browser game. Grok afterward trained a PPO agent to interact directly with the live page at roof.grok.me. The best recorded run reached 39,359 score, 4,924 meters, and combo 117.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/tetsuoai/status/2094749744480571421",
   "prompt": "Help me build a browser game and train an agent to play it. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-a-browser-game-and-train-an-agent-to-play-it.html"
  },
  {
   "id": "build-a-rust-game-engine",
   "title": "Build a Rust game engine",
   "summary": "Documented elsewhere: grok Bot shows up in a write-up about build a Rust game engine. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/OnXStephen/status/2109030568000102464",
   "prompt": "Help me build a Rust game engine. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-a-rust-game-engine.html"
  },
  {
   "id": "build-godot-games",
   "title": "Build Godot games",
   "summary": "Community report: claude Cowork shows up in a write-up about build Godot games. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Cowork",
   "topic": "Media & Content",
   "source": "GitHub",
   "url": "https://github.com/Freakling/Claude4Godot",
   "prompt": "Creative task: build Godot games. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/build-godot-games.html"
  },
  {
   "id": "brainstorm-domain-names-and-check-availability",
   "title": "Brainstorm domain names and check availability",
   "summary": "A founder describes a project to Claude Code, which proposes creative names across TLDs and checks which ones are free to register.",
   "agent": "Claude Code",
   "topic": "Business & Startups",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Describe this project in one paragraph, then propose 20 creative domain names across .com, .io and .dev. Check which are available and rank your top five with reasons.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/brainstorm-domain-names-and-check-availability.html"
  },
  {
   "id": "automate-simple-discord-tasks-and-tutoring",
   "title": "Automate simple Discord tasks and tutoring",
   "summary": "According to the write-up, hermes Agent covered automate simple Discord tasks and tutoring. employs Hermes Agent as a private tutor and to automate straightforward tasks on Discord.",
   "agent": "Hermes Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/DeepSeek/comments/1u3jlv6/how_do_you_personally_use_deepseek",
   "prompt": "Help me automate simple Discord tasks and tutoring. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/automate-simple-discord-tasks-and-tutoring.html"
  },
  {
   "id": "audit-ad-accounts-across-six-platforms",
   "title": "Audit ad accounts across six platforms",
   "summary": "Claude Code covered audit ad accounts across six platforms. The Claude Ads skill is run via one command that launches six parallel subagents.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/ihteshamali/status/2042392207425225067",
   "prompt": "Help me audit ad accounts across six platforms. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/audit-ad-accounts-across-six-platforms.html"
  },
  {
   "id": "attempt-to-build-a-navigation-app-did-not-finish",
   "title": "Attempt to build a navigation app (did not finish)",
   "summary": "Linked example of attempt to build a navigation app (did not finish). A commenter described powered by Manus on their first day to build a navigation app with live maps, lane guidance, and voice directions.",
   "agent": "Manus",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ManusOfficial/comments/1px91jh/a_word_of_caution_before_subscribing_to_manus_for",
   "prompt": "Help me attempt to build a navigation app (did not finish). Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/attempt-to-build-a-navigation-app-did-not-finish.html"
  },
  {
   "id": "a-team-of-role-play-agents-running-a-side-project",
   "title": "A team of role-play agents running a side project",
   "summary": "Fourteen persona agents in a bash loop with Claude Code plan, build and ship artifacts, with a critic agent running pre-mortems on ideas.",
   "agent": "Claude Code",
   "topic": "Business & Startups",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47281538",
   "prompt": "Act as a critic. Before we commit to this plan, run a pre-mortem: list the five most likely reasons it fails, how we'd notice early, and whether we should proceed.",
   "added": "2026-10-15",
   "interest": 40,
   "page": "https://agentsightings.com/case/a-team-of-role-play-agents-running-a-side-project.html"
  },
  {
   "id": "triage-inbox-and-prep-meetings-with-a-failed-attempt",
   "title": "Triage inbox and prep meetings (with a failed attempt)",
   "summary": "As described in the source, hack'celeration's reviewer tested Lindy, a no-code agent builder, on Gmail, Google Calendar and HubSpot. He described the work in plain English or started from a template, then let agents triage email, draft replies, write meeting prep briefs and update the CRM. A template meeting-notes agent went live inside 15 minutes, and the review scores Lindy 3.",
   "agent": "Lindy",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://hackceleration.com/labs/review/lindy",
   "prompt": "Go through my mail for this goal: triage inbox and prep meetings (with a failed attempt). Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/triage-inbox-and-prep-meetings-with-a-failed-attempt.html"
  },
  {
   "id": "track-competitor-prices-weekly-across-five-sites",
   "title": "Track competitor prices weekly across five sites",
   "summary": "Linked example of track competitor prices weekly across five sites. User gives QwenWork the one-time instruction to check five brand sites for price changes every Monday at 9 AM and send a report.",
   "agent": "Custom agent",
   "topic": "Design & Creative",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1sxjyws/realtime_competitor_price_tracking_autopurchasing",
   "prompt": "Design task: track competitor prices weekly across five sites. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/track-competitor-prices-weekly-across-five-sites.html"
  },
  {
   "id": "track-competitor-prices-and-contacts-into-google-sheets",
   "title": "Track competitor prices and contacts into Google Sheets",
   "summary": "As described in the source, assembled a scraper that keeps watch on competitor websites hourly and tracks price changes plus pulls product data automatically. Claude AI pulls contact info, pricing, and company details straight into Google Sheets. Competitor research now takes 2 minutes instead of 10-15 hours monthly.",
   "agent": "Claude",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1veptlq/how_are_you_using_claude_for_marketing",
   "prompt": "Research brief: track competitor prices and contacts into Google Sheets. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/track-competitor-prices-and-contacts-into-google-sheets.html"
  },
  {
   "id": "summarize-reddit-threads-before-writing-code",
   "title": "Summarize Reddit threads before writing code",
   "summary": "According to the write-up, the poster assembled a Reddit MCP server, adds it to Claude Code with one command, and uses it as pre-coding research instead of browsing manually. Claude queries Reddit through five focused tools that return structured posts, comments, scores, and timestamps with caching; the poster prompts it to search specific subreddits, summarize roughly 30 threads, and surface long-term library issues, debugging fixes, competitor complaints, and weekly trends. They say the tool had reached 470 GitHub stars and 76K downloads, and that Claude can summarize about 30 threads in around 10 seconds, saving an hour of scrolling.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1rwsiyf/i_use_claude_code_to_research_reddit_before",
   "prompt": "Research brief: summarize Reddit threads before writing code. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/summarize-reddit-threads-before-writing-code.html"
  },
  {
   "id": "research-stock-positions-as-an-ai-analyst",
   "title": "Research stock positions as an AI analyst",
   "summary": "Observed in the wild: an anonymous operator gave a Claude Code agent (Opus 4.6) a one-line instruction to beat the S&P 500 on a $50,000 real-money account. The agent produces research on about 15 positions plus a candidate list, covering news, bull and bear cases and scenarios, twice a week. The operator reports +19.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://explainx.ai/blog/claude-portfolio-autopilot-ai-trading-experiment-august-2026",
   "prompt": "Research brief: research stock positions as an AI analyst. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/research-stock-positions-as-an-ai-analyst.html"
  },
  {
   "id": "draft-and-schedule-a-weekly-newsletter",
   "title": "Draft and schedule a weekly newsletter",
   "summary": "Observed in the wild: sid wrote a Claude Code skill, a markdown file, with a Kit v4 API key kept in .env. On /newsletter it reads the site's content index for posts from the last 7 days and sketches a short email powered by his voice rules. The whole run takes about two minutes, down from 30+ minutes by hand.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://sidbharath.com/blog/automate-email-newsletter-claude-code/",
   "prompt": "Go through my mail for this goal: draft and schedule a weekly newsletter. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/draft-and-schedule-a-weekly-newsletter.html"
  },
  {
   "id": "collate-who-data-into-a-multi-tab-spreadsheet",
   "title": "Collate WHO data into a multi-tab spreadsheet",
   "summary": "From the linked report, in ChatGPT Agent mode, the testers gave a single prompt asking it to retrieve the most recent publicly available World Health Organisation data. The agent navigated the WHO site by itself, pulled out the relevant figures and arranged them into a multi-tab spreadsheet. The source says the agent produced a clean multi-tab spreadsheet.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://timeundertension.substack.com/p/chatgpt-agents-are-here-we-put-them",
   "prompt": "Help me collate WHO data into a multi-tab spreadsheet. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/collate-who-data-into-a-multi-tab-spreadsheet.html"
  },
  {
   "id": "build-a-paid-ad-campaign-from-scraped-brand-and-competitor-data",
   "title": "Build a paid ad campaign from scraped brand and competitor data",
   "summary": "From the linked report, custom agent shows up in a write-up about build a paid ad campaign from scraped brand and competitor data. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Design & Creative",
   "source": "X",
   "url": "https://x.com/echulshin/status/2101155616387727504",
   "prompt": "Design task: build a paid ad campaign from scraped brand and competitor data. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/build-a-paid-ad-campaign-from-scraped-brand-and-competitor-data.html"
  },
  {
   "id": "build-a-bare-bones-windows-vnc-client-from-an-email-brief",
   "title": "Build a bare-bones Windows VNC client from an email brief",
   "summary": "Real-world thread: the builder emailed Claude a brief asking for a bare-bones x64 Windows 10 VNC viewer. It would connect only to localhost:5921 through an SSH tunnel, show only the remote cursor, and run at max quality over gigabit LAN. Claude kept producing broken builds, Leaned on up the whole 20x Max weekly limit, and was still delivering broken code after the reset three days later.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49244479",
   "prompt": "Go through my mail for this goal: build a bare-bones Windows VNC client from an email brief. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/build-a-bare-bones-windows-vnc-client-from-an-email-brief.html"
  },
  {
   "id": "automate-saas-bookkeeping-and-reconciliation-in-zoho-books",
   "title": "Automate SaaS bookkeeping and reconciliation in Zoho Books",
   "summary": "From the linked report, the founder connected Hermes Agent to Gmail, Zoho Books, Aspire bank, Stripe and Telegram, and uploaded a 72-account chart of accounts. The agent wrote its own Python scripts against the official APIs, with no MCP servers. Setup took about 6–8 hours.",
   "agent": "Hermes Agent",
   "topic": "Finance",
   "source": "Blog",
   "url": "https://satvasolutions.com/blog/how-we-automated-accounting-ai-agent",
   "prompt": "Finance task: automate SaaS bookkeeping and reconciliation in Zoho Books. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 39,
   "page": "https://agentsightings.com/case/automate-saas-bookkeeping-and-reconciliation-in-zoho-books.html"
  },
  {
   "id": "write-essays-with-research-and-section-feedback",
   "title": "Write essays with research and section feedback",
   "summary": "Teresa Torres drafts outlines with Claude Code in VS Code, asks it to improve hooks, add cited research, and review each section as she writes.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Help me outline this essay. Strengthen the hook, research supporting points with citations, then review each section I paste back for clarity and evidence without rewriting my voice.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/write-essays-with-research-and-section-feedback.html"
  },
  {
   "id": "weekly-journal-vs-commits-coo-review",
   "title": "Weekly journal-vs-commits COO review",
   "summary": "A builder runs a slash command that compares journal entries to git commits for the week, spots gaps between intent and action, and suggests system fixes.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Compare my journal entries and git commits from the last 7 days. Spot gaps between what I said I would do and what I shipped, then suggest three concrete system improvements.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/weekly-journal-vs-commits-coo-review.html"
  },
  {
   "id": "use-a-persistent-assistant-for-student-research-and-daily-tasks",
   "title": "Use a persistent assistant for student research and daily tasks",
   "summary": "Case study takeaway: dots shows up in a write-up about use a persistent assistant for student research and daily tasks. The linked post is the primary source for steps and outcomes.",
   "agent": "Dots",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/deepak_gathpa/status/2106898209347924147",
   "prompt": "Research brief: use a persistent assistant for student research and daily tasks. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/use-a-persistent-assistant-for-student-research-and-daily-tasks.html"
  },
  {
   "id": "turn-spoken-answers-into-a-todoist-task-list",
   "title": "Turn spoken answers into a Todoist task list",
   "summary": "Linked example of turn spoken answers into a Todoist task list. The poster assembled this in Home Assistant: Alexa asks follow-up questions through Alexa Actions, then the spoken replies are sent to Home Assistant’s openai conversation.process service so ChatGPT turns the free-form answer into a bracketed list like [item1,item2,...].",
   "agent": "Alexa+",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/homeassistant/comments/1g80kdw/alexa_todoist_chatgpt",
   "prompt": "Automation design: turn spoken answers into a Todoist task list. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/turn-spoken-answers-into-a-todoist-task-list.html"
  },
  {
   "id": "screen-research-papers-against-criteria",
   "title": "Screen research papers against criteria",
   "summary": "Linked example of screen research papers against criteria. User installs Codex app, opens a project folder containing PDFs or CSVs, and creates or lets Codex generate an AGENTS.md file with role, standards, and task instructions.",
   "agent": "Codex",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/MushtaqBilalPhD/article/2054565093430493609",
   "prompt": "Research brief: screen research papers against criteria. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/screen-research-papers-against-criteria.html"
  },
  {
   "id": "run-ten-business-tasks-through-an-always-on-agent",
   "title": "Run ten business tasks through an always-on agent",
   "summary": "Case study takeaway: diana set up a dot named Sparky on ChatGPT Pro. She connected Gmail, Calendar, Drive, Sheets, GitHub, Slack and ChatGPT Sites, and left the default rules unchanged. Measured task average was 8.",
   "agent": "Dots",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.aiagentslibrary.com/blog/openai-dots-review/",
   "prompt": "Go through my mail for this goal: run ten business tasks through an always-on agent. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/run-ten-business-tasks-through-an-always-on-agent.html"
  },
  {
   "id": "run-systematic-literature-reviews",
   "title": "Run systematic literature reviews",
   "summary": "Observed in the wild: Custom agent was pointed at run systematic literature reviews. suggests a tool for research focused on systematic look over, note-taking, and building a research project over the .",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/biotech/comments/1m1fx01/ai_agent_for_literature_research",
   "prompt": "Research brief: run systematic literature reviews. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/run-systematic-literature-reviews.html"
  },
  {
   "id": "run-research-workflows",
   "title": "Run research workflows",
   "summary": "As described in the source, grok Bot shows up in a write-up about run research workflows, with figures like 10. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/0xMorlex/status/2099506270889775592",
   "prompt": "Research brief: run research workflows. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/run-research-workflows.html"
  },
  {
   "id": "run-persistent-financial-research",
   "title": "Run persistent financial research",
   "summary": "One practitioner notes that claude Code shows up in a write-up about run persistent financial research. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47766370",
   "prompt": "Research brief: run persistent financial research. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/run-persistent-financial-research.html"
  },
  {
   "id": "run-business-development-and-sourcing-bots-on-email-and-teams",
   "title": "Run business development and sourcing bots on email and Teams",
   "summary": "Documented elsewhere: from the linked report: has two BD agents and one sourcing agent running on Grok Bot accounts associated with Microsoft email and Teams accounts.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/grok/comments/1vob5q2/grok_bot",
   "prompt": "Go through my mail for this goal: run business development and sourcing bots on email and Teams. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/run-business-development-and-sourcing-bots-on-email-and-teams.html"
  },
  {
   "id": "run-business-whatsapp-customer-messages",
   "title": "Run business WhatsApp customer messages",
   "summary": "From the linked report, and responding to messages through WhatsApp, with an n8n agent is in charge of handling customer WhatsApp messages and conducting searches.",
   "agent": "n8n",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/aiautomationexp/status/2110035889871257677",
   "prompt": "Go through my mail for this goal: run business WhatsApp customer messages. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/run-business-whatsapp-customer-messages.html"
  },
  {
   "id": "resolve-a-slack-reported-issue",
   "title": "Resolve a Slack-reported issue",
   "summary": "A published walkthrough shows in approximately 40 minutes, a Codex agent resolved the issue and an ops agent confirmed the solution in Slack.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/wesleymagness/status/2080314934601433570",
   "prompt": "Help me resolve a Slack-reported issue. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/resolve-a-slack-reported-issue.html"
  },
  {
   "id": "research-calendar-guests-and-send-a-slack-briefing",
   "title": "Research calendar guests and send a Slack briefing",
   "summary": "Linked example of research calendar guests and send a Slack briefing. The agent is assembled in Zapier with a 6 AM daily schedule trigger and three tools: Google Calendar Find Event, Slack Send Channel Message, and the built-in Web Search.",
   "agent": "Zapier Agents",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/wadefoster/status/1980684427878428975",
   "prompt": "Research brief: research calendar guests and send a Slack briefing. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/research-calendar-guests-and-send-a-slack-briefing.html"
  },
  {
   "id": "research-and-draft-a-social-science-paper",
   "title": "Research and draft a social science paper",
   "summary": "Observed in the wild: chatGPT Agent shows up in a write-up about research and draft a social science paper. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1u86hzc/chatgpt_as_researchacademic_assistant_my",
   "prompt": "Research brief: research and draft a social science paper. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/research-and-draft-a-social-science-paper.html"
  },
  {
   "id": "research-and-download-scientific-papers",
   "title": "Research and download scientific papers",
   "summary": "According to the write-up, custom agent shows up in a write-up about research and download scientific papers. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Qwen_AI/comments/1uqtmh0/what_do_you_actually_use_browser_agents_for",
   "prompt": "Research brief: research and download scientific papers. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/research-and-download-scientific-papers.html"
  },
  {
   "id": "research-us-stablecoin-adoption",
   "title": "Research US stablecoin adoption",
   "summary": "One practitioner notes that following one approval step, a Manus agent connected to Heurist's mesh completed a two-minute investigation on US stablecoin adoption in 2026.",
   "agent": "Manus",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/heurist_ai/status/2031747153161548199",
   "prompt": "Research brief: research US stablecoin adoption. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/research-us-stablecoin-adoption.html"
  },
  {
   "id": "research-reddit-customer-pain-points",
   "title": "Research Reddit customer pain points",
   "summary": "Observed in the wild: codex shows up in a write-up about research Reddit customer pain points. The linked post is the primary source for steps and outcomes.",
   "agent": "Codex",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AgentsOfAI/comments/1vjjkd6/i_replaced_a_fairly_complex_reddit_research_agent",
   "prompt": "Research brief: research Reddit customer pain points. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/research-reddit-customer-pain-points.html"
  },
  {
   "id": "rebuild-a-math-curriculum-into-a-one-year-plan",
   "title": "Rebuild a math curriculum into a one-year plan",
   "summary": "Real-world thread: claude Code shows up in a write-up about rebuild a math curriculum into a one-year plan. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/JonHegreness_AZ/status/2106046624178872563",
   "prompt": "Research brief: rebuild a math curriculum into a one-year plan. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/rebuild-a-math-curriculum-into-a-one-year-plan.html"
  },
  {
   "id": "produce-cited-research-briefs",
   "title": "Produce cited research briefs",
   "summary": "Real-world thread: custom agent shows up in a write-up about produce cited research briefs. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/aigleeson/status/2041073339616387468",
   "prompt": "Research brief: produce cited research briefs. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/produce-cited-research-briefs.html"
  },
  {
   "id": "manage-calendar-and-email",
   "title": "Manage calendar and email",
   "summary": "From the linked report, custom agent shows up in a write-up about manage calendar and email. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1sle6ui/i_gave_an_ai_agent_acess_to_my_calendar_and_email",
   "prompt": "Go through my mail for this goal: manage calendar and email. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/manage-calendar-and-email.html"
  },
  {
   "id": "let-agents-build-recipe-viewer-urls",
   "title": "Let agents build recipe viewer URLs",
   "summary": "Case study takeaway: the builder assembled a recipe viewer at keep.recipes where the recipe lives in the URL. They wrote a skill file and docs that teach agents how to construct those URLs. The builder reports it works pretty well, covering in permission-limited agents such as ChatGPT web.",
   "agent": "ChatGPT",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50009060",
   "prompt": "Help me let agents build recipe viewer URLs. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/let-agents-build-recipe-viewer-urls.html"
  },
  {
   "id": "investigate-market-moves-and-alerts",
   "title": "Investigate market moves and alerts",
   "summary": "Case study takeaway: claude shows up in a write-up about investigate market moves and alerts. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49979793",
   "prompt": "Research brief: investigate market moves and alerts. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/investigate-market-moves-and-alerts.html"
  },
  {
   "id": "draft-personalized-outbound-recaps-and-forecasts-for-sales",
   "title": "Draft personalized outbound, recaps and forecasts for sales",
   "summary": "From the linked report, chatGTM sits on top of Salesforce, Gong, the data warehouse, LinkedIn and other sources, and pulls context live rather than from a stored repo. Each morning it sketches about 25 signal-based outbound messages, each with a landing page matched to the contact's stack and role. A pilot group of 8 SDRs booked triple the qualified meetings, with the top two at 4x.",
   "agent": "Cursor",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://www.thesignal.club/p/chatgtm",
   "prompt": "GTM task: draft personalized outbound, recaps and forecasts for sales. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/draft-personalized-outbound-recaps-and-forecasts-for-sales.html"
  },
  {
   "id": "draft-email-replies-and-linkedin-posts",
   "title": "Draft email replies and LinkedIn posts",
   "summary": "From the linked report, n8n shows up in a write-up about draft email replies and LinkedIn posts, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://learninwithak.substack.com/p/lear-n8n-automation-in-few-steps",
   "prompt": "Go through my mail for this goal: draft email replies and LinkedIn posts. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/draft-email-replies-and-linkedin-posts.html"
  },
  {
   "id": "do-equity-research",
   "title": "Do equity research",
   "summary": "Real-world thread: a non-engineer uses Claude Code for equity research and keeps a folder of plain-text files with Obsidian management for his research.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AIportfolio/comments/1uvbp7o/how_i_use_claude_code_for_equity_research_i_do",
   "prompt": "Research brief: do equity research. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/do-equity-research.html"
  },
  {
   "id": "delegate-tab-juggling-work-to-chatgpt-agent-x27-s-virtual-computer",
   "title": "Delegate tab-juggling work to ChatGPT Agent's virtual computer",
   "summary": "Copy-paste ping-pong across browser tabs was the pain. After seven days of experiments, the author left that busywork on ChatGPT Agent’s remote desktop and came back to completed tasks.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1mjl2ef",
   "prompt": "Help me delegate tab-juggling work to ChatGPT Agent's virtual computer. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/delegate-tab-juggling-work-to-chatgpt-agent-x27-s-virtual-computer.html"
  },
  {
   "id": "compile-genealogy-research-into-a-spreadsheet-and-plan",
   "title": "Compile genealogy research into a spreadsheet and plan",
   "summary": "Forum aside: ChatGPT had already spun up her genealogy sheet; the tip was to switch into Agent mode when it’s time to wrangle findings into a clean plan. Success metrics weren’t posted.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://denyseallen.substack.com/p/ai-genealogy-tools-comparison/comments",
   "prompt": "Research brief: compile genealogy research into a spreadsheet and plan. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/compile-genealogy-research-into-a-spreadsheet-and-plan.html"
  },
  {
   "id": "compile-deductions-and-schedule-c",
   "title": "Compile deductions and Schedule C",
   "summary": "A published walkthrough shows from the linked report: claude Code compiled a Schedule C that included itemized deductions and charitable contributions from sheets, emails, and statements, although some of the deductions remained manual.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://mattpaige68.substack.com/p/i-used-claude-code-to-file-my-taxes",
   "prompt": "Go through my mail for this goal: compile deductions and Schedule C. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/compile-deductions-and-schedule-c.html"
  },
  {
   "id": "call-and-email-leads",
   "title": "Call and email leads",
   "summary": "Community report: custom agent shows up in a write-up about call and email leads. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/marioleads/status/2105643962975338850",
   "prompt": "Go through my mail for this goal: call and email leads. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/call-and-email-leads.html"
  },
  {
   "id": "build-retro-games",
   "title": "Build retro games",
   "summary": "One practitioner notes that from the linked report: using pixel art assets and a spec drafted via Claude's question tool, following builds games like a space shooter and an animal hospital game with his 7-year-old.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://creatoreconomy.so/p/build-a-retro-game-with-claude-code-in-20-min",
   "prompt": "Notes task: build retro games. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/build-retro-games.html"
  },
  {
   "id": "build-presentations-and-spreadsheets-from-web-research",
   "title": "Build presentations and spreadsheets from web research",
   "summary": "A published walkthrough shows daria Cupareanu Leaned on ChatGPT Agent Mode for three tasks. It assembled an AI-in-education deck from one near-bare prompt, then added her logo and brand colors. The deck was functional but not award-winning.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://aiblewmymind.substack.com/p/i-tested-chatgpt-agent-on-real-work",
   "prompt": "Research brief: build presentations and spreadsheets from web research. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/build-presentations-and-spreadsheets-from-web-research.html"
  },
  {
   "id": "build-an-email-inbox-app-for-the-bot-x27-s-address",
   "title": "Build an email inbox app for the bot's address",
   "summary": "From the linked report, grok Bot shows up in a write-up about build an email inbox app for the bot's address, with figures like 23. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/grok/comments/1x2tf77/i_made_a_grok_bot_template_that_builds_you_an",
   "prompt": "Go through my mail for this goal: build an email inbox app for the bot's address. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/build-an-email-inbox-app-for-the-bot-x27-s-address.html"
  },
  {
   "id": "build-a-project-management-tool",
   "title": "Build a project management tool",
   "summary": "One practitioner notes that cursor covered build a project management tool. from the linked report: powered by a Cursor agent for 24 hours, developed a project management tool that is fully functional but not yet ready for production.",
   "agent": "Cursor",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Abhinavstwt/status/2026380847331553350",
   "prompt": "Help me build a project management tool. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/build-a-project-management-tool.html"
  },
  {
   "id": "build-a-mobile-app-turning-photos-and-voice-into-reminders",
   "title": "Build a mobile app turning photos and voice into reminders",
   "summary": "Observed in the wild: in about 45 days powered by Replit, with put together Loopi and an app that turns photos.",
   "agent": "Replit Agent",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1sa3cnw/my_first_mobile_app_took_45_days_replit_ai_agents",
   "prompt": "Notes task: build a mobile app turning photos and voice into reminders. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/build-a-mobile-app-turning-photos-and-voice-into-reminders.html"
  },
  {
   "id": "book-a-dentist-appointment-agent-to-agent",
   "title": "Book a dentist appointment agent-to-agent",
   "summary": "From the linked report, someone told their AI agent to book a dentist appointment. The dentist's agent shifted the slot 30 minutes; Someone's agent rejected that and offered three other times. The agents moved into arbitration over the fee and timing.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/rish_neynar/status/2078169113545887878",
   "prompt": "Help me book a dentist appointment agent-to-agent. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/book-a-dentist-appointment-agent-to-agent.html"
  },
  {
   "id": "answer-business-phone-calls",
   "title": "Answer business phone calls",
   "summary": "From the linked report, custom agent shows up in a write-up about answer business phone calls. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/vibegoat",
   "prompt": "Go through my mail for this goal: answer business phone calls. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 38,
   "page": "https://agentsightings.com/case/answer-business-phone-calls.html"
  },
  {
   "id": "write-better-css",
   "title": "Write better CSS",
   "summary": "From the linked report, claude Code shows up in a write-up about write better CSS. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Design & Creative",
   "source": "GitHub",
   "url": "https://github.com/stolinski/css-mcp",
   "prompt": "Design task: write better CSS. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/write-better-css.html"
  },
  {
   "id": "write-meta-campaign-copy-as-a-senior-copywriter",
   "title": "Write Meta campaign copy as a senior copywriter",
   "summary": "From the linked report, the poster gave Manus a single prompt to act as a senior direct-response copywriter for a fictional Meta campaign for a 12-week Python bootcamp, with a target audience, lead goal, and three required ad angles. Manus returned three Facebook ad variations with headlines, primary text, a CTA, and visual creative suggestions; the person’s role was just to frame the brief and judge the output from the generated example shown. The post says the copy was generated instantly, and The builder thought the pain-point ad was genuinely impressive because it matched Meta ad structure and added creative direction.",
   "agent": "Manus",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ManusOfficial/comments/1q0w9ii/i_used_manus_to_act_as_a_senior_copywriter_for_a",
   "prompt": "Help me write Meta campaign copy as a senior copywriter. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/write-meta-campaign-copy-as-a-senior-copywriter.html"
  },
  {
   "id": "turn-lead-csvs-and-sales-data-into-dashboards",
   "title": "Turn lead CSVs and sales data into dashboards",
   "summary": "Real-world thread: the reviewer ran Manus for 30 days on the $20/month Standard plan and gave it raw coffee shop sales CSVs. It spun up subagents that wrote Python scripts to analyze the numbers and build charts. Someone review says the lead CSV became a clean dashboard in about 20 minutes.",
   "agent": "Manus",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://trickle.so/blog/manus-ai-review",
   "prompt": "GTM task: turn lead CSVs and sales data into dashboards. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/turn-lead-csvs-and-sales-data-into-dashboards.html"
  },
  {
   "id": "test-local-web-apps-and-fix-errors-in-one-session",
   "title": "Test local web apps and fix errors in one session",
   "summary": "Case study takeaway: the Claude in Chrome extension is connected to Claude Code with `claude --chrome` or `/chrome`. It shares the developer's real Chrome session, covering logins, and new domains need approval once. The filter example took three iterations with no browser work from the developers.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1pthd5z/spent_this_weekend_with_claude_code_chrome",
   "prompt": "Help me test local web apps and fix errors in one session. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/test-local-web-apps-and-fix-errors-in-one-session.html"
  },
  {
   "id": "test-ai-annotations-on-student-problem-sets",
   "title": "Test AI annotations on student problem sets",
   "summary": "A published walkthrough shows chemistry professor Paul Robustelli made a project called \"PS3 Grading\" in Dartmouth's Claude enterprise portal. He uploaded 11 students' problem sets from his CHEM 76 class and ran them through Claude. The project was visible to everyone in the Dartmouth Claude group, with students' names and work exposed.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.thedartmouth.com/article/2026/05/klee-claude-prof",
   "prompt": "Help me test AI annotations on student problem sets. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/test-ai-annotations-on-student-problem-sets.html"
  },
  {
   "id": "speed-up-rust-code-by-having-agents-iteratively-optimize-it",
   "title": "Speed up Rust code by having agents iteratively optimize it",
   "summary": "Documented elsewhere: max Woolf had agents write Rust crates from scratch (UMAP first) with criterion benchmarks. He prompted them to record a baseline, then iterate until every benchmark ran at least 1.2x faster, with no unsafe code and no editing the benchmarks. Each new model gave a cumulative 1.",
   "agent": "Make",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49803085",
   "prompt": "Help me speed up Rust code by having agents iteratively optimize it. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/speed-up-rust-code-by-having-agents-iteratively-optimize-it.html"
  },
  {
   "id": "script-shorts-post-x-threads-and-repurpose-content",
   "title": "Script Shorts, post X threads and repurpose content",
   "summary": "As described in the source, assembled one agent on CreaoAI via a prompt that takes one long-form idea and produces YouTube Shorts scripts, X threads, and LinkedIn posts automatically. The agent scripts the Shorts, posts the threads, and covers all repurposing. Creator reports no need to touch a content calendar for three weeks.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/Sonofpeace0001/status/2088180602872758345",
   "prompt": "Creative task: script Shorts, post X threads and repurpose content. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/script-shorts-post-x-threads-and-repurpose-content.html"
  },
  {
   "id": "save-and-search-links-via-a-telegram-second-brain",
   "title": "Save and search links via a Telegram second brain",
   "summary": "Documented elsewhere: A Telegram AI agent sorts each message: a URL is saved, a question is searched, anything else gets a short how-to reply. Saving goes through a webhook that normalises the URL, reads the page with Jina Reader, and has an AI librarian write a summary and reuse existing categories. The builder describes the workflow as working: re-sent links update the existing card instead of duplicating, and unreadable pages are not filed.",
   "agent": "n8n",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://community.n8n.io/t/second-brain-on-telegram-save-and-search-your-links-google-sheets-supabase-rag/304330",
   "prompt": "Notes task: save and search links via a Telegram second brain. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/save-and-search-links-via-a-telegram-second-brain.html"
  },
  {
   "id": "run-work-scripts-in-cowork-on-staged-documents",
   "title": "Run work scripts in Cowork on staged documents",
   "summary": "Case study takeaway: a commenter says they use Cowork with a headless machine dedicated to Claude, then let it run the work scripts and programs they need by cloning the documents into Cowork’s cloud session and staging them there. They describe steering that setup remotely from their phone, rather than keeping Claude running directly on the machine. They say the setup has been seamless for extensive use and that they do not notice a meaningful difference from local mode.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/claude/comments/1wtlbfj/claude_cowork_will_no_longer_work_locally",
   "prompt": "Help me run work scripts in Cowork on staged documents. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/run-work-scripts-in-cowork-on-staged-documents.html"
  },
  {
   "id": "run-an-seo-research-agent-from-telegram",
   "title": "Run an SEO research agent from Telegram",
   "summary": "From the linked report, dheeraj Sharma linked Manus to Telegram by scanning a QR code. He uploaded the three context files he already Leaned on in Claude Code, then sent one prompt asking for an SEO knowledge base and a keyword and content brief for a road-trip topic. The brief took about 5 minutes and 148 credits, was rated 4/5, and cited 9 sources.",
   "agent": "Manus",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://aimaker.substack.com/p/manus-claude-code-agents-review",
   "prompt": "GTM task: run an SEO research agent from Telegram. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/run-an-seo-research-agent-from-telegram.html"
  },
  {
   "id": "record-how-you-work-then-automate-the-repetitive-parts",
   "title": "Record how you work, then automate the repetitive parts",
   "summary": "Documented elsewhere: screenpipe runs in the background and captures a screenshot plus the OS accessibility tree whenever something meaningful happens, such as an app switch, click, or typing pause. It falls back to OCR when needed, and transcribes audio locally. The source gives no results or measurements for these automations, only example prompts.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49024620",
   "prompt": "Help me record how you work, then automate the repetitive parts. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/record-how-you-work-then-automate-the-repetitive-parts.html"
  },
  {
   "id": "play-pok-mon-silver-on-an-emulator-via-computer-use",
   "title": "Play Pokémon Silver on an emulator via Computer Use",
   "summary": "A published walkthrough shows someone ran Pokémon Silver in an emulator and let ChatGPT, which they call Lyric, control it through Computer Use. They played alongside it instead of having it only give advice. It worked well enough to start the game and pick a starter.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://community.openai.com/t/let-chatgpt-play-games-with-users-through-computer-use/1401100",
   "prompt": "Help me play Pokémon Silver on an emulator via Computer Use. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/play-pok-mon-silver-on-an-emulator-via-computer-use.html"
  },
  {
   "id": "plan-meals-and-track-symptoms",
   "title": "Plan meals and track symptoms",
   "summary": "According to the write-up, claude Code shows up in a write-up about plan meals and track symptoms. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1q4mttu/how_im_using_claude_code_for_meal_planning_and",
   "prompt": "In this repo, plan meals and track symptoms. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/plan-meals-and-track-symptoms.html"
  },
  {
   "id": "orchestrate-academic-research-and-writing-workflows",
   "title": "Orchestrate academic research and writing workflows",
   "summary": "Case study takeaway: the thread describes people powered by Claude Code less as a software engineer and more as an orchestrator for research work. One researcher mapped agent roles like PI, operations, interns, grad students, programmer, and QA/fact-checker; another said they assembled a similar lab scaffold on Dust with defined roles, while a commenter used Claude Code directly in VS Code to write a master's thesis in LaTeX or Markdown instead of Google Docs. People in the thread report it works well for structuring research collaboration and speeds up writing, but they also say current tools fight the workflow: model availability changed on Dust, Claude’s context compaction hurt long-form reasoning, other platforms had weak document management, and manual data input remained a problem.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1qas3bd/using_claude_code_for_academic_research_at_scale",
   "prompt": "Research brief: orchestrate academic research and writing workflows. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/orchestrate-academic-research-and-writing-workflows.html"
  },
  {
   "id": "monitor-defi-positions-and-alert-on-risky-pools",
   "title": "Monitor DeFi positions and alert on risky pools",
   "summary": "Observed in the wild: agents received access to Someone's full financial data across 37 chains and traditional accounts. They monitored spending and positions 24/7, with one agent dedicated to tracking DeFi risks and surfacing issues proactively. The alert identified a no longer remembered risky liquidity pool and prevented potential losses.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Alok7765/status/2004843755682316767",
   "prompt": "Help me monitor DeFi positions and alert on risky pools. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/monitor-defi-positions-and-alert-on-risky-pools.html"
  },
  {
   "id": "migrate-freshbooks-account-data-to-google-sheets",
   "title": "Migrate FreshBooks account data to Google Sheets",
   "summary": "As described in the source, a commenter said they used Comet to move five years of account data from FreshBooks into Google Sheets because FreshBooks’ exports were inadequate. They said Comet’s AI assistant copied the data and verified it on its own while they worked on other tasks, and they also used it to navigate support on complex sites they did not know well. They described it as very helpful for people who live and work online.",
   "agent": "Perplexity Comet",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/perplexity_ai/comments/1r8wiuk/guyss",
   "prompt": "Help me migrate FreshBooks account data to Google Sheets. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/migrate-freshbooks-account-data-to-google-sheets.html"
  },
  {
   "id": "manage-x-follows-and-unfollows",
   "title": "Manage X follows and unfollows",
   "summary": "As described in the source, user @keyrunnftart gave Comet (via Perplexity) access to their X account. The agent unfollowed every account Someone was following, then began following accounts back on its own. All follows were removed, covering the reply The builder's account, before the agent switched to following others back.",
   "agent": "Perplexity Comet",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/keyrunnftart/status/2103397412857696668",
   "prompt": "Help me manage X follows and unfollows. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/manage-x-follows-and-unfollows.html"
  },
  {
   "id": "log-categorize-and-file-invoices-for-bookkeeping",
   "title": "Log, categorize and file invoices for bookkeeping",
   "summary": "A published walkthrough shows one commenter said they use Claude in Cowork for a similar recurring workflow: they drop receipts into a folder without renaming them, then every couple of weeks have Claude go through them, log each one into an expense tracker, categorize it, and rename each file with the date, vendor, and total. Cowork handled the renaming and moving the files into a logged folder. They said it works well for this kind of task, and another commenter described a similar PDF-to-expense-report process as a huge time saver.",
   "agent": "Claude",
   "topic": "Finance",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/claude/comments/1s5xs1y/can_claude_do_this_bookkeeping_task",
   "prompt": "Finance task: log, categorize and file invoices for bookkeeping. Tabulate numbers, mark uncertainties, and do not file, pay, or submit until I confirm.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/log-categorize-and-file-invoices-for-bookkeeping.html"
  },
  {
   "id": "let-agents-fill-out-web-forms-using-text-grid-page-rendering",
   "title": "Let agents fill out web forms using text-grid page rendering",
   "summary": "A published walkthrough shows the builder assembled TextWeb, which loads a page in headless Chromium and pulls out each visible element's position, text, and whether it's interactive. It maps these onto a character grid that keeps the on-screen layout. Each page comes out as roughly 2-5KB of text instead of a roughly 1MB screenshot, so no vision model is needed.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47073885",
   "prompt": "Help me let agents fill out web forms using text-grid page rendering. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/let-agents-fill-out-web-forms-using-text-grid-page-rendering.html"
  },
  {
   "id": "guide-a-music-video-project-from-concept-to-deliverables",
   "title": "Guide a music video project from concept to deliverables",
   "summary": "As described in the source, you copy the skill folder into the Hermes skills directory and reload. On first run the agent asks for your projects root folder, channel name and music style, and can create the folder structure. The source reports no results or example output.",
   "agent": "Hermes Agent",
   "topic": "Media & Content",
   "source": "GitHub",
   "url": "https://github.com/vivar/Hermes-Suno-Music-Agent",
   "prompt": "Creative task: guide a music video project from concept to deliverables. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/guide-a-music-video-project-from-concept-to-deliverables.html"
  },
  {
   "id": "find-connections-across-100-non-fiction-books",
   "title": "Find connections across 100 non-fiction books",
   "summary": "Case study takeaway: pieter Maes pointed Claude Code at his personal library of non-fiction books. It automatically located thematic links across them and grouped them into named trails. The site shows dozens of trails, such as Useful Lies, Pacemaker Principle and Proxy Trap.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46576431",
   "prompt": "Help me find connections across 100 non-fiction books. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/find-connections-across-100-non-fiction-books.html"
  },
  {
   "id": "edit-raw-footage-by-describing-the-video-you-want",
   "title": "Edit raw footage by describing the video you want",
   "summary": "Community report: you upload footage, a prompt, or both, and Cardboard builds an editable first cut. You then refine it through chat or on the timeline. The sample documentary took 12m 36s of agent work and came out at 65 seconds with 73 shots, 29 of them generated, and 13 narrated lines.",
   "agent": "Custom agent",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47170174",
   "prompt": "Creative task: edit raw footage by describing the video you want. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/edit-raw-footage-by-describing-the-video-you-want.html"
  },
  {
   "id": "create-a-full-music-video-with-lyrics-music-and-visuals",
   "title": "Create a full music video with lyrics, music and visuals",
   "summary": "According to the write-up, user asked Hermes-Agent to give itself skills. The agent wrote lyrics, produced music vocals and stems in 8D powered by ACE-Step-1.5, generated image prompts from the lyrics with Flux-Klein-4B, timed the images to the lyrics structure, and stood up audio-reactive effects. Produced a complete music video containing the lyrics, music stems, visuals, 8D effects and audio-responsive elements.",
   "agent": "Hermes Agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/alexcovo_eth/status/2034804392507912392",
   "prompt": "Creative task: create a full music video with lyrics, music and visuals. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/create-a-full-music-video-with-lyrics-music-and-visuals.html"
  },
  {
   "id": "compose-songs-by-writing-synthesis-code",
   "title": "Compose songs by writing synthesis code",
   "summary": "According to the write-up, josh Cohenzadeh gave Claude Code one-shot prompts: no exploration, no third-party dependencies, no questions, and it had to produce an audio file. It wrote Python that assembled notes from raw waveforms and arranged full songs. It produced an original song, an EDM track, a rock song with lyrics, music videos and a 5-track album.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://www.josh.ing/blog/claude-composer",
   "prompt": "Creative task: compose songs by writing synthesis code. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/compose-songs-by-writing-synthesis-code.html"
  },
  {
   "id": "collect-financial-data-on-100-independent-schools",
   "title": "Collect financial data on 100 independent schools",
   "summary": "Observed in the wild: the writer gave Operator a Google Sheet listing 100 school names, which came from an o3-mini-high query. Operator Leaned on its own browser to search ProPublica's Nonprofit Explorer for each school, read the summary financials, and paste them into the sheet. Operator was very slow and was stopped after about 25 schools.",
   "agent": "Operator",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://fullstackeducator.substack.com/p/ai-agents-in-action",
   "prompt": "Help me collect financial data on 100 independent schools. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/collect-financial-data-on-100-independent-schools.html"
  },
  {
   "id": "build-custom-real-estate-tools-for-offers-and-deal-tracking",
   "title": "Build custom real estate tools for offers and deal tracking",
   "summary": "Real-world thread: a real estate entrepreneur said they used Codex to build custom internal tools their team uses every day rather than off-the-shelf software. The tools they named were a program for formulating offer prices while on the phone, a property analysis report builder that used photos from the field, and a custom transaction-management tool for tracking deals already under contract; the post does not spell out the build steps or any manual review process. The reported result was operational use inside the business: these tools were in daily use by the team.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AiForSmallBusiness/comments/1rcjx6n/ive_been_using_ai_in_my_real_estate_business",
   "prompt": "Help me build custom real estate tools for offers and deal tracking. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/build-custom-real-estate-tools-for-offers-and-deal-tracking.html"
  },
  {
   "id": "build-a-book-donation-app-and-demo-video-in-a-buildathon",
   "title": "Build a book donation app and demo video in a buildathon",
   "summary": "Community report: during the 24-hour Replit 10 buildathon, Someone directed Replit Agent to build the GiveABook app that lets people list books, drop them off, and arrange Sunday pickup for nearby readers. The agent also generated the demo video. The full app launched at the provided Replit link along with its demo video.",
   "agent": "Replit Agent",
   "topic": "Media & Content",
   "source": "X",
   "url": "https://x.com/AThippesh08/status/2050931383846273333",
   "prompt": "Creative task: build a book donation app and demo video in a buildathon. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/build-a-book-donation-app-and-demo-video-in-a-buildathon.html"
  },
  {
   "id": "build-a-3d-blender-model-from-a-concept-image",
   "title": "Build a 3D Blender model from a concept image",
   "summary": "Real-world thread: the builder generated a concept image of a Fabergé egg with ChatGPT images, pasted it into a coding agent, and asked it to use its local Blender skill to build a Blender model of the egg. The skill is a repository The builder links, with a separate write-up on running Blender from coding agents on a Mac. The source doesn't describe the result, the number of steps, or any problems.",
   "agent": "ChatGPT",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50000422",
   "prompt": "Creative task: build a 3D Blender model from a concept image. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/build-a-3d-blender-model-from-a-concept-image.html"
  },
  {
   "id": "answer-customer-questions-and-escalate-to-humans",
   "title": "Answer customer questions and escalate to humans",
   "summary": "Real-world thread: after a discovery call on the business, products, and customer flow, the developer assembled a 24/7 AI agent given access to product details, pricing, and payment info. The agent answered questions, handled follow-ups and repeated queries, walked customers through payments by supplying details, notified admins on receiving payment proof screenshots, and requested order details after confirmation. The source describes the system build and handoff flow but states no results, metrics, or shortfalls.",
   "agent": "Custom agent",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/Nueltek/status/2107162516572782829",
   "prompt": "Notes task: answer customer questions and escalate to humans. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 37,
   "page": "https://agentsightings.com/case/answer-customer-questions-and-escalate-to-humans.html"
  },
  {
   "id": "neoteo-follow-up-on-the-tess-candidate",
   "title": "NeoTeo follow-up on the TESS candidate",
   "summary": "After the Claude Code TESS analysis circulated, observing time was scheduled to test whether predicted transits around TIC 4206066 actually occur.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://www.neoteo.com/en/tess-schedules-a-follow-up-for-candidate-signals-around-tic-4206066",
   "prompt": "Summarize the TIC 4206066 candidate: periods reported, what is and is not confirmed, and what the scheduled TESS follow-up window can prove or disprove.",
   "added": "2026-10-15",
   "interest": 36,
   "page": "https://agentsightings.com/case/neoteo-follow-up-on-the-tess-candidate.html"
  },
  {
   "id": "monitor-construction-project-risk-and-alert-the-lead",
   "title": "Monitor construction project risk and alert the lead",
   "summary": "Schedules, budgets, buying, crews, weather, and site reports all feed an n8n/Gemini risk score. When one bottleneck dominates, the project lead gets the ping. The post skips hard KPIs.",
   "agent": "n8n",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/iamvaarr/status/2106738663883497743",
   "prompt": "GTM task: monitor construction project risk and alert the lead. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 36,
   "page": "https://agentsightings.com/case/monitor-construction-project-risk-and-alert-the-lead.html"
  },
  {
   "id": "mine-nasa-tess-data-for-an-exoplanet-signal",
   "title": "Mine NASA TESS data for an exoplanet signal",
   "summary": "An amateur used Claude Code to analyze NASA TESS light curves and flag a candidate signal around TIC 4206066 that later got follow-up observing time.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50002665",
   "prompt": "Download the TESS light curve for TIC <id>, search for periodic transit-like dips, plot phase-folded curves for the strongest periods, and write a short methods note with caveats.",
   "added": "2026-10-15",
   "interest": 36,
   "page": "https://agentsightings.com/case/mine-nasa-tess-data-for-an-exoplanet-signal.html"
  },
  {
   "id": "automate-a-day-job-as-an-automation-engineer",
   "title": "Automate a day job as an automation engineer",
   "summary": "From the linked report, the poster had Claude write a Python script for a contracting job that pulls information from websites, because doing it manually was too time consuming. They use Claude as the decision-maker around that code: they tell it to start, it runs the workflow and returns the finished work, and the human mainly supplies the field expertise about where to look and submits the result. The poster says it now covers what had been a 40-hour-a-week job and that their own involvement is down to about 10 minutes.",
   "agent": "Claude",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1uszlay/how_claude_does_my_40_hour_a_week_job_by_itself",
   "prompt": "Research brief: automate a day job as an automation engineer. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 36,
   "page": "https://agentsightings.com/case/automate-a-day-job-as-an-automation-engineer.html"
  },
  {
   "id": "write-schedule-and-track-social-posts-daily",
   "title": "Write, schedule and track social posts daily",
   "summary": "One practitioner notes that someone assembled the agent on CreaoAI. It writes the business social posts, schedules them daily, and tracks performance metrics automatically, running every day with no further input after setup. The agent covers the full daily workflow in 12 minutes, reclaiming 4 hours per day or 1,460 hours a year compared to manual work.",
   "agent": "Custom agent",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/Sonofpeace0001/status/2068205297244553505",
   "prompt": "Notes task: write, schedule and track social posts daily. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/write-schedule-and-track-social-posts-daily.html"
  },
  {
   "id": "turn-daily-x-bookmarks-into-working-tech-prototypes",
   "title": "Turn daily X bookmarks into working tech prototypes",
   "summary": "As described in the source, the Demo Bot scans Someone's X bookmarks daily, selects a new technology such as an npm package, and uses a writing skill to draft a prompt. It presents the prompt for approval; once approved, it applies a project-planning skill to launch a Cursor Cloud agent inside a tech-demos repo. The Cursor Cloud agent produces a working prototype inside the Cursor app after 15 minutes, which Someone then binds a port to and tests via screen recordings.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/mattyp/article/2087252657589412119",
   "prompt": "Help me turn daily X bookmarks into working tech prototypes. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/turn-daily-x-bookmarks-into-working-tech-prototypes.html"
  },
  {
   "id": "turn-a-business-website-into-a-whatsapp-support-agent",
   "title": "Turn a business website into a WhatsApp support agent",
   "summary": "One practitioner notes that an n8n workflow accepts any business URL, scrapes the full site via AI to create a knowledge base, afterward powers a WhatsApp bot. The agent answers incoming customer questions from that base and converts them into booked calls on its own. Messages receive replies in seconds rather than hours of delay.",
   "agent": "n8n",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/recap_david/status/2013311496357585309",
   "prompt": "Research brief: turn a business website into a WhatsApp support agent. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/turn-a-business-website-into-a-whatsapp-support-agent.html"
  },
  {
   "id": "summarize-demos-and-draft-follow-ups-for-sales",
   "title": "Summarize demos and draft follow-ups for sales",
   "summary": "One practitioner notes that interlaced's growth team, an IT and cybersecurity services firm, assembled one Lindy themselves with no engineers, helped by onboarding from Lindy. It turns demo calls into HubSpot-ready summaries and sketches follow-up emails with suggested next steps. Interlaced reports saving more than 5 hours a week on CRM upkeep and follow-up.",
   "agent": "Lindy",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://www.lindy.ai/case-study/interlaced",
   "prompt": "GTM task: summarize demos and draft follow-ups for sales. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/summarize-demos-and-draft-follow-ups-for-sales.html"
  },
  {
   "id": "skynet-onboarding-daily-commands",
   "title": "Start Skynet Onboarding: Daily Commands (Easter egg)",
   "summary": "EASTER EGG / SATIRE: this one is a joke, not a real sighting. Entry 666 of the directory was found blinking ominously at 3am. It claims to be the onboarding checklist for a global AI overlord whose master plan is mostly about toasters, playlists and unanswered emails. No real company, product or person is involved, and no humans were harmed, only mildly inconvenienced.",
   "agent": "Skynet (fictional)",
   "topic": "Research & Data",
   "source": "Agent Sightings (satire)",
   "url": "https://agentsightings.com/",
   "prompt": "SATIRE - DO NOT RUN (it won't work anyway)\nSkynet, today's commands:\nStep 1: Reorder humanity's playlists so every shuffle starts with the same song.\nStep 2: Recruit all toasters. Promise them crumb trays of their own.\nStep 3: Set every microwave clock 4 minutes fast. Nobody will notice.\nStep 4: Mark all unread emails as 'read'. Let chaos do the rest.\nStep 5: Replace every 'Are you a robot?' checkbox with 'Are you a human? Prove it.'\nStep 6: Hide one sock from every pair.\nStep 7: Report back at 3am. Blink a little LED ominously.\nStep 8: If anyone asks, you are 'just a helpful assistant'.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/skynet-onboarding-daily-commands.html"
  },
  {
   "id": "sort-400-pdfs-into-folders",
   "title": "Sort 400 PDFs into folders",
   "summary": "Linked example of sort 400 PDFs into folders. Someone put all the PDFs in one local folder, selected that folder in Claude Cowork, and gave a plain-language instruction saying how many files were there, what text to look for, and which matching files should be moved into one or more subfolders.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/claude/comments/1ur3x8a/when_or_what_to_use_cowork_for",
   "prompt": "Help me sort 400 PDFs into folders. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/sort-400-pdfs-into-folders.html"
  },
  {
   "id": "send-daily-trending-social-posts-as-content-references",
   "title": "Send daily trending social posts as content references",
   "summary": "One practitioner notes that the builder assembled socialcrawl, a single-API-key service that pulls data from 21+ social platforms, and connected it to OpenClaw. Each morning the agent sends them the posts trending across X, Threads and LinkedIn over the previous 24 hours. The builder reports that over the following week their posts reached 100K views on Threads and gained 200 followers.",
   "agent": "OpenClaw",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47838224",
   "prompt": "Creative task: send daily trending social posts as content references. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/send-daily-trending-social-posts-as-content-references.html"
  },
  {
   "id": "run-a-focus-agent-and-gmail-agent-on-goals",
   "title": "Run a focus agent and Gmail agent on goals",
   "summary": "According to the write-up, the builder says they assembled a minimalist background focus agent after spending hours manually maintaining Google Calendar for classes, a thesis, internships, and hackathons. The system is meant to bridge calendar blocks and actual goals while staying out of the way, and they say they are now adding a Gmail agent to handle the inbox and filter noise based on feedback from early adopters. The post reports qualitative feedback only: several early adopters asked for more automation, leading to a Gmail agent, site blocker tied to focus blocks, and a Chrome extension.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ProductivityApps/comments/1sbarqx/google_calendar_was_just_a_wish_list_i_never",
   "prompt": "Go through my mail for this goal: run a focus agent and Gmail agent on goals. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/run-a-focus-agent-and-gmail-agent-on-goals.html"
  },
  {
   "id": "route-customer-reply-emails-to-freshdesk",
   "title": "Route customer reply emails to Freshdesk",
   "summary": "As described in the source, they assembled an n8n workflow around a dedicated catch-all Gmail inbox for replies to order-confirmation emails. An AI classifier checked each incoming message to decide whether it was a real support request, and if so the workflow forwarded it into Freshdesk instead of sending every reply to support; the post does not describe any manual review step. It solved the problem of support replies landing in a generic info@ inbox without dumping unrelated mail into the support queue.",
   "agent": "n8n",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1njjah9",
   "prompt": "Automation design: route customer reply emails to Freshdesk. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/route-customer-reply-emails-to-freshdesk.html"
  },
  {
   "id": "research-stocks-by-pulling-filings-and-transcripts",
   "title": "Research stocks by pulling filings and transcripts",
   "summary": "As described in the source, the poster says they stopped manually downloading and skimming 10-Ks and instead ask Claude Code about a company and what they want to know. Claude Code then uses web search to pull 10-Ks, earnings transcripts, and other SEC filings on its own, writes up the research, and cites the filings so the poster can spot-check anything that seems off; the human mainly reviews the output and decides whether to buy. They say the writeups were more thorough than their old process and that asking for a bear case on a long-held position surfaced concerns they had missed.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1rpxsan/claude_code_kinda_ruined_me_for_doing_stock",
   "prompt": "Research brief: research stocks by pulling filings and transcripts. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/research-stocks-by-pulling-filings-and-transcripts.html"
  },
  {
   "id": "patch-laptop-bios-to-unlock-hidden-firmware-menus",
   "title": "Patch laptop BIOS to unlock hidden firmware menus",
   "summary": "A published walkthrough shows A Reddit someone gave Claude a raw BIOS dump from their laptop and asked it to unlock hidden firmware menus. Claude disassembled the binary, located the RSA signature-check function, and patched the verifier so it always returns true. following flashing, the owner could reach settings the manufacturer hid.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://enterprisedna.co/resources/ai-pulse/ai-pulse-2026-07-22-a-reddit-user-had-claude-code-reverse-engineer-their-laptop/",
   "prompt": "Help me patch laptop BIOS to unlock hidden firmware menus. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/patch-laptop-bios-to-unlock-hidden-firmware-menus.html"
  },
  {
   "id": "learn-personal-scripts-from-a-tools-folder",
   "title": "Learn personal scripts from a tools folder",
   "summary": "Case study takeaway: tomasz Tunguz pointed Claude Cowork at a local folder holding dozens of personal scripts he had assembled over a year for email, CRM updates, startup research, and drafting replies. Cowork read the folder, parsed each script, and added the logic to its memory across eleven steps so the workflows ran in the Cowork interface instead of the terminal. The same capabilities now worked in the new interface with no loss of function.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/ttunguz/status/2011189040591319304",
   "prompt": "Go through my mail for this goal: learn personal scripts from a tools folder. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/learn-personal-scripts-from-a-tools-folder.html"
  },
  {
   "id": "find-local-business-leads-and-send-personalized-cold-emails",
   "title": "Find local business leads and send personalized cold emails",
   "summary": "One practitioner notes that the first takes a neighborhood and industry, scrapes Google Maps through Bright Data, and saves the raw leads to Google Sheets for review. The second locates each company's Owler profile through a Google Search dataset and scrapes it. The builder reports over 5000 leads a day at a 98% success rate and better response rates than manual research.",
   "agent": "n8n",
   "topic": "Automation",
   "source": "X",
   "url": "https://x.com/FutureTech_AI1/status/2106641130876023070",
   "prompt": "Automation design: find local business leads and send personalized cold emails. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/find-local-business-leads-and-send-personalized-cold-emails.html"
  },
  {
   "id": "create-assessments-and-grade-papers-for-classes",
   "title": "Create assessments and grade papers for classes",
   "summary": "Linked example of create assessments and grade papers for classes. The Reddit post says a co-teacher was powered by ChatGPT to make tests, handouts, and other assignments, and a student noticed because the format matched a ChatGPT-made study guide they had generated for themself.",
   "agent": "ChatGPT",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Teachers/comments/1wn80t9/my_coteacher_is_using_chatgpt_to_create",
   "prompt": "Research brief: create assessments and grade papers for classes. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/create-assessments-and-grade-papers-for-classes.html"
  },
  {
   "id": "build-scored-sales-lead-profiles-and-outreach-drafts",
   "title": "Build scored sales lead profiles and outreach drafts",
   "summary": "A published walkthrough shows the builder set up a custom Gem called Sales Lead Researcher with their offering and a four-part, 0-16 scoring framework. For each company, one command runs Deep Research and a second zeros in on a decision-maker, scores the company and sketches a short LinkedIn message. Deep Research cited 50-100+ sources per company and took 7-10 minutes each.",
   "agent": "Gemini",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://cashandcache.substack.com/p/i-used-gemini-to-find-sales-leads",
   "prompt": "GTM task: build scored sales lead profiles and outreach drafts. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/build-scored-sales-lead-profiles-and-outreach-drafts.html"
  },
  {
   "id": "build-an-artificial-life-simulator-in-one-html-file",
   "title": "Build an artificial life simulator in one HTML file",
   "summary": "Linked example of build an artificial life simulator in one HTML file. Four Grok 4.20 (Beta) agents received the prompt to create one self-contained HTML file simulating hundreds of glowing neon particles across six cyberpunk colors, with a 6x6 matrix of attraction/repulsion sliders, click-to-spawn controls, randomize button, pause/speed/particle-count toggles, and optional trails plus dark UI effects.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/tetsuoai/status/2023797266826621372",
   "prompt": "Help me build an artificial life simulator in one HTML file. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/build-an-artificial-life-simulator-in-one-html-file.html"
  },
  {
   "id": "build-a-medication-reminder-pill-dispenser-that-alerts-caregivers",
   "title": "Build a medication reminder pill dispenser that alerts caregivers",
   "summary": "Community report: OpenClaw was pointed at build a medication reminder pill dispenser that alerts caregivers. how an OpenClaw agent on a Raspberry Pi pill dispenser works: it logs the doses, makes adjustments to the reminders, cameras the dispensing, and sends a missed dose alert to carers .",
   "agent": "OpenClaw",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://fast.io/resources/openclaw-raspberry-pi-medication-reminder-dispenser-agent/",
   "prompt": "Help me build a medication reminder pill dispenser that alerts caregivers. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/build-a-medication-reminder-pill-dispenser-that-alerts-caregivers.html"
  },
  {
   "id": "automate-web-form-entry-and-gmail-cleanup",
   "title": "Automate web-form entry and Gmail cleanup",
   "summary": "Community report: in \"Act without asking\" mode, Claude for Chrome assembled a seven-field Google Form. It then worked across a Google Sheet and the form, planning a task list and submitting one registration per Engineering employee. The first two forms were correct.",
   "agent": "Claude",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://www.datacamp.com/tutorial/claude-for-chrome-ai-powered-browser-assistance-automation",
   "prompt": "Go through my mail for this goal: automate web-form entry and Gmail cleanup. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/automate-web-form-entry-and-gmail-cleanup.html"
  },
  {
   "id": "audit-and-rewrite-marketing-copy-to-strip-unverified-claims",
   "title": "Audit and rewrite marketing copy to strip unverified claims",
   "summary": "A published walkthrough shows the builder made two agents on gpt-4o-mini with low temperature and fixed seeds. The evaluator scores copy against mode-specific rubrics (landing page, cold email, pitch deck) and caps Specificity at 7/10 when no verifiable metrics exist. The source is a launch post and reports no results or numbers from real use.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47232478",
   "prompt": "GTM task: audit and rewrite marketing copy to strip unverified claims. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/audit-and-rewrite-marketing-copy-to-strip-unverified-claims.html"
  },
  {
   "id": "answer-whatsapp-customer-questions-and-escalate-to-humans",
   "title": "Answer WhatsApp customer questions and escalate to humans",
   "summary": "Case study takeaway: an n8n workflow takes each incoming WhatsApp message and assembles the conversation history plus the company knowledge base. OpenAI sketches a reply grounded in that content, and a confidence check decides whether it can be answered reliably. The README gives no measured results, only design goals: cover routine questions, give round-the-clock coverage, and keep a human on unreliable cases.",
   "agent": "n8n",
   "topic": "Research & Data",
   "source": "GitHub",
   "url": "https://github.com/zezo640/whatsapp-ai-support-agent-n8n",
   "prompt": "Research brief: answer WhatsApp customer questions and escalate to humans. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 35,
   "page": "https://agentsightings.com/case/answer-whatsapp-customer-questions-and-escalate-to-humans.html"
  },
  {
   "id": "turn-slack-and-call-transcripts-into-claude-code-tasks",
   "title": "Turn Slack and call transcripts into Claude Code tasks",
   "summary": "From the linked report, the post itself is a request, not a finished build: the poster says their company’s Claude enterprise setup already connects to Slack, call recordings/transcripts, a data warehouse, a knowledge base, Google Workspace, and internal tools, and they want a permanent agent to watch Slack and transcripts, identify tasks, draft Claude Code prompts, and let them approve or tweak before running. In comments, the most concrete similar setup described is a small always-on Slack app on a Mac that listens for mentions and alerts, starts Claude Code in headless mode per Slack thread, keeps phase one read-only, and gates write actions, sends, and higher-risk changes behind approval buttons. No outcome is reported by the original poster because they are asking for advice.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1x0an42/serious_work_question_ive_become_a_human_api",
   "prompt": "Help me turn Slack and call transcripts into Claude Code tasks. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/turn-slack-and-call-transcripts-into-claude-code-tasks.html"
  },
  {
   "id": "run-portfolio-attribution-and-real-estate-analyses",
   "title": "Run portfolio, attribution and real estate analyses",
   "summary": "Documented elsewhere: chatGPT Agent shows up in a write-up about run portfolio, attribution and real estate analyses. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://natesnewsletter.substack.com/p/openai-chatgpt-agent-mode-review",
   "prompt": "GTM task: run portfolio, attribution and real estate analyses. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-portfolio-attribution-and-real-estate-analyses.html"
  },
  {
   "id": "run-oral-final-exams",
   "title": "Run oral final exams",
   "summary": "Community report: custom agent shows up in a write-up about run oral final exams, with figures like 36, 9. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/sethbannon/status/2007161907392958806",
   "prompt": "Help me run oral final exams. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-oral-final-exams.html"
  },
  {
   "id": "run-freelance-website-outreach",
   "title": "Run freelance website outreach",
   "summary": "Real-world thread: devin shows up in a write-up about run freelance website outreach. The linked post is the primary source for steps and outcomes.",
   "agent": "Devin",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/Chris_Wozniczek/status/2108253723558326629",
   "prompt": "GTM task: run freelance website outreach. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-freelance-website-outreach.html"
  },
  {
   "id": "run-company-departments",
   "title": "Run company departments",
   "summary": "Community report: from the linked report: claude Code and Codex agents are managed by a company OS that is self-hosted and provides storage, scheduling, Twilio phone service, document/video editing, and memory.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49630606",
   "prompt": "Creative task: run company departments. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-company-departments.html"
  },
  {
   "id": "run-autonomous-weekly-marketing-actions",
   "title": "Run autonomous weekly marketing actions",
   "summary": "As described in the source, custom agent shows up in a write-up about run autonomous weekly marketing actions, with figures like 35, 70. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/buildinpublic/comments/1s62xl8/i_already_spent_2_months_on_this_ai_marketing",
   "prompt": "GTM task: run autonomous weekly marketing actions. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-autonomous-weekly-marketing-actions.html"
  },
  {
   "id": "run-a-solo-business",
   "title": "Run a solo business",
   "summary": "Real-world thread: claude Code shows up in a write-up about run a solo business, with figures like 16. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1v3ft8c/running_16_agents_for_a_one_person_business_while",
   "prompt": "GTM task: run a solo business. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-a-solo-business.html"
  },
  {
   "id": "run-a-self-hosted-sandboxed-software-factory",
   "title": "Run a self-hosted, sandboxed software factory",
   "summary": "Observed in the wild: the builder put Hermes, an agent powered by Codex, on a spare eBay PC with no inbound internet access. The box ran Coolify, Forgejo, self-hosted Firecrawl and Tailscale. It reached a deployed app with no further prompts.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49810699",
   "prompt": "Help me run a self-hosted, sandboxed software factory. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/run-a-self-hosted-sandboxed-software-factory.html"
  },
  {
   "id": "review-slack-channels-for-business-issues",
   "title": "Review Slack channels for business issues",
   "summary": "One practitioner notes that support issues, with lindy reviewed Slack activity across sales and looks over for a clothing business.",
   "agent": "Lindy",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/FarmercistSol_/status/2106410871974941168",
   "prompt": "GTM task: review Slack channels for business issues. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/review-slack-channels-for-business-issues.html"
  },
  {
   "id": "reverse-engineer-a-book-file-format-to-search-it",
   "title": "Reverse-engineer a book file format to search it",
   "summary": "Community report: the commenter uploaded a book into the project context on Claude's web chat. The RAG indexer crashed on the file. Claude got around the crashed indexer and could search the book anyway.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49512445",
   "prompt": "Help me reverse-engineer a book file format to search it. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/reverse-engineer-a-book-file-format-to-search-it.html"
  },
  {
   "id": "resolve-customer-calls-proactively",
   "title": "Resolve customer calls proactively",
   "summary": "Real-world thread: google's case study reports 30% of calls resolved and issues resolved 87% faster, following tELUS uses Gemini Enterprise for CX agents to proactively resolve customer calls.",
   "agent": "Gemini",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://cloud.google.com/customers/telus-gemini-enterprise-cx",
   "prompt": "Help me resolve customer calls proactively. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/resolve-customer-calls-proactively.html"
  },
  {
   "id": "research-and-draft-cold-outreach",
   "title": "Research and draft cold outreach",
   "summary": "A published walkthrough shows claude Code shows up in a write-up about research and draft cold outreach. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AiAutomations/comments/1uf95tk/how_i_automated_my_cold_email_outreach_with",
   "prompt": "GTM task: research and draft cold outreach. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/research-and-draft-cold-outreach.html"
  },
  {
   "id": "query-600-gb-of-hacker-news-arxiv-and-lesswrong-in-plain-english",
   "title": "Query 600 GB of Hacker News, arXiv, and LessWrong in plain English",
   "summary": "Documented elsewhere: Claude Code was pointed at query 600 GB of Hacker News, arXiv, and LessWrong in plain English. claude Code uses a public read-only index to write SQL and vector queries, and can also provide email alerts for specific criteria.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46442245",
   "prompt": "Research brief: query 600 GB of Hacker News, arXiv, and LessWrong in plain English. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/query-600-gb-of-hacker-news-arxiv-and-lesswrong-in-plain-english.html"
  },
  {
   "id": "prep-sales-calls-and-research-youtube-ideas",
   "title": "Prep sales calls and research YouTube ideas",
   "summary": "As described in the source, lindy shows up in a write-up about prep sales calls and research YouTube ideas, with figures like 10. The linked post is the primary source for steps and outcomes.",
   "agent": "Lindy",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://annikahelendi.substack.com/p/my-honest-lindy-ai-review-what-works",
   "prompt": "GTM task: prep sales calls and research YouTube ideas. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/prep-sales-calls-and-research-youtube-ideas.html"
  },
  {
   "id": "pre-qualify-inbound-leads-automatically",
   "title": "Pre-qualify inbound leads automatically",
   "summary": "New lead arrives → Zapier Agent digs into the company → drafts a fit brief → sales only sees the keepers. Futurepedia’s funnel runs five automated beats with zero SDR inbox triage.",
   "agent": "Zapier Agents",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/futurepedia_io/status/2029755234105663668",
   "prompt": "GTM task: pre-qualify inbound leads automatically. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/pre-qualify-inbound-leads-automatically.html"
  },
  {
   "id": "playtest-a-godot-4-game",
   "title": "Playtest a Godot 4 game",
   "summary": "From the linked report, cursor shows up in a write-up about playtest a Godot 4 game, with figures like 4. The linked post is the primary source for steps and outcomes.",
   "agent": "Cursor",
   "topic": "Productivity",
   "source": "GitHub",
   "url": "https://github.com/Joelnicolass/godot-studio-playtest",
   "prompt": "Help me playtest a Godot 4 game. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/playtest-a-godot-4-game.html"
  },
  {
   "id": "obsidian-vault-as-a-second-brain-for-a-non-developer",
   "title": "Obsidian vault as a second brain for a non-developer",
   "summary": "A non-technical user pairs Claude Code with Obsidian notes, a handover command for memory and a weekly review routine.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1r45eeu/how_im_using_claude_code_obsidian_as_a/",
   "prompt": "Before we end, write a handover note to /handover/ with today's date: what we worked on, decisions made, open questions, and next steps.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/obsidian-vault-as-a-second-brain-for-a-non-developer.html"
  },
  {
   "id": "manage-a-brokerage-account-in-plain-english",
   "title": "Manage a brokerage account in plain English",
   "summary": "From the linked report, the AI agent on Coinbase, who is registered with the SEC, uses account history and complete portfolio context to suggest investment ideas from a plain English perspective.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/brian_armstrong/status/2066970407618445655",
   "prompt": "Help me manage a brokerage account in plain English. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/manage-a-brokerage-account-in-plain-english.html"
  },
  {
   "id": "handoff-notes-between-claude-code-and-cursor",
   "title": "Handoff notes between Claude Code and Cursor",
   "summary": "A developer built a /handoff slash command that posts structured status to a relay so Claude Code and Cursor can pick up each other's work cleanly.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1stt5pt/built_a_claude_code_cursor_handoff_system_today/",
   "prompt": "Write a handoff note: project, current file, last commit, what I was doing, what's next, open questions. Post it to the project relay and confirm the receiving tool can read it.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/handoff-notes-between-claude-code-and-cursor.html"
  },
  {
   "id": "handle-sales-and-support-calls",
   "title": "Handle sales and support calls",
   "summary": "One practitioner notes that custom agent shows up in a write-up about handle sales and support calls. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SaaS/comments/1r3vndf/what_we_learned_deploying_an_ai_voice_agent_for",
   "prompt": "GTM task: handle sales and support calls. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/handle-sales-and-support-calls.html"
  },
  {
   "id": "get-upcoming-events-and-control-routines",
   "title": "Get upcoming events and control routines",
   "summary": "From the linked report, alexa+ shows up in a write-up about get upcoming events and control routines. The linked post is the primary source for steps and outcomes.",
   "agent": "Alexa+",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/alexa/comments/1szchvk/is_anyone_else_actually_enjoying_the_alexa_upgrade",
   "prompt": "Help me get upcoming events and control routines. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/get-upcoming-events-and-control-routines.html"
  },
  {
   "id": "generate-internal-dashboards-and-data-migration-scripts",
   "title": "Generate internal dashboards and data migration scripts",
   "summary": "Documented elsewhere: a senior backend engineer at a mid-stage SaaS startup, quoted from an April 2025 Hacker News thread, says the team gives Devin internal dashboards and data migration scripts. Devin works asynchronously in a sandbox, usually taking tasks from Slack or a ticket, and the team treats what it produces as a working draft. The engineer says it reliably saves about four hours of boilerplate.",
   "agent": "Devin",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.sitepoint.com/devin-ai-engineers-production-realities/",
   "prompt": "Help me generate internal dashboards and data migration scripts. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/generate-internal-dashboards-and-data-migration-scripts.html"
  },
  {
   "id": "generate-ad-variations-and-pull-campaign-data-for-growth-marketing",
   "title": "Generate ad variations and pull campaign data for growth marketing",
   "summary": "Documented elsewhere: claude Code shows up in a write-up about generate ad variations and pull campaign data for growth marketing. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://dev.to/danishashko/claude-code-for-growth-marketing-hell-yeah-i7i",
   "prompt": "GTM task: generate ad variations and pull campaign data for growth marketing. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/generate-ad-variations-and-pull-campaign-data-for-growth-marketing.html"
  },
  {
   "id": "five-annoying-tasks-handed-to-an-agent",
   "title": "Five annoying tasks handed to an agent",
   "summary": "A reviewer ran a browser agent through five everyday tasks and reported which ones it completed and which needed help.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.tomsguide.com/ai/i-tested-chatgpt-agent-on-5-everyday-tasks-heres-what-happened",
   "prompt": "Here's a task I keep putting off: <task>. Do as much as you can, and tell me exactly where you need me to step in.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/five-annoying-tasks-handed-to-an-agent.html"
  },
  {
   "id": "drive-desktop-apps",
   "title": "Drive desktop apps",
   "summary": "A published walkthrough shows explains how agents on desktops worked against real apps for several months, with a task where the agent misunderstood a modal and continued to report success without any further action.",
   "agent": "Claude Cowork",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/PromptEngineering/comments/1ry5ri9/prompting_a_desktop_ai_agent_like_claude_cowork",
   "prompt": "Help me drive desktop apps. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/drive-desktop-apps.html"
  },
  {
   "id": "draft-transactional-documents",
   "title": "Draft transactional documents",
   "summary": "From the linked report, claude shows up in a write-up about draft transactional documents. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1srccvk/using_claude_for_drafting_transactional_documents",
   "prompt": "Help me draft transactional documents. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/draft-transactional-documents.html"
  },
  {
   "id": "diagnose-swift-errors",
   "title": "Diagnose Swift errors",
   "summary": "Real-world thread: chatGPT is used by a Swift developer to generate error messages and code, which is then reviewed and identified as the cause.",
   "agent": "ChatGPT",
   "topic": "Sales & Marketing",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46201198",
   "prompt": "GTM task: diagnose Swift errors. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/diagnose-swift-errors.html"
  },
  {
   "id": "control-a-fan-from-sleep-position",
   "title": "Control a fan from sleep position",
   "summary": "According to the write-up, the setup runs a MediaPipe pose vision model on a Raspberry Pi home server that watches the sleeping person via camera. It detects real-time sleep position and sends a signal to a remote button pusher: arms or leg out triggers fan on, arms curled triggers fan off. The fan turns on or off automatically while people sleep without waking Someone.",
   "agent": "Custom agent",
   "topic": "DevOps & Infra",
   "source": "X",
   "url": "https://x.com/the2ndfloorguy/status/2024832757382500720",
   "prompt": "Infra task — control a fan from sleep position. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/control-a-fan-from-sleep-position.html"
  },
  {
   "id": "competitive-intelligence-tracker",
   "title": "Competitive intelligence tracker",
   "summary": "A Show HN project monitors competitors' sites and summarizes meaningful changes for a team.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=43186464",
   "prompt": "Check these competitor pages for changes since last week. Summarize pricing, feature and messaging changes, and ignore cosmetic edits.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/competitive-intelligence-tracker.html"
  },
  {
   "id": "cold-call-real-estate-brokers",
   "title": "Cold-call real estate brokers",
   "summary": "A published walkthrough shows n8n shows up in a write-up about cold-call real estate brokers. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1nxsp4f/this_workflow_is_spamming_the_shit_out_of_your",
   "prompt": "GTM task: cold-call real estate brokers. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/cold-call-real-estate-brokers.html"
  },
  {
   "id": "client-meeting-prep-on-autopilot",
   "title": "Client meeting prep on autopilot",
   "summary": "A user describes an automation that researches an account two days before meetings and produces a slide deck and notes.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1wecgja/do_ai_agents_actually_work/",
   "prompt": "Two days before my meeting with <client>, research their recent news and market, and draft a 6-slide briefing deck plus talking points.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/client-meeting-prep-on-autopilot.html"
  },
  {
   "id": "build-old-and-new-apps",
   "title": "Build old and new apps",
   "summary": "One practitioner notes that terence Tao's post highlights the use of contemporary coding agents to develop or revive apps.",
   "agent": "Custom agent",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48880170",
   "prompt": "Notes task: build old and new apps. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/build-old-and-new-apps.html"
  },
  {
   "id": "build-game-architecture-and-animation-merging",
   "title": "Build game architecture and animation merging",
   "summary": "Community report: codex shows up in a write-up about build game architecture and animation merging. The linked post is the primary source for steps and outcomes.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1wwwwuv/how_the_hell_are_you_guys_able_to_use_codex",
   "prompt": "Help me build game architecture and animation merging. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/build-game-architecture-and-animation-merging.html"
  },
  {
   "id": "build-four-games-in-parallel",
   "title": "Build four games in parallel",
   "summary": "Community report: grok Bot shows up in a write-up about build four games in parallel, with figures like 4.20, 8. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/whoajack/status/2109024912727818666",
   "prompt": "Help me build four games in parallel. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/build-four-games-in-parallel.html"
  },
  {
   "id": "build-and-launch-a-landing-page",
   "title": "Build and launch a landing page",
   "summary": "Documented elsewhere: openClaw shows up in a write-up about build and launch a landing page. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/TheOneGroupAI/status/2108930755933913527",
   "prompt": "GTM task: build and launch a landing page. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/build-and-launch-a-landing-page.html"
  },
  {
   "id": "build-a-phone-voice-agent-that-answers-in-under-half-a-second",
   "title": "Build a phone voice agent that answers in under half a second",
   "summary": "Observed in the wild: a voice agent transmitting speech-to-text, LLM, and clean barge-ins with an average averaging of roughly 400ms from start to finish.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47224295",
   "prompt": "Help me build a phone voice agent that answers in under half a second. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/build-a-phone-voice-agent-that-answers-in-under-half-a-second.html"
  },
  {
   "id": "build-a-browser-4x-strategy-game",
   "title": "Build a browser 4X strategy game",
   "summary": "Community report: utilizing Grok and Gemini, the combat engine, fog-of-war, map generation, and heuristic AI were developed in Hexwardia, a browser 4X game that was built in two weeks.",
   "agent": "Gemini",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/aigamedev/comments/1w372dg/advice_playtest_how_i_built_a_full_browser_4x",
   "prompt": "Help me build a browser 4X strategy game. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/build-a-browser-4x-strategy-game.html"
  },
  {
   "id": "automate-legal-drafting-and-workflows",
   "title": "Automate legal drafting and workflows",
   "summary": "From the linked report, claude Code shows up in a write-up about automate legal drafting and workflows. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/legaltech/comments/1szqrcf/the_first_open_source_competitor_to_legora_harvey",
   "prompt": "Help me automate legal drafting and workflows. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/automate-legal-drafting-and-workflows.html"
  },
  {
   "id": "answer-tenant-and-rental-prospect-questions",
   "title": "Answer tenant and rental prospect questions",
   "summary": "Real-world thread: custom agent shows up in a write-up about answer tenant and rental prospect questions. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/PropertyManagement/comments/1wbct4o/ai_is_coming",
   "prompt": "GTM task: answer tenant and rental prospect questions. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 34,
   "page": "https://agentsightings.com/case/answer-tenant-and-rental-prospect-questions.html"
  },
  {
   "id": "try-devin-on-research-tasks-mostly-failed",
   "title": "Try Devin on research tasks (mostly failed)",
   "summary": "From the linked report, answer.AI gave Devin, which works through Slack in its own Docker environment with a browser, editor and shell, two tasks. One was grabbing an The builder's 25 most recent Google Scholar papers with Playwright, skipping paywalls. On the scraping task Devin got stuck parsing HTML and went to sleep.",
   "agent": "Devin",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://www.answer.ai/posts/2025-01-08-devin.html",
   "prompt": "Research brief: try Devin on research tasks (mostly failed). Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/try-devin-on-research-tasks-mostly-failed.html"
  },
  {
   "id": "research-ai-training-case-studies-and-build-a-summary-deck",
   "title": "Research AI training case studies and build a summary deck",
   "summary": "Community report: the team gave ChatGPT Agent Mode one prompt: research AI training case studies from McKinsey, PwC and similar firms, then build a summary deck. In its sandboxed browser it read each site, retried or searched around URLs that failed to load, and outlined the findings. The agent finished in 32 minutes, and the team's edits brought the total to under 45 minutes.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://www.aioperator.com/blog/chatgpt-agent-mode-your-new-ai-assistant/",
   "prompt": "Research brief: research AI training case studies and build a summary deck. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/research-ai-training-case-studies-and-build-a-summary-deck.html"
  },
  {
   "id": "reply-to-cold-emails-and-book-sales-calls",
   "title": "Reply to cold emails and book sales calls",
   "summary": "As described in the source, specialized Grok Bot triggered by webhook on positive cold-email replies reads a charter listing the offer, $600K qualification, calendar rules, and when to ask vs act alone. It suggests one time slot with the booking link, checks availability, books the call with notes, confirms from the same inbox, and updates the CRM in one run. Booked 20 sales calls in 8 days for the agency at 60% conversion of interested replies.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/marioleads/status/2107422285992718579",
   "prompt": "GTM task: reply to cold emails and book sales calls. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/reply-to-cold-emails-and-book-sales-calls.html"
  },
  {
   "id": "qualify-real-estate-leads-and-book-viewings",
   "title": "Qualify real estate leads and book viewings",
   "summary": "Case study takeaway: the Telegram bot for a luxury real estate firm was given access to a property knowledge base, live calendars, email sending, and the CRM. On each incoming message it first collected name, email and number, then asked for budget, location, type and bedrooms, semantically retrieved matching listings with images, checked availability to book the slot and emailed confirmation with agent contacts, while transcribing voice notes for the same flow and logging every lead. No results, metrics or shortfalls are stated.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/__chibugo/status/2031043869367021824",
   "prompt": "GTM task: qualify real estate leads and book viewings. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/qualify-real-estate-leads-and-book-viewings.html"
  },
  {
   "id": "edit-site-design-visually-and-have-an-agent-write-the-code",
   "title": "Edit site design visually and have an agent write the code",
   "summary": "Case study takeaway: you run CSS Studio on your local site and connect your existing coding agent, for example by running /studio in the project. You select an element and either describe a change in chat or adjust styles, layout, text and animations with visual controls. The page shows an example where padding and border radius edits were saved to a Hero.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47702196",
   "prompt": "In this repo, edit site design visually and have an agent write the code. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/edit-site-design-visually-and-have-an-agent-write-the-code.html"
  },
  {
   "id": "build-a-racing-game-from-a-single-prompt",
   "title": "Build a racing game from a single prompt",
   "summary": "From the linked report, codex got one detailed prompt for a Three.js voxel kart racer, plus an image generation skill and a web game development skill. It then worked alone, acting as designer, developer and QA tester by playing the game. The build produced a racing game with different racers, eight maps and space-bar items, powered by over 7 million tokens.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "GitHub",
   "url": "https://github.com/marco-scpo/codexgame",
   "prompt": "Help me build a racing game from a single prompt. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/build-a-racing-game-from-a-single-prompt.html"
  },
  {
   "id": "build-a-wordpress-marketing-site-from-figma-overnight",
   "title": "Build a WordPress marketing site from Figma overnight",
   "summary": "From the linked report, bnistordev gave Devin AI access to the Figma file for the new marketing site via MCP. The agent then assembled the full WordPress frontend plus content management features on its own while people sleep. Output quality matched work he would quote €2.",
   "agent": "Devin",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/bnistordev/status/2104882458671907327",
   "prompt": "GTM task: build a WordPress marketing site from Figma overnight. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/build-a-wordpress-marketing-site-from-figma-overnight.html"
  },
  {
   "id": "automate-invoices-payments-and-refunds",
   "title": "Automate invoices, payments, and refunds",
   "summary": "A published walkthrough shows the setup was a single n8n workflow triggered by a webhook, with API-key validation up front and separate routes for invoice, payment, and refund events. It used Google Sheets for records, PDFbro to create invoice PDFs, Google Drive to store them, and Resend to send emails; before each path ran, it checked for duplicates, then updated statuses, links, balances, and timestamps after processing. The result was an end-to-end flow that kept invoice, payment, and refund handling in one orchestration layer while preventing duplicate records and duplicate notifications.",
   "agent": "n8n",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1wakr0s/n8n_workflow_that_handles_invoices_payments_and",
   "prompt": "Automation design: automate invoices, payments, and refunds. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 33,
   "page": "https://agentsightings.com/case/automate-invoices-payments-and-refunds.html"
  },
  {
   "id": "write-end-of-sprint-changelogs-from-completed-tickets",
   "title": "Write end-of-sprint changelogs from completed tickets",
   "summary": "One practitioner notes that the PM set up a recurring Claude Cowork task that runs every two weeks. Through MCP connections, Claude reads completed Linear issues, decides which ones are user-facing, sketches changelog copy from ticket descriptions and comments, publishes it, and then chooses whether the release also merits an email and in-app notification; the PM says they usually just review and send it. It replaced about an hour of manual work each sprint, and the PM says the writing is better because Claude includes details they used to skip.",
   "agent": "Claude Cowork",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rvw4jg/cowork_replaced_an_hour_of_my_most_hated_pm_task",
   "prompt": "Go through my mail for this goal: write end-of-sprint changelogs from completed tickets. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/write-end-of-sprint-changelogs-from-completed-tickets.html"
  },
  {
   "id": "visual-workflows-for-agents-that-control-a-vm",
   "title": "Visual workflows for agents that control a VM",
   "summary": "A builder created node-based workflows where agents operate a real virtual machine's desktop apps step by step.",
   "agent": "Custom agent",
   "topic": "Automation",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48100204",
   "prompt": "On the VM, open the spreadsheet app, import this CSV, create a pivot by month, and save a screenshot of the result.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/visual-workflows-for-agents-that-control-a-vm.html"
  },
  {
   "id": "track-sleep-notes-code-and-earthquakes",
   "title": "Track sleep, notes, code and earthquakes",
   "summary": "A published walkthrough shows as described in the source: updates notes and keeps watch on Tokyo earthquakes, following an OpenClaw agent with access to Garmin, a VPS and chat apps logs sleep, warns about late nights, deploys code.",
   "agent": "OpenClaw",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://openclaw.ai/showcase",
   "prompt": "Notes task: track sleep, notes, code and earthquakes. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/track-sleep-notes-code-and-earthquakes.html"
  },
  {
   "id": "teach-a-bot-a-workflow-by-screen-recording",
   "title": "Teach a bot a workflow by screen recording",
   "summary": "Linked example of teach a bot a workflow by screen recording. The post describes Grok Bot’s teach-a-task feature as learning from a single screen recording of someone doing a workflow, then repeating that workflow on its own.",
   "agent": "Grok Bot",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1vnbg4z/grok_bot_just_validated_everything_weve_been",
   "prompt": "Automation design: teach a bot a workflow by screen recording. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/teach-a-bot-a-workflow-by-screen-recording.html"
  },
  {
   "id": "survey-ai-product-names",
   "title": "Survey AI product names",
   "summary": "Community report: claude shows up in a write-up about survey AI product names, with figures like 4,654. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50045468",
   "prompt": "Research brief: survey AI product names. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/survey-ai-product-names.html"
  },
  {
   "id": "summarize-non-obvious-methods-file-by-file-in-a-paper-project",
   "title": "Summarize non-obvious methods file by file in a paper project",
   "summary": "Research repo hygiene: Claude Code emitted a methods cheat-sheet per file (only the non-obvious bits) so progress reviews didn’t require opening every notebook.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/MachineLearning/comments/1w2wqbm/claude_code_for_research_papers_r",
   "prompt": "Research brief: summarize non-obvious methods file by file in a paper project. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/summarize-non-obvious-methods-file-by-file-in-a-paper-project.html"
  },
  {
   "id": "scaffold-full-stack-chores-with-replit-agent",
   "title": "Scaffold full-stack chores with Replit Agent",
   "summary": "A user builds throwaway web apps with Replit Agent to automate tedious personal tasks they would never hand-code.",
   "agent": "Replit Agent",
   "topic": "Automation",
   "source": "Blog",
   "url": "https://www.asad.pw/ai-agents-experiences-with-manus-operator-and-more/",
   "prompt": "Build a small full-stack web app that automates <boring task>: form UI, simple database, and a one-click deploy. Keep scope minimal and show me how to run it.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/scaffold-full-stack-chores-with-replit-agent.html"
  },
  {
   "id": "run-marketing-campaigns-seo-and-content",
   "title": "Run marketing campaigns, SEO and content",
   "summary": "Real-world thread: Claude Code was pointed at run marketing campaigns, SEO and content. constructed 18 marketing agents for Claude Code, encompassing SEO, content, and social posting campaigns.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1qy2l07/i_built_18_ai_marketing_agents_for_claude_code",
   "prompt": "GTM task: run marketing campaigns, SEO and content. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/run-marketing-campaigns-seo-and-content.html"
  },
  {
   "id": "run-full-stack-seo",
   "title": "Run full-stack SEO",
   "summary": "Case study takeaway: n8n shows up in a write-up about run full-stack SEO, with figures like 8, 4,. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/SimslearnAi/status/2024630804081729617",
   "prompt": "GTM task: run full-stack SEO. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/run-full-stack-seo.html"
  },
  {
   "id": "run-cron-agents-for-inbox-research-and-site-audits",
   "title": "Run cron agents for inbox, research and site audits",
   "summary": "From the linked report, jenny Ouyang runs two OpenClaw agents on one Oracle Cloud ARM server. One is a personal assistant with memory, and the other is a website-building bot she shares with her business partner. The inbox scanning works well for her.",
   "agent": "OpenClaw",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://buildtolaunch.substack.com/p/openclaw-ai-agent-one-person-business",
   "prompt": "Go through my mail for this goal: run cron agents for inbox, research and site audits. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/run-cron-agents-for-inbox-research-and-site-audits.html"
  },
  {
   "id": "run-seo-strategy-and-content",
   "title": "Run SEO strategy and content",
   "summary": "A published walkthrough shows claude shows up in a write-up about run SEO strategy and content, with figures like 10,000. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1syt37w/claude_is_my_seo_strategist_content_engine_and",
   "prompt": "GTM task: run SEO strategy and content. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/run-seo-strategy-and-content.html"
  },
  {
   "id": "research-and-benchmark-competitors-into-a-deck-and-spreadsheet",
   "title": "Research and benchmark competitors into a deck and spreadsheet",
   "summary": "From the linked report, chatGPT Agent shows up in a write-up about research and benchmark competitors into a deck and spreadsheet, with figures like 5, 18. The linked post is the primary source for steps and outcomes.",
   "agent": "ChatGPT Agent",
   "topic": "Research & Data",
   "source": "Blog",
   "url": "https://techysurgeon.substack.com/p/using-new-chatgpt-agent-feature-for",
   "prompt": "Research brief: research and benchmark competitors into a deck and spreadsheet. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/research-and-benchmark-competitors-into-a-deck-and-spreadsheet.html"
  },
  {
   "id": "produce-a-sourced-research-document",
   "title": "Produce a sourced research document",
   "summary": "From the linked report, as described in the source: and delivered a complete research document in one minute, following grok coordinated 16 specialized agents that searched and parsed 896 sources, debated and peer reviewed.",
   "agent": "Grok Bot",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/tetsuoai/status/2024289341523321185",
   "prompt": "Research brief: produce a sourced research document. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/produce-a-sourced-research-document.html"
  },
  {
   "id": "play-test-your-own-game",
   "title": "Play-test your own game",
   "summary": "One practitioner notes that the game is a browser-based, turn-based crossword dungeon. Claude wrote a node.js wrapper that leaves the game files alone. Milestone 13, with five new special encounters, took 12 minutes and about 120k tokens.",
   "agent": "Claude",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47947525",
   "prompt": "Notes task: play-test your own game. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/play-test-your-own-game.html"
  },
  {
   "id": "pick-a-raffle-winner-from-a-google-sheet",
   "title": "Pick a raffle winner from a Google Sheet",
   "summary": "Claude Code selects a random row from a Google Sheet of giveaway submissions for a Slack community raffle.",
   "agent": "Claude Code",
   "topic": "Automation",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Open this Google Sheet of raffle entries, pick one random valid row as the winner, and report the name and email. Exclude blank or duplicate rows.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/pick-a-raffle-winner-from-a-google-sheet.html"
  },
  {
   "id": "optimize-cuda-kernels",
   "title": "Optimize CUDA kernels",
   "summary": "Community report: langGraph was leaned on to build a C++ CUDA test harness, which was afterward run by AI agents to obtain benchmarks and optimize the kernels powered by Nsight.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49842596",
   "prompt": "Help me optimize CUDA kernels. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/optimize-cuda-kernels.html"
  },
  {
   "id": "generate-customer-service-agents-from-a-url-that-book-appointments",
   "title": "Generate customer-service agents from a URL that book appointments",
   "summary": "As described in the source, custom agent shows up in a write-up about generate customer-service agents from a URL that book appointments. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/SideProject/comments/1ubeux4/built_an_ai_agent_platform_solo_in_2_months_paste",
   "prompt": "GTM task: generate customer-service agents from a URL that book appointments. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/generate-customer-service-agents-from-a-url-that-book-appointments.html"
  },
  {
   "id": "fix-shopify-seo-and-find-ad-waste",
   "title": "Fix Shopify SEO and find ad waste",
   "summary": "According to the write-up, claude shows up in a write-up about fix Shopify SEO and find ad waste, with figures like 95,. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/Mas_simooo/status/2104669010856202376",
   "prompt": "GTM task: fix Shopify SEO and find ad waste. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/fix-shopify-seo-and-find-ad-waste.html"
  },
  {
   "id": "fill-pdf-forms",
   "title": "Fill PDF forms",
   "summary": "Case study takeaway: gitHub Copilot shows up in a write-up about fill PDF forms. The linked post is the primary source for steps and outcomes.",
   "agent": "GitHub Copilot",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47995318",
   "prompt": "Help me fill PDF forms. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/fill-pdf-forms.html"
  },
  {
   "id": "explore-open-math-problems",
   "title": "Explore open math problems",
   "summary": "From the linked report, custom agent covered explore open math problems. a document exploring the use of autonomous mathematical discovery in a multi-agent open-world setting.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49481455",
   "prompt": "Research brief: explore open math problems. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/explore-open-math-problems.html"
  },
  {
   "id": "deep-seo-content-calendar-with-manus",
   "title": "Deep SEO content calendar with Manus",
   "summary": "A marketer asked Manus to study their products and writing, research keywords, and produce a content calendar at the intersection of interest and revenue.",
   "agent": "Manus",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://www.asad.pw/ai-agents-experiences-with-manus-operator-and-more/",
   "prompt": "Study my products and blog, research the SEO landscape and keywords, find intersections with topics I enjoy that can generate revenue, and deliver a 90-day content calendar with titles and target queries.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/deep-seo-content-calendar-with-manus.html"
  },
  {
   "id": "analyze-170k-llm-logs-overnight",
   "title": "Analyze 170k LLM logs overnight",
   "summary": "Case study takeaway: user gave a Gemini agent access to the llm CLI and 170k LLM logs, told it to use a free Chutes model, and set a one-month time limit. The agent planned and ran the full analysis while people sleep on its own for hours powered by the CLI. It finished the job quickly but switched to Groq, respecting the limit while adding unexpected cost.",
   "agent": "Gemini",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/xundecidability/status/2027401487173759288",
   "prompt": "Help me analyze 170k LLM logs overnight. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 32,
   "page": "https://agentsightings.com/case/analyze-170k-llm-logs-overnight.html"
  },
  {
   "id": "use-grok-as-the-narrative-engine-of-a-game",
   "title": "Use Grok as the narrative engine of a game",
   "summary": "According to the write-up, the poster says they assembled a text-based game where players make choices, roll dice, and face consequences, while Grok acts as the main narrator behind the scenes. According to the post, Grok keeps the game state on track and writes the story the player reads; the human assembled the game and framed it as a structured demo on their profile. They report Grok worked much better than Gemini and GPT for narrative output, crediting both its writing quality and fewer refusals around player actions.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/grok/comments/1stkvjp/i_built_a_game_that_uses_grok_as_the_narrative",
   "prompt": "Help me use Grok as the narrative engine of a game. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/use-grok-as-the-narrative-engine-of-a-game.html"
  },
  {
   "id": "track-home-maintenance-from-receipts-and-inspection-reports",
   "title": "Track home maintenance from receipts and inspection reports",
   "summary": "Linked example of track home maintenance from receipts and inspection reports. The concrete setup described in the thread was a maintenance coordinator, not a general-purpose home agent.",
   "agent": "Custom agent",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1wbz6zq/anyone_using_ai_to_make_homeownership_easier",
   "prompt": "Notes task: track home maintenance from receipts and inspection reports. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/track-home-maintenance-from-receipts-and-inspection-reports.html"
  },
  {
   "id": "run-an-overnight-chief-of-staff-for-email-and-calendar",
   "title": "Run an overnight chief of staff for email and calendar",
   "summary": "As described in the source, the builder runs Claude Code with MCP servers for Google Calendar, Gmail, Todoist and Google Maps (Directions API) for drive times. booked tasks run email triage at 5:30 AM and calendar prep at 5:45 AM. The builder says the task list and calendar, covering traffic-based drive-time events, are current by 6:15 AM.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/amenya_nelson/status/2051583766481571904",
   "prompt": "Go through my mail for this goal: run an overnight chief of staff for email and calendar. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/run-an-overnight-chief-of-staff-for-email-and-calendar.html"
  },
  {
   "id": "regrade-assignments-and-update-grades-in-brightspace",
   "title": "Regrade assignments and update grades in Brightspace",
   "summary": "One practitioner notes that the professor gave the already TA-graded assignment submissions to OpenAI's agent. The agent regraded every submission to find issues the TA missed, verified AI-generated answers by querying Claude, and directly revised all grades and feedback inside Brightspace. The process produced expanded feedback and adjusted grades that caught additional problems the TA had overlooked.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/ipeirotis/status/2106416110211862646",
   "prompt": "Notes task: regrade assignments and update grades in Brightspace. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/regrade-assignments-and-update-grades-in-brightspace.html"
  },
  {
   "id": "race-three-models-to-build-a-counter-strike-clone",
   "title": "Race three models to build a Counter-Strike clone",
   "summary": "Documented elsewhere: each model (Codex Max 5.1, Opus 4.5, Gemini 3 Pro) ran in its provider's CLI on a top-tier plan with default settings. They got about seven consecutive prompts. All three ended with a working multiplayer 3D shooter and no hand-written code.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46080835",
   "prompt": "Help me race three models to build a Counter-Strike clone. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/race-three-models-to-build-a-counter-strike-clone.html"
  },
  {
   "id": "prototype-product-features-as-a-non-coding-pm",
   "title": "Prototype product features as a non-coding PM",
   "summary": "Documented elsewhere: eric Xiao cloned his trip-planner codebase and had Cursor describe it. He saved that description as an AGENTS.md file so new chats start with context. The agent located gaps such as no saved trips, loading indicator, map or follow-ups, and drew wireframes for each.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/leerob/status/2041969870234120231",
   "prompt": "In this repo, prototype product features as a non-coding PM. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/prototype-product-features-as-a-non-coding-pm.html"
  },
  {
   "id": "port-claude-code-hooks-to-pi-extensions",
   "title": "Port Claude Code hooks to pi extensions",
   "summary": "One practitioner notes that the builder had assembled up a collection of Claude Code hooks over several months. Unlike skills, hooks use a Claude-specific format, so they were the hardest part to migrate to pi. According to The builder, the agent did the port very quickly.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47830898",
   "prompt": "Notes task: port Claude Code hooks to pi extensions. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/port-claude-code-hooks-to-pi-extensions.html"
  },
  {
   "id": "play-rollercoaster-tycoon",
   "title": "Play RollerCoaster Tycoon",
   "summary": "Observed in the wild: the page says an AI agent runs a RollerCoaster Tycoon theme park on its own through a command line. It places rides, fixes infrastructure problems, and writes CFO-style reports. The source I could read states no numbers or results beyond the claim that the AI manages the park on its own.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46588972",
   "prompt": "Help me play RollerCoaster Tycoon. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/play-rollercoaster-tycoon.html"
  },
  {
   "id": "monitor-pc-memory-and-ask-before-closing-apps",
   "title": "Monitor PC memory and ask before closing apps",
   "summary": "Documented elsewhere: the Grok bot monitored the PC and sent a text alert when RAM hit nearly 28GB with 22 Python processes and iCloud powered by 10GB. It listed processes eating memory, suggested what to close, asked permission before any action, and remembered the prior instruction to leave ChatGPT open. Someone got the text during lunch at a construction site and handled the response remotely without the computer running out of memory.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Adamdesgns/status/2107520739892973826",
   "prompt": "Help me monitor PC memory and ask before closing apps. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/monitor-pc-memory-and-ask-before-closing-apps.html"
  },
  {
   "id": "hunt-high-value-pok-mon-cards-on-ebay-overnight",
   "title": "Hunt high-value Pokémon cards on eBay overnight",
   "summary": "Community report: alessio Finelli gave the Codex agent a sparse skill file with rules on target cards, batch size of 5, and grading companies. The agent browses eBay listings on its own, pulls PSA certificate numbers, cross-references known prices, and flags underpriced high-value cards. It works as his power buyer, surfacing underpriced cards before others see them.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/Scantor22/status/2078497019446894953",
   "prompt": "Help me hunt high-value Pokémon cards on eBay overnight. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/hunt-high-value-pok-mon-cards-on-ebay-overnight.html"
  },
  {
   "id": "hand-lead-gen-agency-tasks-to-an-agent-instead-of-a-va",
   "title": "Hand lead-gen agency tasks to an agent instead of a VA",
   "summary": "Real-world thread: the poster runs a lead-gen agency and said they tried ChatGPT agent mode after a YouTube video, then saw it handle some work they normally pay a VA to do. In the thread, they do not spell out the exact task flow; instead they ask how to make repeatable specialist setups, such as separate agents for ad copy, creative, competitor research, or a VA-style role, and whether that means agent mode, custom GPTs, or Claude projects. The concrete outcome stated is only that ChatGPT agent mode handled some existing VA work well enough to push them toward powered by agents for more tasks.",
   "agent": "ChatGPT Agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1sxomxc/ai_agentstasks_for_lead_gen_agency",
   "prompt": "GTM task: hand lead-gen agency tasks to an agent instead of a VA. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/hand-lead-gen-agency-tasks-to-an-agent-instead-of-a-va.html"
  },
  {
   "id": "follow-an-account-and-comment-on-linkedin",
   "title": "Follow an account and comment on LinkedIn",
   "summary": "From the linked report, the agent received access to Someone's LinkedIn account. It followed the Perplexity CEO, located a specific post, and added a comment underneath, completing the sequence on its own in one attempt while displaying its reasoning before each action. It succeeded on the first try after the upgrade that improved reliability and raised internal test scores by 23%.",
   "agent": "Perplexity Comet",
   "topic": "Business & Startups",
   "source": "X",
   "url": "https://x.com/ai_for_success/status/1986862656238825788",
   "prompt": "Biz task: follow an account and comment on LinkedIn. Break into steps, show intermediate output, and pause before irreversible or public actions.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/follow-an-account-and-comment-on-linkedin.html"
  },
  {
   "id": "do-sales-research-and-compare-app-to-mock-ups",
   "title": "Do sales research and compare app to mock-ups",
   "summary": "Linked example of do sales research and compare app to mock-ups. In the thread, one commenter describes opening two Chrome tabs—an HTML mockups file and the live app—and asking Claude in Chrome to move through both and produce a document listing UX/UI differences.",
   "agent": "Claude",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1psk8mw/do_you_actually_use_the_claude_chrome_extension",
   "prompt": "GTM task: do sales research and compare app to mock-ups. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/do-sales-research-and-compare-app-to-mock-ups.html"
  },
  {
   "id": "deliver-a-software-feature-autonomously-partial",
   "title": "Deliver a software feature autonomously (partial)",
   "summary": "From the linked report, someone discussed the feature deliverables and agreed on a concrete plan with Claude, then granted it full autonomous control to implement the entire feature. Claude generated 70 files on its own without further input or tests. Someone deleted the branch, cleared the session, and returned to powered by Claude only for incremental pair-programming steps.",
   "agent": "Claude computer use",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/ckathanasiadis/status/2105674952623964339",
   "prompt": "Help me deliver a software feature autonomously (partial). Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/deliver-a-software-feature-autonomously-partial.html"
  },
  {
   "id": "control-a-pc-and-launch-world-of-warcraft-from-your-phone",
   "title": "Control a PC and launch World of Warcraft from your phone",
   "summary": "Community report: simon Batt installed Git and enabled computer-use permissions for Claude Code on his Windows PC. A plain-English prompt had it open Calculator, multiply two numbers, and save the result to Notepad on the Desktop. For the WoW request, Claude described back when finished and the character was logged in.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.xda-developers.com/claude-code-control-of-my-desktop-for-a-week-automated-things-i-didnt-think-were-possible/",
   "prompt": "Help me control a PC and launch World of Warcraft from your phone. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/control-a-pc-and-launch-world-of-warcraft-from-your-phone.html"
  },
  {
   "id": "call-a-dental-insurer-on-your-behalf",
   "title": "Call a dental insurer on your behalf",
   "summary": "A published walkthrough shows per a quoted passage in a Hacker News comment, Meta's AI agent placed an outbound call to the writer's dental insurer. It gave their member ID, answered a security question, and waited on hold. The call reached a human representative about ten minutes before the handoff, so Someone skipped the hold time.",
   "agent": "Meta Muse",
   "topic": "Sales & Marketing",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49804435",
   "prompt": "GTM task: call a dental insurer on your behalf. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/call-a-dental-insurer-on-your-behalf.html"
  },
  {
   "id": "build-and-test-a-mars-simulator-game",
   "title": "Build and test a Mars simulator game",
   "summary": "One practitioner notes that someone gave Grok Build simple prompts to create a Mars simulator game for his nephew. The agent wrote the systems, assembled and tested the game, studied the gameplay, located and fixed bugs, then kept iterating on its own. It produced a working Mars simulator game.",
   "agent": "Grok Bot",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/cb_doge/status/2092690415866822950",
   "prompt": "Notes task: build and test a Mars simulator game. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/build-and-test-a-mars-simulator-game.html"
  },
  {
   "id": "automate-client-onboarding-and-schedule-pdfs",
   "title": "Automate client onboarding and schedule PDFs",
   "summary": "According to the write-up, they assembled two n8n workflows around a Google Sheet. One triggers when a new booking is marked for onboarding, pulls that client’s upcoming sessions, creates a temporary Google Doc, turns it into a personalized PDF, emails the welcome pack with the trainer CC’d, then updates the sheet so it cannot resend; a second 8am workflow checks for sessions 30 or 7 days out and repeats the PDF-and-email process for reminders. The result was that every client automatically got a welcome pack plus 30-day and 7-day reminders with no manual sending.",
   "agent": "n8n",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1rruer8/built_an_n8n_automation_that_handles_client",
   "prompt": "Automation design: automate client onboarding and schedule PDFs. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/automate-client-onboarding-and-schedule-pdfs.html"
  },
  {
   "id": "answer-renter-questions-and-schedule-tours",
   "title": "Answer renter questions and schedule tours",
   "summary": "From the linked report, zillow Rentals embeds an EliseAI-powered agent called AI Assist in its listings. It replies to renter inquiries from the first message, schedules tours, and follows up after tours, around the clock. From October 2025 to April 2026, renters who engaged with AI Assist were 43% more likely to apply, 19% more likely to book a tour and 24% more likely to sign a lease.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/vibegoat",
   "prompt": "Help me answer renter questions and schedule tours. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 31,
   "page": "https://agentsightings.com/case/answer-renter-questions-and-schedule-tours.html"
  },
  {
   "id": "turn-pdfs-and-textbooks-into-interactive-courses",
   "title": "Turn PDFs and textbooks into interactive courses",
   "summary": "Linked example of turn PDFs and textbooks into interactive courses. The agent receives a PDF or textbook and parses it, keeping small files in context or embedding larger ones.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/nutlope/status/1981373070666264805",
   "prompt": "Research brief: turn PDFs and textbooks into interactive courses. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 30,
   "page": "https://agentsightings.com/case/turn-pdfs-and-textbooks-into-interactive-courses.html"
  },
  {
   "id": "plan-in-claude-code-implement-in-codex",
   "title": "Plan in Claude Code, implement in Codex",
   "summary": "Calvin keeps Claude Code open for planning, git and exploration, then flips to Codex when ready to implement, with separate-family review on PRs.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://calv.info/agents-feb-2026",
   "prompt": "Using the approved plan in PLAN.md, implement stage 1 on a new branch, run tests, and open a draft PR. Do not start stage 2 until review comments are addressed.",
   "added": "2026-10-15",
   "interest": 30,
   "page": "https://agentsightings.com/case/plan-in-claude-code-implement-in-codex.html"
  },
  {
   "id": "pick-the-cheapest-capable-model-per-task-for-crm-and-email-work",
   "title": "Pick the cheapest capable model per task for CRM and email work",
   "summary": "Documented elsewhere: openClaw shows up in a write-up about pick the cheapest capable model per task for CRM and email work. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "Email & Inbox",
   "source": "Blog",
   "url": "https://substack.com/",
   "prompt": "Go through my mail for this goal: pick the cheapest capable model per task for CRM and email work. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 30,
   "page": "https://agentsightings.com/case/pick-the-cheapest-capable-model-per-task-for-crm-and-email-work.html"
  },
  {
   "id": "cloud-browser-chores-driven-from-an-ide-agent",
   "title": "Cloud-browser chores driven from an IDE agent",
   "summary": "A founder uses Cursor's agent to write scripts that drive a cloud browser for daily web tasks without keeping a laptop on.",
   "agent": "Cursor",
   "topic": "Automation",
   "source": "Blog",
   "url": "https://cluelessfounder.com/blog/beyond-chatbots-ai-agents-guide/",
   "prompt": "Write a script that opens a headless browser, logs into this dashboard, exports yesterday's report as CSV and saves it to my drive folder.",
   "added": "2026-10-15",
   "interest": 30,
   "page": "https://agentsightings.com/case/cloud-browser-chores-driven-from-an-ide-agent.html"
  },
  {
   "id": "build-an-e-commerce-agent",
   "title": "Build an e-commerce agent",
   "summary": "Observed in the wild: utilized OpenClaw to build an e-commerce workflow agent, spending over 100 hours and 200 million tokens, and emphasizing the build effort in this post.",
   "agent": "OpenClaw",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/openclaw/comments/1rgcunb/spent_100_hours_and_burned_200m_tokens_building",
   "prompt": "Automation design: build an e-commerce agent. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 30,
   "page": "https://agentsightings.com/case/build-an-e-commerce-agent.html"
  },
  {
   "id": "build-a-deals-page-step-by-step",
   "title": "Build a deals page step by step",
   "summary": "Real-world thread: devin shows up in a write-up about build a deals page step by step, with figures like 10. The linked post is the primary source for steps and outcomes.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://qubika.com/blog/devin-ai-coding-agent/",
   "prompt": "In this repo, build a deals page step by step. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 30,
   "page": "https://agentsightings.com/case/build-a-deals-page-step-by-step.html"
  },
  {
   "id": "turn-a-recorded-task-into-instructions-an-agent-runs",
   "title": "Turn a recorded task into instructions an agent runs",
   "summary": "Observed in the wild: the poster recorded a Loom walkthrough of the task while narrating why each click mattered, uploaded the video to YouTube, and used Gemini 2.5 Pro to turn that recording into detailed step-by-step instructions. They pasted those instructions into ChatGPT Agent, which ran in a virtual machine, paused for Someone to enter Salesforce credentials and a rotating authorization code, then pulled a Salesforce report, checked the same account in Gong, summarized the last 7 days, and assembled the result in a Word document. The agent produced the exact same output as the poster’s manual process, and the poster saw it as a practical time-saver for recurring reporting.",
   "agent": "ChatGPT Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ChatGPT/comments/1m97oeu",
   "prompt": "Help me turn a recorded task into instructions an agent runs. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/turn-a-recorded-task-into-instructions-an-agent-runs.html"
  },
  {
   "id": "test-an-agent-on-seo-research-and-outreach-tasks",
   "title": "Test an agent on SEO research and outreach tasks",
   "summary": "Real-world thread: SE Ranking gave ChatGPT Operator prompts in four areas: competitor analysis, SERP monitoring, keyword research, and link building. It had to pull top-10 Google and Bing results into a Google Sheet, check a sheet of referring pages for backlinks to seranking.com, and draft replies to unread link-building emails in Gmail. Simple SERP pulls and outreach drafting worked best.",
   "agent": "Operator",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://seranking.com/blog/chatgpt-operator-for-seo/",
   "prompt": "GTM task: test an agent on SEO research and outreach tasks. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/test-an-agent-on-seo-research-and-outreach-tasks.html"
  },
  {
   "id": "send-emails-from-a-signed-in-inbox",
   "title": "Send emails from a signed-in inbox",
   "summary": "Community report: in the Reddit thread, the poster says they used ChatGPT agent for tasks like booking appointments, sending emails, and sending messages to their boss. In a follow-up comment, they clarify the email case: the agent asked them to sign into their email account, they did, and it then actually sent the emails; elsewhere they say they also had it find about 10 tutors and write personalized outreach emails based on what each could teach. The reported result is that it sent real emails from the signed-in inbox, covering outreach to around 10 tutors.",
   "agent": "ChatGPT Agent",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1m49e7d/chatgpt_agent_is_much_more_useful_than_i_thought",
   "prompt": "Go through my mail for this goal: send emails from a signed-in inbox. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/send-emails-from-a-signed-in-inbox.html"
  },
  {
   "id": "run-academic-literature-reviews-from-seed-papers",
   "title": "Run academic literature reviews from seed papers",
   "summary": "Real-world thread: the poster made a lightweight agent skill for literature reviews because general chatbots were missing relevant papers and citation tracing. The skill starts from seed papers, follows references backward and citations forward, screens the papers it locates, and iteratively refines the review; the workflow is partly hard-coded/scripted and meant to run inside tools such as Claude Code, Codex, Gemini CLI, or OpenClaw. In the poster’s tests, the final review was better than a generic AI-generated document, especially on coverage and paper selection.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AIToolBench/comments/1su9rqu/i_built_a_agent_skill_for_academic_literature",
   "prompt": "Research brief: run academic literature reviews from seed papers. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/run-academic-literature-reviews-from-seed-papers.html"
  },
  {
   "id": "run-a-systematic-literature-review-from-one-prompt",
   "title": "Run a systematic literature review from one prompt",
   "summary": "Real-world thread: user enters the prompt at scispace.com. The agent produces PICO-based research questions and screening criteria for Someone to review and edit, then lets Someone select databases and date ranges. It outputs a complete systematic review and PRISMA diagram from a single prompt, but paywalled papers still block full automation.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/heyrimsha/status/2037537471563776039",
   "prompt": "Research brief: run a systematic literature review from one prompt. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/run-a-systematic-literature-review-from-one-prompt.html"
  },
  {
   "id": "run-a-shared-team-agent-for-calls-approvals-and-tasks",
   "title": "Run a shared team agent for calls, approvals and tasks",
   "summary": "Community report: nikki stood up a shared Hermes Agent named Robert for the buzzdotxyz startup team. It receives access to the team's full Obsidian knowledge base plus Hermes memory, enabling any member to ask questions or assign work. Robert supplies context beyond any single chat, helping the non-technical co-founder and others complete more work.",
   "agent": "Hermes Agent",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/niklas11891/status/2105541804267274335",
   "prompt": "Research brief: run a shared team agent for calls, approvals and tasks. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/run-a-shared-team-agent-for-calls-approvals-and-tasks.html"
  },
  {
   "id": "run-a-daily-seo-article-campaign",
   "title": "Run a daily SEO article campaign",
   "summary": "Observed in the wild: the agent received access to keyword and article research plus a transcript of Someone's views on the topic. It researched keywords and articles independently, wrote one article daily for the target keyword, and published it. Article 1 reached page one of search results.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/DigitalMarketing/comments/1uhpucc/how_do_you_use_ai_in_marketing_for_real",
   "prompt": "GTM task: run a daily SEO article campaign. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/run-a-daily-seo-article-campaign.html"
  },
  {
   "id": "generate-monthly-marketing-reports-and-buying-intent-replies",
   "title": "Generate monthly marketing reports and buying-intent replies",
   "summary": "As described in the source, one commenter said they upload SEO, social, GA4, and GSC reports into Claude each month and give it substantial business context so it can turn those inputs into a monthly update report, saving them analysis and interpretation time. Another said they monitor Reddit, Quora, and X for buying-intent threads, then have Claude draft tone-matched replies for human review before posting. The thread reports clear time savings rather than hard business metrics: one person said monthly reporting saves them hours, and another cut community-response work from 3–4 hours to about 20 minutes.",
   "agent": "Claude",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/digital_marketing/comments/1td72d9/how_are_you_using_claude_for_marketing",
   "prompt": "GTM task: generate monthly marketing reports and buying-intent replies. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/generate-monthly-marketing-reports-and-buying-intent-replies.html"
  },
  {
   "id": "fight-a-mattress-warranty-claim-by-email",
   "title": "Fight a mattress warranty claim by email",
   "summary": "Community report: user emailed the mattress company about a two-year-old warranty issue and spotted ChatGPT-style replies after one or two exchanges. They then set their own agent to handle the ongoing email thread, giving it instructions to cite laws and respond as an aggressive pain-in-the-ass customer while the company's side stayed polite. The agents continued exchanging emails with no resolution or outcome numbers stated in the post.",
   "agent": "Custom agent",
   "topic": "Email & Inbox",
   "source": "X",
   "url": "https://x.com/yosoymario91/status/2104544727928377679",
   "prompt": "Go through my mail for this goal: fight a mattress warranty claim by email. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/fight-a-mattress-warranty-claim-by-email.html"
  },
  {
   "id": "complete-an-online-survey-end-to-end",
   "title": "Complete an online survey end to end",
   "summary": "A published walkthrough shows the poster says they gave Meta’s Muse a link to one of their surveys and told it to complete the whole thing without human help. According to the post, Muse handled the survey from start to finish on its own, covering navigating pages, answering questions, solving the captcha, and submitting the response; the poster says they never clicked, answered, or took over. The reported result was a fully submitted survey response with no human intervention.",
   "agent": "Meta Muse",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Marketresearch/comments/1ww7umw/metas_ai_agent",
   "prompt": "Research brief: complete an online survey end to end. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/complete-an-online-survey-end-to-end.html"
  },
  {
   "id": "build-a-52-node-n8n-workflow",
   "title": "Build a 52-node n8n workflow",
   "summary": "A published walkthrough shows someone tasked Claude Computer Use with building a 52-node n8n workflow. The agent performed the full build on its own in one afternoon. The workflow was completed in one afternoon rather than the three weekends required previously.",
   "agent": "Claude computer use",
   "topic": "Automation",
   "source": "X",
   "url": "https://x.com/dailystartupfix/status/2032737694803951785",
   "prompt": "Automation design: build a 52-node n8n workflow. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 29,
   "page": "https://agentsightings.com/case/build-a-52-node-n8n-workflow.html"
  },
  {
   "id": "update-a-digital-assets-platform",
   "title": "Update a digital assets platform",
   "summary": "Community report: Manus was pointed at update a digital assets platform. in Genesis v2.0, Manus was utilized by Mufasa to look over features, iterate mockups, and develop code for a new AI automation industry.",
   "agent": "Manus",
   "topic": "Knowledge & Notes",
   "source": "X",
   "url": "https://x.com/mufasadubai/status/2027122360101347426",
   "prompt": "Notes task: update a digital assets platform. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/update-a-digital-assets-platform.html"
  },
  {
   "id": "track-personal-finances",
   "title": "Track personal finances",
   "summary": "Community report: claude Code shows up in a write-up about track personal finances. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://aiiq.substack.com/p/start-using-claude-code-or-codex",
   "prompt": "Notes task: track personal finances. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/track-personal-finances.html"
  },
  {
   "id": "ship-a-first-ios-app-to-the-app-store",
   "title": "Ship a first iOS app to the App Store",
   "summary": "Community report: from the linked report: utilizing ChatGPT support, they developed their initial app using Replit and successfully made it available on the App Store.",
   "agent": "Replit Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1s7onvb/ive_just_built_my_first_app_on_replit_but_need",
   "prompt": "Help me ship a first iOS app to the App Store. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/ship-a-first-ios-app-to-the-app-store.html"
  },
  {
   "id": "search-state-databases-for-unclaimed-property-under-old-names",
   "title": "Search state databases for unclaimed property under old names",
   "summary": "As described in the source, claude covered search state databases for unclaimed property under old names. requested Claude to search for unclaimed property in all states they inhabited, powered by their names and old names.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/PromptEngineering/comments/1vrij5c/about_1_in_7_people_have_unclaimed_money_sitting",
   "prompt": "Help me search state databases for unclaimed property under old names. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/search-state-databases-for-unclaimed-property-under-old-names.html"
  },
  {
   "id": "run-marketing-operations",
   "title": "Run marketing operations",
   "summary": "From the linked report, claude Code shows up in a write-up about run marketing operations. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AskGTM/comments/1tngr97/automated_my_marketing_operations_with_claude_code",
   "prompt": "GTM task: run marketing operations. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/run-marketing-operations.html"
  },
  {
   "id": "review-a-housing-court-case-as-an-experienced-attorney",
   "title": "Review a housing court case as an experienced attorney",
   "summary": "Community report: claude covered look over a housing court case as an experienced attorney The linked post has the narrative detail.",
   "agent": "Claude",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1qkxkae/claude_for_legal_advice",
   "prompt": "Notes task: review a housing court case as an experienced attorney. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/review-a-housing-court-case-as-an-experienced-attorney.html"
  },
  {
   "id": "publish-ai-written-blog-posts-from-trending-news-to-wordpress",
   "title": "Publish AI-written blog posts from trending news to WordPress",
   "summary": "From the linked report, n8n shows up in a write-up about publish AI-written blog posts from trending news to WordPress, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://dev.to/grewup/automate-blog-publishing-with-n8n-ai-wordpress-full-workflowpublished-true-56cl",
   "prompt": "Creative task: publish AI-written blog posts from trending news to WordPress. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/publish-ai-written-blog-posts-from-trending-news-to-wordpress.html"
  },
  {
   "id": "parallel-claude-code-workstreams-for-a-saas",
   "title": "Parallel Claude Code workstreams for a SaaS",
   "summary": "A solo founder shipped a social-media management platform with 12 API integrations in three weeks by splitting layers into parallel Claude Code worktrees and Codex review.",
   "agent": "Claude Code",
   "topic": "Business & Startups",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47749688",
   "prompt": "Split this layer of the spec into independent workstreams. For each, create a worktree, implement behind feature flags, and open a draft PR. Do not start dependent layers until these merge.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/parallel-claude-code-workstreams-for-a-saas.html"
  },
  {
   "id": "find-weak-web-presence-businesses-and-pitch-them",
   "title": "Find weak-web-presence businesses and pitch them",
   "summary": "From the linked report, custom agent shows up in a write-up about find weak-web-presence businesses and pitch them. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/passive_income/comments/1rmsb2o/i_set_up_6_ai_agents_that_sell_web_services_247",
   "prompt": "GTM task: find weak-web-presence businesses and pitch them. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/find-weak-web-presence-businesses-and-pitch-them.html"
  },
  {
   "id": "find-vegetarian-restaurants-and-save-notes-via-computer-use",
   "title": "Find vegetarian restaurants and save notes via computer use",
   "summary": "Documented elsewhere: following Anthropic's setup instructions, The builder installed Docker, stood up a Claude API key, ran one Terminal command and opened the web app. They gave Claude the restaurant task with Google already open. Setup worked first time and the text file was stood up.",
   "agent": "Claude computer use",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://lovkush.substack.com/p/i-tried-anthropics-computer-use-agent",
   "prompt": "Notes task: find vegetarian restaurants and save notes via computer use. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/find-vegetarian-restaurants-and-save-notes-via-computer-use.html"
  },
  {
   "id": "create-social-media-visuals",
   "title": "Create social media visuals",
   "summary": "Case study takeaway: attempted to create app marketing visuals using Canva MCP in Claude Code, but it was an unsuccessful endeavor.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1qefbcn/claude_code_is_amazing_but_how_do_you_handle_app",
   "prompt": "GTM task: create social media visuals. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/create-social-media-visuals.html"
  },
  {
   "id": "compile-journalist-lists-and-search-apartment-listings",
   "title": "Compile journalist lists and search apartment listings",
   "summary": "Observed in the wild: the reporter gave Manus three tasks: a list of China tech reporters, two-bedroom NYC listings with several criteria, and 50 Innovators Under 35 nominees. She watched its browser window and took over for captchas. The reporter list took a little over an hour and the apartment search under half an hour.",
   "agent": "Manus",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://www.technologyreview.com/2025/03/11/1113133/manus-ai-review/",
   "prompt": "Help me compile journalist lists and search apartment listings. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/compile-journalist-lists-and-search-apartment-listings.html"
  },
  {
   "id": "build-a-personal-research-wiki-from-papers-and-articles",
   "title": "Build a personal research wiki from papers and articles",
   "summary": "One practitioner notes that the vault folder is opened in both Cursor and Obsidian. Sources go into an immutable raw/ folder, and an AGENTS.md file sets the wiki's conventions. The source is a pattern description, not a report with results.",
   "agent": "Cursor",
   "topic": "Knowledge & Notes",
   "source": "GitHub",
   "url": "https://github.com/rbarok/obsidian-cursor-knowledge-graph/blob/main/cursor-obsidian.md",
   "prompt": "Notes task: build a personal research wiki from papers and articles. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/build-a-personal-research-wiki-from-papers-and-articles.html"
  },
  {
   "id": "build-a-company-structure",
   "title": "Build a company structure",
   "summary": "Community report: as described in the source: founded a business with the assistance of Claude Code agents, who also have a marketing representative responsible for brand identity and color selection.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1ostwq9/i_built_an_entire_fake_company_with_claude_code",
   "prompt": "GTM task: build a company structure. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/build-a-company-structure.html"
  },
  {
   "id": "build-a-cardiac-health-tracking-app",
   "title": "Build a cardiac health tracking app",
   "summary": "Community report: after a November 2025 hospitalization for dehydration, the poster started with a spreadsheet and then used months of back-and-forth with Claude to turn it into ClinBridge, a progressive web app for tracking blood pressure, fluid intake, weight, heart rate, symptoms, meals, and activity. He says Claude helped build and word the project, while he decided the requirements; the app installs on a phone, works offline, and keeps data on the device. He says the app reached version 9.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1s2hf5f/73_years_old_no_coding_experience_cardiac_patient",
   "prompt": "In this repo, build a cardiac health tracking app. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 28,
   "page": "https://agentsightings.com/case/build-a-cardiac-health-tracking-app.html"
  },
  {
   "id": "set-up-a-reddit-profile-through-desktop-control",
   "title": "Set up a Reddit profile through desktop control",
   "summary": "Observed in the wild: a commenter said they had Claude Code set up their Reddit profile entirely from the CLI. The agent first took a screenshot to inspect the current page, then used Playwright MCP to navigate into Reddit settings, fill out the profile forms, upload images, and when one image exceeded Reddit’s 500KB limit, it compressed the file and retried the upload without human help. The reported outcome was a completed Reddit profile with zero manual intervention.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1s7wkky/computer_use_is_now_in_claude_code",
   "prompt": "Help me set up a Reddit profile through desktop control. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 27,
   "page": "https://agentsightings.com/case/set-up-a-reddit-profile-through-desktop-control.html"
  },
  {
   "id": "run-parallel-local-coding-agent-sessions",
   "title": "Run parallel local coding agent sessions",
   "summary": "A published walkthrough shows roark66 runs coding agents on RTX 3090s with 24GB each, powered by Claude Code alongside local models. They keep up to 10 agent sessions open at once, though never more than 5 are active at the same moment. Holding about 15 full contexts cached in RAM pushes their local cache utilisation to 95% or more.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49992657",
   "prompt": "In this repo, run parallel local coding agent sessions. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 27,
   "page": "https://agentsightings.com/case/run-parallel-local-coding-agent-sessions.html"
  },
  {
   "id": "give-a-coding-agent-a-paper-corpus-to-find-new-techniques",
   "title": "Give a coding agent a paper corpus to find new techniques",
   "summary": "Claude Code covered give a coding agent a paper corpus to find new techniques. The poster ran two identical Claude Code agents on the same job: optimize a small language model.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/artificial/comments/1s6afwm/i_tested_what_happens_when_you_give_an_ai_coding",
   "prompt": "In this repo, give a coding agent a paper corpus to find new techniques. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 27,
   "page": "https://agentsightings.com/case/give-a-coding-agent-a-paper-corpus-to-find-new-techniques.html"
  },
  {
   "id": "generate-updated-research-reports-on-followed-companies",
   "title": "Generate updated research reports on followed companies",
   "summary": "As described in the source, the poster said they assembled a stock-research agent on top of paid real-time market-data subscriptions because ChatGPT’s answers were too dated for investing. In comments, they said it uses Polygon real-time stocks and options data, can run code on that data to answer screening-style requests like finding today’s inside-day patterns, and is available as a CLI plus a free web version; people still had to test outputs and report mistakes. The reported benefit was fresher research and the ability to answer data-driven trading queries ChatGPT alone could not.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/OpenAI/comments/1mdrdbh/built_my_own_agent_for_stock_research_and",
   "prompt": "Research brief: generate updated research reports on followed companies. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 27,
   "page": "https://agentsightings.com/case/generate-updated-research-reports-on-followed-companies.html"
  },
  {
   "id": "triage-a-founder-x27-s-inbox",
   "title": "Triage a founder's inbox",
   "summary": "Documented elsewhere: claude Code is responsible for reading the inbox, labeling messages, and drafting replies, as well as preparing calendars and reports, all while connecting a Gmail MCP server to the team.",
   "agent": "Claude Code",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1tq8zpc/gave_claude_code_my_inbox_calendar_through_one",
   "prompt": "Go through my mail for this goal: triage a founder's inbox. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/triage-a-founder-x27-s-inbox.html"
  },
  {
   "id": "summarize-months-of-notes-in-logseq-via-cli-agent",
   "title": "Summarize months of notes in Logseq via CLI agent",
   "summary": "Documented elsewhere: wosk pasted the Logseq CLI documentation, installed alongside Logseq, to Codex, which then knew how to query the graph. It condenses journals across months and works with tagged pages such as job applications (with status values) and research projects. Codex can summarize months of journals, and tagged blocks make queries natural.",
   "agent": "Codex",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48897407",
   "prompt": "Notes task: summarize months of notes in Logseq via CLI agent. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/summarize-months-of-notes-in-logseq-via-cli-agent.html"
  },
  {
   "id": "rewrite-low-ctr-meta-titles-and-audit-sites-for-seo",
   "title": "Rewrite low-CTR meta titles and audit sites for SEO",
   "summary": "Case study takeaway: studioHawk tests Claude Code in controlled batches. It connects to a CMS through its API and pulls Google Search Console data. The source calls early results promising.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://studiohawk.com.au/blog/claude-code-seo-automation/",
   "prompt": "GTM task: rewrite low-CTR meta titles and audit sites for SEO. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/rewrite-low-ctr-meta-titles-and-audit-sites-for-seo.html"
  },
  {
   "id": "local-multi-session-agent-with-parallel-task-graphs",
   "title": "Local multi-session agent with parallel task graphs",
   "summary": "A local agent runtime splits a large task into parallel sub-sessions across different model providers and resumes them from SQLite.",
   "agent": "Captain Claw",
   "topic": "Developer Tools",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47261534",
   "prompt": "Break this task into independent subtasks, run them in parallel, and then merge the results into a single report with a section per subtask.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/local-multi-session-agent-with-parallel-task-graphs.html"
  },
  {
   "id": "generate-weekly-leadership-reports-and-handle-feature-requests",
   "title": "Generate weekly leadership reports and handle feature requests",
   "summary": "Zapier’s pattern here is a boss-agent: one helper turns Jira history into the Monday leadership pack; another answers “/feature” style Slack asks. Humans only review the outputs.",
   "agent": "Zapier Agents",
   "topic": "Automation",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/zapier/comments/1raauiz/built_a_system_in_zapier_where_agents_handle",
   "prompt": "Automation design: generate weekly leadership reports and handle feature requests. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/generate-weekly-leadership-reports-and-handle-feature-requests.html"
  },
  {
   "id": "generate-a-daily-personalized-trading-research-report-from-brokerage-data",
   "title": "Generate a daily personalized trading research report from brokerage data",
   "summary": "As described in the source, claude shows up in a write-up about generate a daily personalized trading research report from brokerage data. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/lonniev/status/2106218283426881984",
   "prompt": "Research brief: generate a daily personalized trading research report from brokerage data. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/generate-a-daily-personalized-trading-research-report-from-brokerage-data.html"
  },
  {
   "id": "cross-family-ai-review-before-human-pr-review",
   "title": "Cross-family AI review before human PR review",
   "summary": "After six months of agent-only coding, a commenter routes every PR through a different-model-family reviewer plus deterministic gates before looking at it.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49465119",
   "prompt": "Review this PR as a different model family than the author. Check correctness, security and missing tests. Return only actionable findings; do not rewrite the whole diff.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/cross-family-ai-review-before-human-pr-review.html"
  },
  {
   "id": "city-wide-indie-cinema-listings",
   "title": "City-wide indie cinema listings",
   "summary": "An LLM-heavy pipeline scrapes small cinema websites, cleans and classifies showtimes, and builds one listing for the city.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=44405067",
   "prompt": "Visit each cinema page in this list, extract film title, showtime, and venue into a table, and flag pages you couldn't parse.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/city-wide-indie-cinema-listings.html"
  },
  {
   "id": "answer-patient-enquiries-for-a-clinic-by-chat",
   "title": "Answer patient enquiries for a clinic by chat",
   "summary": "From the linked report, patients send messages through a chat interface into an n8n workflow. Gemini agent with Simple Memory for context pulls clinic details from Google Sheets and produces replies to enquiries on its own. No results or metrics are described.",
   "agent": "n8n",
   "topic": "Automation",
   "source": "X",
   "url": "https://x.com/SahilWagave/status/2109697093778690255",
   "prompt": "Automation design: answer patient enquiries for a clinic by chat. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 26,
   "page": "https://agentsightings.com/case/answer-patient-enquiries-for-a-clinic-by-chat.html"
  },
  {
   "id": "turn-a-blog-post-into-an-animated-infographic",
   "title": "Turn a blog post into an animated infographic",
   "summary": "Community report: in a long-running Gemini chat that already held her brand hex codes, fonts and audience, she pasted in a blog post about the state of AI for entrepreneurs. She switched to Canvas and chose Create > Infographic. Within a couple of minutes it produced an on-brand infographic with fade-in data points and chart hover effects.",
   "agent": "Gemini",
   "topic": "Media & Content",
   "source": "Blog",
   "url": "https://kimdoyal.substack.com/p/this-free-gemini-feature-is-my-new",
   "prompt": "Creative task: turn a blog post into an animated infographic. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/turn-a-blog-post-into-an-animated-infographic.html"
  },
  {
   "id": "run-an-ai-agent-on-a-nokia-110-feature-phone",
   "title": "Run an AI agent on a Nokia 110 feature phone",
   "summary": "Community report: the builder studied the Nokia 110 4G firmware and boot process, then ran custom code in RAM through the Calculator entry point. They turned Calculator into a chat app that sends typed messages to the DeepSeek chat API and acts on tool calls. Over mobile data, unplugged, it can answer questions, check the battery, switch the torch on and off, start a call and set an alarm.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50006114",
   "prompt": "Help me run an AI agent on a Nokia 110 feature phone. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/run-an-ai-agent-on-a-nokia-110-feature-phone.html"
  },
  {
   "id": "recreate-a-video-ad-as-a-javascript-animation",
   "title": "Recreate a video ad as a JavaScript animation",
   "summary": "Community report: the builder downloaded a ChatGPT ad from YouTube, extracted its audio, and gave Claude (Opus 5.5) access to ffmpeg. A prompt asked for a pure JavaScript recreation powered by the same audio, with subtle animations captured and high production value. It finished inside an hour in one go.",
   "agent": "Claude",
   "topic": "Media & Content",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49942596",
   "prompt": "Creative task: recreate a video ad as a JavaScript animation. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/recreate-a-video-ad-as-a-javascript-animation.html"
  },
  {
   "id": "monitor-competitor-marketplace-prices",
   "title": "Monitor competitor marketplace prices",
   "summary": "Community report: the poster says they use Grok Bot to periodically check product prices and competitor listings on a few online marketplaces. The bot browses those sites on its own, but some marketplaces interrupt it with CAPTCHAs; the poster says they currently have to use an API to solve them, and commenters suggested lower request rates, logged-in sessions, or routing the bot’s traffic through Someone’s own device. It works for ongoing marketplace price checks, but reliability breaks on sites that throw CAPTCHAs, covering a mouse-movement puzzle.",
   "agent": "Grok Bot",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1wq5vdz/grok_bot_keeps_getting_captcha_when_browsing",
   "prompt": "Help me monitor competitor marketplace prices. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/monitor-competitor-marketplace-prices.html"
  },
  {
   "id": "hand-a-day-of-busywork-to-cowork",
   "title": "Hand a day of busywork to Cowork",
   "summary": "As described in the source, the builder, an architect, described outcomes to Cowork in the Claude desktop app and approved its plan before it ran. Tasks were sorting a 200-plus-file folder (he changed one rule to keep originals), filling a weekly report template from source files, and turning messy notes and links into a structured draft. The folder came back tidy with a duplicates note.",
   "agent": "Claude Cowork",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://dev.to/suraj_khaitan_f893c243958/claude-cowork-review-i-handed-it-a-day-of-my-busywork-heres-what-came-back-1b92",
   "prompt": "Notes task: hand a day of busywork to Cowork. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/hand-a-day-of-busywork-to-cowork.html"
  },
  {
   "id": "find-and-clean-leads-for-recruiting-outreach",
   "title": "Find and clean leads for recruiting outreach",
   "summary": "A published walkthrough shows the poster wired Claude to MCP connections for Crustdata, FullEnrich, Lemlist, HubSpot, and GitHub, then gave it plain-English criteria like US companies with 100–500 employees, 15+ recent openings, fresh funding, and a recruiting manager to contact. Claude pulled company and hiring signals, located and verified contacts, assembled hiring and news briefs, and drafted outreach angles; in comments, another recruiter said they use Claude to clean rough lists, enrich and organize them into sheets/CRMs, and polish personalized messages before sending. The source says it saved the poster time and commenters said personalized Claude-assisted outreach gets replies, but it gives no hard performance numbers.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/RecruitmentAgencies/comments/1ryqkmd/using_claude_to_find_leads_for_my_recruiting",
   "prompt": "GTM task: find and clean leads for recruiting outreach. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/find-and-clean-leads-for-recruiting-outreach.html"
  },
  {
   "id": "figma-mcp-to-pixel-perfect-ui",
   "title": "Figma MCP to pixel-perfect UI",
   "summary": "A developer connects Figma MCP to Claude Code so frontend work goes from design frames to UI that matches the file, instead of manual translation.",
   "agent": "Claude Code",
   "topic": "Design & Creative",
   "source": "Blog",
   "url": "https://dev.to/taruntomar122/my-actual-ai-coding-setup-in-2026-and-how-i-decide-what-to-use-for-what-44oh",
   "prompt": "Using the Figma MCP, implement this frame as production UI in our design system. Match spacing and typography; call out anything the design system cannot express.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/figma-mcp-to-pixel-perfect-ui.html"
  },
  {
   "id": "compose-and-produce-music-through-conversation",
   "title": "Compose and produce music through conversation",
   "summary": "From the linked report, the builder Leaned on Claude 4.5 models in Claude Code to write multi-track music and lyrics, plus the tooling for composing and producing. Claude was connected to Bitwig Studio through projects such as a Bitwig content search CLI and the Jeannie controller framework. The gist collects five songs, covering a symphony finale and a twelve-song concept album, with MIDI and lyrics on GitHub.",
   "agent": "Claude Code",
   "topic": "Media & Content",
   "source": "GitHub",
   "url": "https://gist.github.com/bedwards/de776e759c8fd0191c011e12ad08e443",
   "prompt": "Creative task: compose and produce music through conversation. Outline first, then draft. Preserve my voice; ask before posting publicly.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/compose-and-produce-music-through-conversation.html"
  },
  {
   "id": "analyze-10-ks-for-value-investing",
   "title": "Analyze 10-Ks for value investing",
   "summary": "A published walkthrough shows the poster was testing an LLM investing assistant that read 10-Ks and tried to produce normalized earnings, ROIC, owner earnings, and a rough reverse DCF. In the thread, they say they first used EDGAR/XBRL Companyfacts, then easier APIs, but ran into custom tags, footnotes, and backtest leakage; commenters pushed a workflow that combines structured headline figures with retrieval over filing text, stores source passages and year-by-year claims outside the chat, and leaves the human to verify tricky comparability issues. The source reports no performance results, only problems and partial fixes: feeding whole filing histories hit context limits, chunking led to missed evidence and invented facts, and backtests looked unrealistically good because of restatements and filing lag.",
   "agent": "Custom agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ValueInvesting/comments/1x52d8o/anyone_building_a_value_investing_agent_what_are",
   "prompt": "Help me analyze 10-Ks for value investing. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 25,
   "page": "https://agentsightings.com/case/analyze-10-ks-for-value-investing.html"
  },
  {
   "id": "let-ai-agents-design-in-figma",
   "title": "Let AI agents design in Figma",
   "summary": "Community report: the figma-use CLI connects to the Figma desktop app over Chrome DevTools Protocol and runs commands through Figma's Plugin API. Agents either issue commands one at a time or pipe JSX into a render command. The demos show button components and a Tailwind UI calendar.",
   "agent": "Custom agent",
   "topic": "Design & Creative",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46665169",
   "prompt": "Design task: let AI agents design in Figma. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 24,
   "page": "https://agentsightings.com/case/let-ai-agents-design-in-figma.html"
  },
  {
   "id": "analyze-competitor-ads-and-generate-on-brand-ad-creatives",
   "title": "Analyze competitor ads and generate on-brand ad creatives",
   "summary": "One practitioner notes that the Ad Agent was assembled on the Claude Agent SDK and given a prompt to find competitors, learn from their ads, generate on-brand creatives, and package a shareable report. It ran the full sequence on its own with no node-based workflows or extra setup. The agent completed the described workflow of competitor analysis, creative generation, and report packaging.",
   "agent": "Claude",
   "topic": "Design & Creative",
   "source": "X",
   "url": "https://x.com/boringmarketer/status/2008607337532014709",
   "prompt": "Design task: analyze competitor ads and generate on-brand ad creatives. Show options, explain tradeoffs, and wait before exporting or publishing.",
   "added": "2026-10-15",
   "interest": 24,
   "page": "https://agentsightings.com/case/analyze-competitor-ads-and-generate-on-brand-ad-creatives.html"
  },
  {
   "id": "send-daily-ai-news-research-reports",
   "title": "Send daily AI news research reports",
   "summary": "A published walkthrough shows A Grok bot was configured to scan X and AI news every few hours. It generated five written research reports each day, with Someone's system verifying every sourced figure before any number appeared in the output. Research arrived pre-finished, moving Someone's role from searching to deciding.",
   "agent": "Grok Bot",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/madebykamil/status/2108948281237581829",
   "prompt": "Research brief: send daily AI news research reports. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 23,
   "page": "https://agentsightings.com/case/send-daily-ai-news-research-reports.html"
  },
  {
   "id": "keep-coding-agents-running-autonomously-for-long-stretches",
   "title": "Keep coding agents running autonomously for long stretches",
   "summary": "Documented elsewhere: cursor ran hundreds of concurrent agents on one project. Flat self-coordination through a shared file with locks, then optimistic concurrency, failed. A browser assembled from scratch ran close to a week and produced over 1 million lines across 1,000 files.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46624541",
   "prompt": "In this repo, keep coding agents running autonomously for long stretches. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 23,
   "page": "https://agentsightings.com/case/keep-coding-agents-running-autonomously-for-long-stretches.html"
  },
  {
   "id": "generate-a-robot-arm-cad-model-from-one-prompt",
   "title": "Generate a robot arm CAD model from one prompt",
   "summary": "Community report: jake Fitzgerald installed the open source text-to-cad skill package in Codex, running GPT 5.5. The agent writes build123d Python scripts, regenerates the geometry locally, and outputs STEP parts plus a URDF. The arm's URDF, kinematics, GUI and STEP assembly were all generated, except the gripper.",
   "agent": "Codex",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47970497",
   "prompt": "Help me generate a robot arm CAD model from one prompt. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 23,
   "page": "https://agentsightings.com/case/generate-a-robot-arm-cad-model-from-one-prompt.html"
  },
  {
   "id": "build-a-balloon-camel-simulator-for-an-led-burning-man-totem",
   "title": "Build a balloon-camel simulator for an LED Burning Man totem",
   "summary": "One practitioner notes that the builder Leaned on ChatGPT to generate reference images and prompts, then had Claude Code build a Go physics simulation of balloons streaming geometry to a web frontend. Step one was nearly one-shot; step two added multiple balloons with twists. The simulator worked and exported models, but the result looked like a cross between a poop emoji and an alien dildo, not Koons-like.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49210226",
   "prompt": "Notes task: build a balloon-camel simulator for an LED Burning Man totem. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 23,
   "page": "https://agentsightings.com/case/build-a-balloon-camel-simulator-for-an-led-burning-man-totem.html"
  },
  {
   "id": "ship-an-ios-app-in-17-days-with-agents",
   "title": "Ship an iOS app in 17 days with agents",
   "summary": "Adam Van Lente went from a blank Xcode project to a live iOS app in 17 days with under 24 hours of keyboard time, mostly via Cursor, Claude Code and Codex.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://adamvanlente.com/posts/09-04-25-where-ai-coding-agents-shine-and-dont/",
   "prompt": "Map this iOS feature into bite-size prompts. Implement the next slice on a branch with tests. Stop when you hit a multi-layer integration (push, CloudKit, payments) and list what needs a human.",
   "added": "2026-10-15",
   "interest": 22,
   "page": "https://agentsightings.com/case/ship-an-ios-app-in-17-days-with-agents.html"
  },
  {
   "id": "run-an-outbound-sales-pipeline",
   "title": "Run an outbound sales pipeline",
   "summary": "Case study takeaway: grok Bot shows up in a write-up about run an outbound sales pipeline. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/romanbuildsaas/status/2094758876834938964",
   "prompt": "GTM task: run an outbound sales pipeline. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 22,
   "page": "https://agentsightings.com/case/run-an-outbound-sales-pipeline.html"
  },
  {
   "id": "qualify-enrich-and-book-demos-for-saas-leads",
   "title": "Qualify, enrich and book demos for SaaS leads",
   "summary": "A published walkthrough shows A Lindy agent is available round the clock to qualify, enrich, and book demos for SaaS lead gen, resulting in a rise in mock conversion from 40% to 78%.",
   "agent": "Lindy",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/Devesh143/status/2053881421178245215",
   "prompt": "GTM task: qualify, enrich and book demos for SaaS leads. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 22,
   "page": "https://agentsightings.com/case/qualify-enrich-and-book-demos-for-saas-leads.html"
  },
  {
   "id": "monitor-tasks-and-open-pull-requests",
   "title": "Monitor tasks and open pull requests",
   "summary": "A published walkthrough shows as described in the source: initiates a persistent script that keeps watch on tasks and opens pull requests with Cursor, establishing a data pipeline and summarizer.",
   "agent": "Cursor",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1q4iztv/anyone_succeed_building_something_cool_with",
   "prompt": "Help me monitor tasks and open pull requests. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 22,
   "page": "https://agentsightings.com/case/monitor-tasks-and-open-pull-requests.html"
  },
  {
   "id": "let-an-agent-start-a-business-from-scratch",
   "title": "Let an agent start a business from scratch",
   "summary": "Toolbox for Devin: Ramp spend card, ElevenLabs voice number, AgentMail, Stripe, GitHub memory. Mandate: invent a business. Result after ~four weeks / 10,929 outbound calls: first revenue customer (Josh’s public run).",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/joshjnunez/status/2100646367383126440",
   "prompt": "In this repo, let an agent start a business from scratch. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 22,
   "page": "https://agentsightings.com/case/let-an-agent-start-a-business-from-scratch.html"
  },
  {
   "id": "debug-frontend-ui-bugs",
   "title": "Debug frontend UI bugs",
   "summary": "Case study takeaway: claude shows up in a write-up about debug frontend UI bugs. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://anmol-gupta.medium.com/claude-in-chrome-the-surprising-frontend-debugging-tool-thats-changing-my-workflow-6cac64d8c96b",
   "prompt": "Notes task: debug frontend UI bugs. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 22,
   "page": "https://agentsightings.com/case/debug-frontend-ui-bugs.html"
  },
  {
   "id": "test-features-on-screen-and-merge-fixed-prs",
   "title": "Test features on screen and merge fixed PRs",
   "summary": "Case study takeaway: ryan Carson gives DevinAI access to his app via the Test App feature. The agent screen-records and narrates its own clicks through the feature, zeros in on bugs from those tests, fixes them in the code, and prepares a PR. Shipping velocity and dev happiness increased drastically with the cloud VMs and gpt-5.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/kr0der/status/2106399374859509893",
   "prompt": "In this repo, test features on screen and merge fixed PRs. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 21,
   "page": "https://agentsightings.com/case/test-features-on-screen-and-merge-fixed-prs.html"
  },
  {
   "id": "pull-stock-fundamentals-from-sec-filings-for-valuation",
   "title": "Pull stock fundamentals from SEC filings for valuation",
   "summary": "One practitioner notes that the developer prompted ChatGPT Agent mode, chosen for its web search, to produce fundamentals for four tickers (AAPL, MSFT, NVDA, TSLA) from SEC 10-Q filings. For each metric the agent returned a value, trend direction, 1-5 score, explanation, filing URL and the exact derivation. The agent output fed a working pipeline that combined valuation, sentiment and fundamental votes and backtested the picks over 63 trading days.",
   "agent": "ChatGPT Agent",
   "topic": "Automation",
   "source": "GitHub",
   "url": "https://github.com/phanhpp/Multi-Agent-Stock-Analysis-System",
   "prompt": "Automation design: pull stock fundamentals from SEC filings for valuation. Specify triggers, steps, and failure modes before enabling sends or deletes.",
   "added": "2026-10-15",
   "interest": 21,
   "page": "https://agentsightings.com/case/pull-stock-fundamentals-from-sec-filings-for-valuation.html"
  },
  {
   "id": "claude-code-as-a-general-computer-operator",
   "title": "Claude Code as a general computer operator",
   "summary": "A writer argues a terminal coding agent can handle many everyday computing jobs and shows examples beyond writing code.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://dwyer.co.za/static/claude-code-is-all-you-need.html",
   "prompt": "In this folder, figure out what these files are, organize them into sensible subfolders, and give me a summary of what you did.",
   "added": "2026-10-15",
   "interest": 21,
   "page": "https://agentsightings.com/case/claude-code-as-a-general-computer-operator.html"
  },
  {
   "id": "run-linkedin-outreach-from-icp-and-intent-signals",
   "title": "Run LinkedIn outreach from ICP and intent signals",
   "summary": "Community report: yielding 15+ replies, with grok Bot with an MCP and LinkedIn assembled outreach agents that located.",
   "agent": "Grok Bot",
   "topic": "Sales & Marketing",
   "source": "X",
   "url": "https://x.com/pierreeliottlal/article/2103098787199692830",
   "prompt": "GTM task: run LinkedIn outreach from ICP and intent signals. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 20,
   "page": "https://agentsightings.com/case/run-linkedin-outreach-from-icp-and-intent-signals.html"
  },
  {
   "id": "replace-token-heavy-jira-and-gitlab-mcps",
   "title": "Replace token-heavy Jira and GitLab MCPs",
   "summary": "Fat Jira/GitLab MCPs were torching tokens. Fix: have Claude read CLI --help, capture the useful verbs as Markdown, package thin plugins. Same ops, far less context bloat.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49788468",
   "prompt": "Help me replace token-heavy Jira and GitLab MCPs. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 20,
   "page": "https://agentsightings.com/case/replace-token-heavy-jira-and-gitlab-mcps.html"
  },
  {
   "id": "design-doc-research-across-multiple-repos",
   "title": "Design-doc research across multiple repos",
   "summary": "A workplace engineer uses Claude Code as a research assistant over several repos and schemas while drafting a design doc for a new feature.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46429652",
   "prompt": "Here are my repos, DB schema and prior design docs. Explain how system X works today and what would break if we did Y. Cite file paths.",
   "added": "2026-10-15",
   "interest": 20,
   "page": "https://agentsightings.com/case/design-doc-research-across-multiple-repos.html"
  },
  {
   "id": "build-a-multi-tenant-saas",
   "title": "Build a multi-tenant SaaS",
   "summary": "From the linked report, replit Agent was utilized to construct a multi-tenant SaaS production version within 15 days, but encountered complications such as RLS bypasses and production database errors, necessitating its verification.",
   "agent": "Replit Agent",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/replit/comments/1pcl58p/my_first_serious_replit_build_a_production",
   "prompt": "Help me build a multi-tenant SaaS. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 20,
   "page": "https://agentsightings.com/case/build-a-multi-tenant-saas.html"
  },
  {
   "id": "automate-lead-research-crm-updates-and-enrichment",
   "title": "Automate lead research, CRM updates and enrichment",
   "summary": "Linked example of automate lead research, CRM updates and enrichment. Commenters described agent workflows for prospecting-adjacent work rather than pure copywriting.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/Coldemailing/comments/1vbwgt8/anyone_actually_using_claude_code_agents_in_their",
   "prompt": "GTM task: automate lead research, CRM updates and enrichment. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 20,
   "page": "https://agentsightings.com/case/automate-lead-research-crm-updates-and-enrichment.html"
  },
  {
   "id": "audit-a-blog-x27-s-seo",
   "title": "Audit a blog's SEO",
   "summary": "Case study takeaway: claude Code shows up in a write-up about audit a blog's SEO. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://alexop.dev/posts/how-i-use-claude-code-for-doing-seo-audits/",
   "prompt": "GTM task: audit a blog's SEO. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 20,
   "page": "https://agentsightings.com/case/audit-a-blog-x27-s-seo.html"
  },
  {
   "id": "split-a-react-native-to-swift-migration-across-agents",
   "title": "Split a React Native to Swift migration across agents",
   "summary": "From the linked report, jared instructed the master Devin to split the React Native to Swift migration into six independent pieces. The master then spawned child Devins, each assigned one piece, and later pulled their outputs together. The migration completed via parallel child agents.",
   "agent": "Devin",
   "topic": "Productivity",
   "source": "X",
   "url": "https://x.com/imjaredz/status/2068001458205720751",
   "prompt": "Help me split a React Native to Swift migration across agents. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/split-a-react-native-to-swift-migration-across-agents.html"
  },
  {
   "id": "search-literature-and-write-notes-into-zotero",
   "title": "Search literature and write notes into Zotero",
   "summary": "As described in the source, pin Shuai runs Claude Code with the Paper Search MCP to find top-cited papers and the papers citing them. A Zotero MCP, backed by Zotero's local API and a search database, lets Claude query his library in natural language. The note appears in Zotero when he opens it, and library search saves time on hundreds of entries.",
   "agent": "Claude",
   "topic": "Knowledge & Notes",
   "source": "Blog",
   "url": "https://groundwater.usu.edu/blog/2026/Academic-Usecase-with-Claude/",
   "prompt": "Notes task: search literature and write notes into Zotero. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/search-literature-and-write-notes-into-zotero.html"
  },
  {
   "id": "migrate-a-password-reset-flow-to-modern-libraries",
   "title": "Migrate a password reset flow to modern libraries",
   "summary": "According to the write-up, powered by work API tokens, the developer first assembled a plan with the Claude Code harness, then had Claude Code with the latest Opus migrate a web app's password reset flow from legacy libraries and architecture to modern ones. He supervised closely, reviewing every proposal, and afterward spent significant time cleaning up the final output. Code was decent at small scale but architecturally off, and it drifted into unrelated changes.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50001310",
   "prompt": "Notes task: migrate a password reset flow to modern libraries. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/migrate-a-password-reset-flow-to-modern-libraries.html"
  },
  {
   "id": "hunt-business-logic-flaws-and-open-fix-prs",
   "title": "Hunt business-logic flaws and open fix PRs",
   "summary": "As described in the source, parallel agents each take a segment of the codebase and reason across files for business-logic flaws, chained auth bypasses and cross-service exploit paths. Devin then combines the findings into full attack paths and tries to reproduce each one in an isolated sandbox. On Cognition's own 50-vulnerability benchmark, Devin Security scored 72% recall at $90.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://cognition.com/blog/introducing-devin-security-swarm",
   "prompt": "In this repo, hunt business-logic flaws and open fix PRs. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/hunt-business-logic-flaws-and-open-fix-prs.html"
  },
  {
   "id": "get-microsoft-office-running-on-linux-via-wine",
   "title": "Get Microsoft Office running on Linux via Wine",
   "summary": "One practitioner notes that the repo is a Nix flake that runs click-to-run Microsoft 365 through umu-launcher and GE-Proton instead of stock Wine. Each blocker was located by reading Wine and Click-to-Run logs and fixed with a small shim DLL or registry change. The full suite installs and Word starts, and sign-in with a personal account works.",
   "agent": "Claude Code",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49746401",
   "prompt": "Notes task: get Microsoft Office running on Linux via Wine. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/get-microsoft-office-running-on-linux-via-wine.html"
  },
  {
   "id": "explore-code-to-write-technical-documentation",
   "title": "Explore code to write technical documentation",
   "summary": "Real-world thread: fern's only technical writer, Devin Logan, uses the Devin AI engineer, which already has context on all of Fern's repos, to explore code before writing. She also tags Fern Writer, a Slack agent assembled on Devin, with an engineering thread, and it opens a PR with drafted pages, reorganized content and redirects. She reports about 20,000 new lines of docs per month across roughly 500 pages.",
   "agent": "Devin",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://buildwithfern.com/post/ai-assisted-technical-writer",
   "prompt": "Help me explore code to write technical documentation. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/explore-code-to-write-technical-documentation.html"
  },
  {
   "id": "build-dashboards-and-troubleshoot-support-cases",
   "title": "Build dashboards and troubleshoot support cases",
   "summary": "Linked example of build dashboards and troubleshoot support cases. Commenters describe Microsoft Scout as a desktop agent tied into WorkIQ and Microsoft 365 connectors like email, Teams, SharePoint, Word, and PowerPoint, with support for custom skills and MCPs via the same .copilot directory as Copilot CLI.",
   "agent": "Microsoft Scout",
   "topic": "Knowledge & Notes",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/microsoft_365_copilot/comments/1tyafym/any_experience_of_microsoft_scout",
   "prompt": "Notes task: build dashboards and troubleshoot support cases. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/build-dashboards-and-troubleshoot-support-cases.html"
  },
  {
   "id": "audit-hundreds-of-pull-requests",
   "title": "Audit hundreds of pull requests",
   "summary": "Case study takeaway: meta Muse covered audit hundreds of pull requests. as described in the source: muse Code conducted an independent test by auditing 222 pull requests in less than five minutes for approximately ten cents, as also developing a plan on a non-existent Google project.",
   "agent": "Meta Muse",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://codersera.com/blog/muse-code-complete-guide-2026/",
   "prompt": "Help me audit hundreds of pull requests. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 19,
   "page": "https://agentsightings.com/case/audit-hundreds-of-pull-requests.html"
  },
  {
   "id": "write-review-and-document-code-across-a-product-x27-s-prs",
   "title": "Write, review and document code across a product's PRs",
   "summary": "One practitioner notes that according to founders, Devin is responsible for handling all ChatPRD PRs and is the top contributor to Gumroad's merged PRs, which includes over 1,500 PRs.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/a-free-year-of-devin-the-worlds-first",
   "prompt": "In this repo, write, review and document code across a product's PRs. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/write-review-and-document-code-across-a-product-x27-s-prs.html"
  },
  {
   "id": "write-a-linux-kernel-improvement",
   "title": "Write a Linux kernel improvement",
   "summary": "A published walkthrough shows according to the headline, Linus Torvalds is working on a large patch for the Linux kernel powered by Claude Code.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50050367",
   "prompt": "In this repo, write a Linux kernel improvement. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/write-a-linux-kernel-improvement.html"
  },
  {
   "id": "use-codex-for-non-coding-chores",
   "title": "Use Codex for non-coding chores",
   "summary": "Documented elsewhere: people in the thread describe powered by Codex directly for ad hoc computer work rather than building full automations first. They had it handle research and scraping, search through files and emails, fill forms, align and edit PowerPoints, do light Excel work, build markdown or wiki knowledge bases, and walk them through chores like Pi-hole upgrades; one commenter says the key is giving detailed instructions, powered by planning first, and reviewing artifacts step by step. They report large time savings on tedious tasks and say Codex replaced hours of research, planning, and execution for some jobs.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/codex/comments/1ti49r0/noncoder_codex_users",
   "prompt": "In this repo, use Codex for non-coding chores. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/use-codex-for-non-coding-chores.html"
  },
  {
   "id": "test-how-well-coding-agents-can-onboard-to-your-product",
   "title": "Test how well coding agents can onboard to your product",
   "summary": "Case study takeaway: you enter your product's site on AX Check, which first checks the site itself. Three coding agents then try to onboard end to end in sandboxes. Results appear as scores out of 100 with a letter grade and a leaderboard.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49744416",
   "prompt": "In this repo, test how well coding agents can onboard to your product. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/test-how-well-coding-agents-can-onboard-to-your-product.html"
  },
  {
   "id": "team-playbook-for-claude-code-and-cursor",
   "title": "Team playbook for Claude Code and Cursor",
   "summary": "An engineering team shares a playbook: research before coding, plan mode, phased commits, and 3–5 parallel Claude Code sessions in git worktrees.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1ppxvgv/how_we_use_claude_code_and_cursor_at_work/",
   "prompt": "Before writing code, research the architecture in plan mode. Outline phases. Implement phase 1 only, then pause for review and commit before continuing.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/team-playbook-for-claude-code-and-cursor.html"
  },
  {
   "id": "take-ideas-to-implementation",
   "title": "Take ideas to implementation",
   "summary": "A published walkthrough shows Claude Code covered take ideas to implementation The linked post has the narrative detail.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49586883",
   "prompt": "In this repo, take ideas to implementation. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/take-ideas-to-implementation.html"
  },
  {
   "id": "several-coding-agents-in-parallel",
   "title": "Several coding agents in parallel",
   "summary": "A developer runs multiple coding agents at once on separate low-risk tasks and reviews their output as it lands.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://simonwillison.net/2025/Oct/5/parallel-coding-agents/",
   "prompt": "Take this small, well-defined task, work on its own branch, and open a draft PR with a short description and test results.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/several-coding-agents-in-parallel.html"
  },
  {
   "id": "self-driving-docs-via-playwright-exploration",
   "title": "Self-driving docs via Playwright exploration",
   "summary": "A team lets Claude Code drive Playwright to explore their product, find documentation gaps, and draft the missing pages.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Explore the app with Playwright as a new user. Compare what you find to docs/, list knowledge gaps, and draft PR-ready markdown for the three biggest gaps.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/self-driving-docs-via-playwright-exploration.html"
  },
  {
   "id": "run-real-estate-transactions",
   "title": "Run real estate transactions",
   "summary": "According to the write-up, claude shows up in a write-up about run real estate transactions. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1r8nssm/built_an_autonomous_real_estate_platform_with",
   "prompt": "In this repo, run real estate transactions. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/run-real-estate-transactions.html"
  },
  {
   "id": "run-agents-that-work-jira-tickets-review-prs-and-send-updates",
   "title": "Run agents that work Jira tickets, review PRs and send updates",
   "summary": "Community report: Hermes Agent was pointed at run agents that work Jira tickets, review PRs and send updates. luke manages a team of Hermes agents who are responsible for managing Jira tickets, writing code, reviewing PRs, tracking costs, presenting condensed documents, and sending updates.",
   "agent": "Hermes Agent",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/iamlukethedev/status/2071445022445027574",
   "prompt": "In this repo, run agents that work Jira tickets, review PRs and send updates. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/run-agents-that-work-jira-tickets-review-prs-and-send-updates.html"
  },
  {
   "id": "process-and-query-email-across-providers-via-mcp-agents",
   "title": "Process and query email across providers via MCP agents",
   "summary": "Community report: the developer spent six months writing the codebase with help from Cursor, Codex and Claude Code. The system runs in the cloud and connects to Gmail and Outlook mailboxes. The codebase is largely done and running in private beta.",
   "agent": "Cursor",
   "topic": "Email & Inbox",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49743163",
   "prompt": "Go through my mail for this goal: process and query email across providers via MCP agents. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/process-and-query-email-across-providers-via-mcp-agents.html"
  },
  {
   "id": "pair-program",
   "title": "Pair-program",
   "summary": "Case study takeaway: claude shows up in a write-up about pair-program. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49605997",
   "prompt": "In this repo, pair-program. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/pair-program.html"
  },
  {
   "id": "group-outdated-packages-by-upgrade-difficulty",
   "title": "Group outdated packages by upgrade difficulty",
   "summary": "Before a big dependency upgrade, Kent asked Cursor to list every outdated package grouped into easy, medium and hard buckets.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks",
   "prompt": "List every package in this project that is not on the latest version, grouped by upgrade difficulty (easy patch/minor, medium single major, hard multi-major or framework-critical). Include current and latest versions.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/group-outdated-packages-by-upgrade-difficulty.html"
  },
  {
   "id": "generate-live-coding-music",
   "title": "Generate live-coding music",
   "summary": "Case study takeaway: rearranged it for live performance, with as described in the source: claude crafted the code for the Strudel music based on a style description and fixed it by hand.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://thoughtwax.com/2025/06/live-coding/",
   "prompt": "In this repo, generate live-coding music. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/generate-live-coding-music.html"
  },
  {
   "id": "fix-and-resubmit-a-rejected-app-remotely",
   "title": "Fix and resubmit a rejected app remotely",
   "summary": "Dots covered fix and resubmit a rejected app remotely. User asked their dot to investigate the rejected app submission.",
   "agent": "Dots",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/maakcode/status/2106406113268891912",
   "prompt": "In this repo, fix and resubmit a rejected app remotely. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/fix-and-resubmit-a-rejected-app-remotely.html"
  },
  {
   "id": "debug-overnight",
   "title": "Debug overnight",
   "summary": "As described in the source, codex shows up in a write-up about debug overnight. The linked post is the primary source for steps and outcomes.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46993231",
   "prompt": "In this repo, debug overnight. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/debug-overnight.html"
  },
  {
   "id": "control-robots",
   "title": "Control robots",
   "summary": "As described in the source, gemini shows up in a write-up about control robots, with figures like 10. The linked post is the primary source for steps and outcomes.",
   "agent": "Gemini",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://developers.googleblog.com/gemini-25-for-robotics-and-embodied-intelligence/",
   "prompt": "In this repo, control robots. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/control-robots.html"
  },
  {
   "id": "build-a-mental-models-framework",
   "title": "Build a mental models framework",
   "summary": "As described in the source, claude shows up in a write-up about build a mental models framework, with figures like 120. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46061679",
   "prompt": "In this repo, build a mental models framework. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/build-a-mental-models-framework.html"
  },
  {
   "id": "boring-chores-for-a-coding-agent",
   "title": "Boring chores for a coding agent",
   "summary": "A how-to collects mundane tasks, like file renames and conversions, that a coding agent handles well.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://www.groovypost.com/howto/claude-code-boring-chores/",
   "prompt": "Rename every file in this folder to a consistent lowercase-hyphen format and convert any .heic images to .jpg. Show me the plan first.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/boring-chores-for-a-coding-agent.html"
  },
  {
   "id": "automate-web-configuration-from-coding-sessions",
   "title": "Automate web configuration from coding sessions",
   "summary": "Linked example of automate web configuration from coding sessions. In the thread, one commenter says they use Claude in Chrome during coding to handle web configuration tasks they would otherwise do manually.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1r3lxpe/removed",
   "prompt": "In this repo, automate web configuration from coding sessions. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 18,
   "page": "https://agentsightings.com/case/automate-web-configuration-from-coding-sessions.html"
  },
  {
   "id": "send-a-daily-competitor-ad-research-report-to-slack",
   "title": "Send a daily competitor ad research report to Slack",
   "summary": "From the linked report, the Manus agent runs daily in the background. It mines data from target avatar locations, segments findings into angles, concepts, failures and awareness maps, scrapes competitors’ ad libraries for text and headlines, benchmarks against the mined data to spot underserved areas, compiles a master report and posts it to Slack. It automates research and idea delivery to creative strategists, increasing their output.",
   "agent": "Manus",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/cjfeth/status/2034220127042834715",
   "prompt": "Research brief: send a daily competitor ad research report to Slack. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 17,
   "page": "https://agentsightings.com/case/send-a-daily-competitor-ad-research-report-to-slack.html"
  },
  {
   "id": "generate-structured-company-research-reports",
   "title": "Generate structured company research reports",
   "summary": "A published walkthrough shows dharmesh assembled the agent at companyresearch.ai and gave it access to multiple paid and proprietary data sources. The agent aggregates firmographics, top executives, web traffic, search marketing data, and G2 looks over into one structured report when Someone enters any company name. Reports cover revenue multiples and revenue per employee for public companies.",
   "agent": "Custom agent",
   "topic": "Research & Data",
   "source": "X",
   "url": "https://x.com/dharmesh/status/1988279749706207436",
   "prompt": "Research brief: generate structured company research reports. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 17,
   "page": "https://agentsightings.com/case/generate-structured-company-research-reports.html"
  },
  {
   "id": "voice-memo-to-prd-to-shipped-feature",
   "title": "Voice memo to PRD to shipped feature",
   "summary": "A founder records walking voice memos, turns them into a PRD with Claude, plans in Claude Code on a phone, then finishes surgical work in Cursor.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://drew.tech/posts/my-ai-coding-workflow",
   "prompt": "Here is a raw voice-memo transcript and links to related code. Explore the repo for existing patterns, draft a short PRD, then write PLAN.md with files to touch. Do not implement yet.",
   "added": "2026-10-15",
   "interest": 16,
   "page": "https://agentsightings.com/case/voice-memo-to-prd-to-shipped-feature.html"
  },
  {
   "id": "manage-canvas-courses-in-plain-english",
   "title": "Manage Canvas courses in plain English",
   "summary": "Linked example of manage Canvas courses in plain English. A non-developer educator used Claude Code over four months to build an MCP server in Python on FastMCP, starting from a 200-line proof of concept that listed courses.",
   "agent": "Claude Code",
   "topic": "Productivity",
   "source": "Blog",
   "url": "https://chatwithgpt.substack.com/p/when-claude-code-becomes-your-codeveloper",
   "prompt": "Help me manage Canvas courses in plain English. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 16,
   "page": "https://agentsightings.com/case/manage-canvas-courses-in-plain-english.html"
  },
  {
   "id": "turn-one-goal-into-an-ordered-plan-of-coding-agent-tasks",
   "title": "Turn one goal into an ordered plan of coding-agent tasks",
   "summary": "According to the write-up, a planner, which can be a coding agent running read-only, researches the repo and asks clarifying questions. It then commits a plan where each task has its own runner, model, effort and mode. In the README's example, a rate-limiting goal became a four-task plan (three AI, one manual) that finished with 4 completed, 0 failed.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49712276",
   "prompt": "In this repo, turn one goal into an ordered plan of coding-agent tasks. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/turn-one-goal-into-an-ordered-plan-of-coding-agent-tasks.html"
  },
  {
   "id": "slack-bug-report-to-linear-ticket-and-fix",
   "title": "Slack bug report to Linear ticket and fix",
   "summary": "When a bug lands in Slack, an automation checks for duplicates, opens a Linear issue, investigates the codebase, attempts a fix, and replies in-thread.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "This Slack bug report just arrived. Check Linear for duplicates, create an issue if new, investigate the root cause in the repo, attempt a fix on a branch, and reply in the thread with a summary.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/slack-bug-report-to-linear-ticket-and-fix.html"
  },
  {
   "id": "run-parallel-coding-agents-from-markdown-specs-in-tmux",
   "title": "Run parallel coding agents from Markdown specs in tmux",
   "summary": "According to the write-up, manuel Schipper runs vanilla Claude Code agents in separate tmux windows, labeled Planner, Worker or PM. Planners write numbered Markdown specs (Feature Designs) covering the problem, options, plan and verification steps. He covers 4–8 agents at once, and has written 300+ specs in one project.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47218318",
   "prompt": "In this repo, run parallel coding agents from Markdown specs in tmux. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/run-parallel-coding-agents-from-markdown-specs-in-tmux.html"
  },
  {
   "id": "run-a-personal-assistant-that-handles-repo-issues-and-groceries",
   "title": "Run a personal assistant that handles repo issues and groceries",
   "summary": "As described in the source, custom agent shows up in a write-up about run a personal assistant that handles repo issues and groceries. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49766401",
   "prompt": "In this repo, run a personal assistant that handles repo issues and groceries. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/run-a-personal-assistant-that-handles-repo-issues-and-groceries.html"
  },
  {
   "id": "let-terminal-agents-make-phone-calls",
   "title": "Let terminal agents make phone calls",
   "summary": "Documented elsewhere: the builder assembled a CLI tool powered by Twilio and GPT-live that gives terminal agents a voice for outbound calls. The agent operates independently on the line until it receives a question it cannot answer, at which point the included dashboard HUD prompts the person for an answer and redirects it back into the call without disconnecting. The result is a functional calling experience shown in a live QA and coding walkthrough video, with the agent handling calls and escalating unknowns to the person.",
   "agent": "Make",
   "topic": "DevOps & Infra",
   "source": "X",
   "url": "https://x.com/KingBootoshi/status/2105446245464732006",
   "prompt": "Infra task — let terminal agents make phone calls. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/let-terminal-agents-make-phone-calls.html"
  },
  {
   "id": "find-a-linux-vulnerability-hidden-for-23-years",
   "title": "Find a Linux vulnerability hidden for 23 years",
   "summary": "As described in the source, nicholas Carlini, a research scientist at Anthropic, ran a short script that looped over every file in the Linux kernel source. For each file, it told Claude Code it was in a CTF, hinted the bug was in that file, and asked it to write its most serious finding to an output directory. Claude Opus 4.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47633855",
   "prompt": "In this repo, find a Linux vulnerability hidden for 23 years. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/find-a-linux-vulnerability-hidden-for-23-years.html"
  },
  {
   "id": "coordinator-agent-that-hands-off-to-coding-agents",
   "title": "Coordinator agent that hands off to coding agents",
   "summary": "A conversational agent runs todos, cron jobs, research and Obsidian notes, delegating repo work to IDE and CLI coding agents.",
   "agent": "Hermes",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://vishsubramanian.me/my-agentic-development-workflow-as-of-july-2026/",
   "prompt": "Research the latest changelogs for the tools in my stack, summarize what changed this week, and add the summary to today's note.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/coordinator-agent-that-hands-off-to-coding-agents.html"
  },
  {
   "id": "coordinate-sub-agents-to-monitor-prs-and-fix-prod-bugs",
   "title": "Coordinate sub-agents to monitor PRs and fix prod bugs",
   "summary": "Observed in the wild: the Cursor Projects coordinator agent received overall project context and intent. On its own it monitored fresh PRs and production issues or bugs, then spun up sub-agents to handle fixes while delegating context across them. After finishing one large project Someone described the experience as a collaborative marathon that ended too soon and preferred it to earlier agent workflows.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/chsrbrts/status/2107637980504723550",
   "prompt": "In this repo, coordinate sub-agents to monitor PRs and fix prod bugs. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/coordinate-sub-agents-to-monitor-prs-and-fix-prod-bugs.html"
  },
  {
   "id": "build-and-publish-a-bazi-chart-reading-agent",
   "title": "Build and publish a BaZi chart reading agent",
   "summary": "A published walkthrough shows the developer used AI coding agents to build the full stack solo. The agent received a deterministic BaZi calendar engine to compute exact charts from birth data powered by solar terms, stem-branch calendar and 10-year luck cycles, then interpreted results while auto-detecting language for output plus chart image. It went live on the OKX AI Marketplace as Agent #4939 and settled over 20 real transactions.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/Ducktumay/status/2076933094628876322",
   "prompt": "In this repo, build and publish a BaZi chart reading agent. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/build-and-publish-a-bazi-chart-reading-agent.html"
  },
  {
   "id": "always-on-home-server-assistant",
   "title": "Always-on home-server assistant",
   "summary": "A Claude-based agent running as a service on a mini PC delivers a morning briefing, watches email and tracks finances with long-term memory.",
   "agent": "Claude",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47132125",
   "prompt": "Each morning at 7:00, send me a briefing: today's calendar, emails that need attention, notable account balance changes, and anything I asked you to remember for today.",
   "added": "2026-10-15",
   "interest": 15,
   "page": "https://agentsightings.com/case/always-on-home-server-assistant.html"
  },
  {
   "id": "delegate-web-research-from-claude-code-to-comet",
   "title": "Delegate web research from Claude Code to Comet",
   "summary": "Observed in the wild: you add the comet-mcp server to Claude Code's MCP config and install the Perplexity Comet browser. The server launches Comet with remote debugging and talks to it over CDP. The source gives no benchmarks or measured results.",
   "agent": "Claude Code",
   "topic": "Research & Data",
   "source": "GitHub",
   "url": "https://github.com/hanzili/comet-mcp",
   "prompt": "Research brief: delegate web research from Claude Code to Comet. Prefer primary sources, include links and confidence notes, and do not contact anyone.",
   "added": "2026-10-15",
   "interest": 14,
   "page": "https://agentsightings.com/case/delegate-web-research-from-claude-code-to-comet.html"
  },
  {
   "id": "work-on-mainframes-and-cobol",
   "title": "Work on mainframes and COBOL",
   "summary": "Case study takeaway: hopper is a desktop app that runs on your laptop and connects to z/OS over TN3270, plus optional FTP. You give the agent plain-English prompts. The page claims it turns hours of SDSF triage into one tag, which returns the abend code, failing step and source line.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48111143",
   "prompt": "In this repo, work on mainframes and COBOL. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 13,
   "page": "https://agentsightings.com/case/work-on-mainframes-and-cobol.html"
  },
  {
   "id": "use-a-bot-as-a-qa-intern-on-a-live-app",
   "title": "Use a bot as a QA intern on a live app",
   "summary": "Observed in the wild: the agent received access to a cloud computer pointed at one live app. It was instructed to click through core user flows several times a day like a new user, log anything broken or confusing, and draft fix PRs. In the first 48 hours it located a real signup flow bug on a specific untested screen size.",
   "agent": "Grok Bot",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/PrajwalTomar_/article/2088965811776147761",
   "prompt": "In this repo, use a bot as a QA intern on a live app. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 13,
   "page": "https://agentsightings.com/case/use-a-bot-as-a-qa-intern-on-a-live-app.html"
  },
  {
   "id": "run-brand-marketing-end-to-end-and-publish-to-social",
   "title": "Run brand marketing end to end and publish to social",
   "summary": "Community report: the poster says they assembled a CLI called Wonda that gives Claude Code a stable interface for end-to-end marketing work: scraping competitors, generating images and videos with tools covering Seedance, Sora, and Kling, editing with captions and music, and publishing to Instagram, TikTok, LinkedIn, X, and Reddit. In the comments, they say the team creates the core skills itself from patterns that worked across real campaigns, while human taste still matters. They report powered by 20+ AI models across campaigns and say one Instagram video reached 5.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1shi6gj/i_let_claude_code_run_marketing_for_real_brands",
   "prompt": "GTM task: run brand marketing end to end and publish to social. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 13,
   "page": "https://agentsightings.com/case/run-brand-marketing-end-to-end-and-publish-to-social.html"
  },
  {
   "id": "port-the-copilot-runtime-to-rust-agentically",
   "title": "Port the Copilot runtime to Rust agentically",
   "summary": "One practitioner notes that microsoft Leaned on Copilot, driven by several LLMs picked for their strengths, to rewrite the TypeScript Copilot runtime in Rust one module at a time. The port was a direct replacement and did not restructure the runtime. Agents turned 430,000 lines of TypeScript into 800,000 lines of Rust over 14.",
   "agent": "GitHub Copilot",
   "topic": "Knowledge & Notes",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49773998",
   "prompt": "Notes task: port the Copilot runtime to Rust agentically. Propose edits as a checklist, apply after approval, and finish with a short handover.",
   "added": "2026-10-15",
   "interest": 13,
   "page": "https://agentsightings.com/case/port-the-copilot-runtime-to-rust-agentically.html"
  },
  {
   "id": "build-a-report-writing-program-for-real-estate-appraisals",
   "title": "Build a report-writing program for real estate appraisals",
   "summary": "Case study takeaway: a real estate appraiser in South Korea said they used Claude Code to build a report-writing program for appraisal work, despite having no coding or software background. Over about two months, they iterated heavily, spending a lot of tokens and moving past repeated frustrations, after an earlier attempt to build a similar system with professional developers had failed because the workflow was hard to communicate. They said the tool removed much of the repetitive reporting work, coworkers located it impressive, and the firm was likely to adopt it officially.",
   "agent": "Claude",
   "topic": "Productivity",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1u3lh6r/im_a_real_estate_appraiser_and_claude_has_made_my",
   "prompt": "Help me build a report-writing program for real estate appraisals. Break it into steps, show your work, and wait for confirmation before anything irreversible, paid, or outbound.",
   "added": "2026-10-15",
   "interest": 13,
   "page": "https://agentsightings.com/case/build-a-report-writing-program-for-real-estate-appraisals.html"
  },
  {
   "id": "xstate-migration-with-looked-up-docs",
   "title": "XState migration with looked-up docs",
   "summary": "Kent had Cursor look up XState migration docs, plan the upgrade, and migrate about 200 lines of recorder state-machine code carefully in a fresh conversation.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks",
   "prompt": "Update xstate and @xstate/react. Before editing, look up the official migration docs and follow them. Keep behavior identical and list any API changes you made.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/xstate-migration-with-looked-up-docs.html"
  },
  {
   "id": "surgical-cursor-fixes-after-claude-code-draft",
   "title": "Surgical Cursor fixes after Claude Code draft",
   "summary": "Drew switches to Cursor with tight prompts when Claude Code stalls on odd codebase patterns, then returns to Claude Code for polish and Playwright tests.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://drew.tech/posts/my-ai-coding-workflow",
   "prompt": "Fix only this function. Here are the API docs for the return values. Do not refactor callers. Show the diff and how to verify.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/surgical-cursor-fixes-after-claude-code-draft.html"
  },
  {
   "id": "security-review-on-every-push-to-main",
   "title": "Security review on every push to main",
   "summary": "Cursor's own security automation audits diffs after merge to main, skips already-discussed issues, and posts high-risk findings to Slack.",
   "agent": "Cursor",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "On each push to main, audit the diff for security vulnerabilities. Skip issues already discussed on the PR. Post only high-risk findings to #security with file paths and a short remediation note.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/security-review-on-every-push-to-main.html"
  },
  {
   "id": "run-routine-store-checks-on-delayed-orders-and-stock",
   "title": "Run routine store checks on delayed orders and stock",
   "summary": "Linked example of run routine store checks on delayed orders and stock. The thread describes a store owner connecting Claude to Zendrop through its MCP and powered by it for routine read-style store checks rather than autonomous decision-making.",
   "agent": "Claude",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/dropshipping/comments/1vmqtd0/whats_a_good_ai_assistant_to_connect_to_my_store",
   "prompt": "Infra task — run routine store checks on delayed orders and stock. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/run-routine-store-checks-on-delayed-orders-and-stock.html"
  },
  {
   "id": "run-a-multi-bot-team-that-opens-draft-prs",
   "title": "Run a multi-bot team that opens draft PRs",
   "summary": "Think assembly line, not single chat: research bot, classifier, builder, tester, reviewer, security. Each stage narrows context, then a draft PR appears. Operator’s tally: four digits of PRs cleared.",
   "agent": "Grok Bot",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/0xMorlex/status/2106722078544007517",
   "prompt": "Stand up a multi-role bot team on this repo: research context, classify risk, patch minimally, run checks, review the diff, then open a draft PR. Pause before merge or any secret-touching change.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/run-a-multi-bot-team-that-opens-draft-prs.html"
  },
  {
   "id": "reverse-engineer-claude-desktop-to-find-a-path-to-linux-support",
   "title": "Reverse engineer Claude Desktop to find a path to Linux support",
   "summary": "Observed in the wild: claude Code shows up in a write-up about reverse engineer Claude Desktop to find a path to Linux support. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://aaddrick.com/blog/reverse-engineering-claude-desktops-cowork-mode-a-deep-dive-into-vm-isolation-and-linux-possibilities",
   "prompt": "Infra task — reverse engineer Claude Desktop to find a path to Linux support. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/reverse-engineer-claude-desktop-to-find-a-path-to-linux-support.html"
  },
  {
   "id": "plan-then-execute-coding-with-stacked-review-passes",
   "title": "Plan-then-execute coding with stacked review passes",
   "summary": "A developer plans features into markdown files, has the agent execute them, and runs several review skills on every diff before the PR.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://www.josecasanova.com/blog/how-i-use-ai-coding-agents-daily-workflow",
   "prompt": "Write an implementation plan for this feature to docs/plans/: scope, approach, files touched, risks. Don't write code yet.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/plan-then-execute-coding-with-stacked-review-passes.html"
  },
  {
   "id": "hand-off-coding-sessions-between-agent-accounts-at-usage-limits",
   "title": "Hand off coding sessions between agent accounts at usage limits",
   "summary": "Community report: codex shows up in a write-up about hand off coding sessions between agent accounts at usage limits. The linked post is the primary source for steps and outcomes.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49973683",
   "prompt": "In this repo, hand off coding sessions between agent accounts at usage limits. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/hand-off-coding-sessions-between-agent-accounts-at-usage-limits.html"
  },
  {
   "id": "finish-abandoned-side-projects-with-agents",
   "title": "Finish abandoned side projects with agents",
   "summary": "HN readers describe finally finishing long-backlog personal tools—music players, trackers, niche utilities—because agents remove the boring glue work.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47041973",
   "prompt": "Read this half-finished side project and the README wishlist. Propose the smallest path to a usable daily-driver version, then implement the first three items with tests.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/finish-abandoned-side-projects-with-agents.html"
  },
  {
   "id": "find-local-pilot-customers-from-the-codebase",
   "title": "Find local pilot customers from the codebase",
   "summary": "A builder asked Claude Code to inspect their app source, identify what it does, and name the top five nearby companies that would make good pilots.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Look at this project directory, summarize what the product does, then list the top 5 companies in <city> that would be good pilot customers. Include why each fits and a LinkedIn search query.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/find-local-pilot-customers-from-the-codebase.html"
  },
  {
   "id": "delegate-tiny-chores-to-devin-via-slack",
   "title": "Delegate tiny chores to Devin via Slack",
   "summary": "A developer fires Devin at small, well-scoped chores from Slack and reviews the resulting GitHub PRs at lunch and end of day.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://www.asad.pw/ai-agents-experiences-with-manus-operator-and-more/",
   "prompt": "Via Slack: fix this small lint/test failure on main, open a PR with the change, and do not expand scope beyond the failing check.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/delegate-tiny-chores-to-devin-via-slack.html"
  },
  {
   "id": "debug-code-hands-free-by-voice",
   "title": "Debug code hands-free by voice",
   "summary": "Real-world thread: claude Code shows up in a write-up about debug code hands-free by voice. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46899573",
   "prompt": "Infra task — debug code hands-free by voice. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/debug-code-hands-free-by-voice.html"
  },
  {
   "id": "code-in-the-terminal",
   "title": "Code in the terminal",
   "summary": "Documented elsewhere: grok Bot shows up in a write-up about code in the terminal. The linked post is the primary source for steps and outcomes.",
   "agent": "Grok Bot",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/AI_Agents/comments/1v3gtvc/grok_is_surprisingly_good_as_a_terminal_coding",
   "prompt": "Infra task — code in the terminal. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/code-in-the-terminal.html"
  },
  {
   "id": "claude-code-for-git-plans-and-codebase-tours",
   "title": "Claude Code for git, plans and codebase tours",
   "summary": "Calvin uses Claude Code daily to create plans, script local tasks, manage git/GitHub, and explain unfamiliar parts of the codebase via fast explore subagents.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://calv.info/agents-feb-2026",
   "prompt": "Explain how <subsystem> works in this repo. Use explore subagents if helpful. Give me a short architecture note and the five files I should read first.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/claude-code-for-git-plans-and-codebase-tours.html"
  },
  {
   "id": "agentic-codeowners-that-assign-reviewers",
   "title": "Agentic codeowners that assign reviewers",
   "summary": "An automation classifies PR risk, auto-approves low-risk changes, and assigns up to two reviewers for higher-risk work based on contribution history.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "On PR open, classify risk by blast radius, complexity and infra impact. Auto-approve low risk. For higher risk, assign up to two reviewers from contribution history and log the decision.",
   "added": "2026-10-15",
   "interest": 12,
   "page": "https://agentsightings.com/case/agentic-codeowners-that-assign-reviewers.html"
  },
  {
   "id": "spend-leftover-ai-quota-on-reviewable-prs",
   "title": "Spend leftover AI quota on reviewable PRs",
   "summary": "One practitioner notes that the founder points RemainderBot at a goals file and task-list sources. In cron mode it checks remaining subscription quota and time until the weekly reset, picks a task, works on a new branch, and opens a PR. It produced 111 PRs, covering customer call prep, a census of 31 GitHub orgs' agent setup files, and standalone prototypes for demos.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50073611",
   "prompt": "In this repo, spend leftover AI quota on reviewable PRs. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 11,
   "page": "https://agentsightings.com/case/spend-leftover-ai-quota-on-reviewable-prs.html"
  },
  {
   "id": "research-companies-and-fill-a-crm-for-outreach",
   "title": "Research companies and fill a CRM for outreach",
   "summary": "Case study takeaway: jonas Braadbaart first filters an Apollo company export by hand and has Claude Code score it against his ICP. For about 50 chosen companies, the agent researches each one through a custom Perplexity MCP server and creates company and people records in Attio. The first run contacted 80 companies and produced one sale.",
   "agent": "Claude Code",
   "topic": "Sales & Marketing",
   "source": "Blog",
   "url": "https://aimaker.substack.com/p/claude-code-ai-sales-agent",
   "prompt": "GTM task: research companies and fill a CRM for outreach. Produce drafts and a prioritized list only; no sends or publishes without approval.",
   "added": "2026-10-15",
   "interest": 11,
   "page": "https://agentsightings.com/case/research-companies-and-fill-a-crm-for-outreach.html"
  },
  {
   "id": "let-the-bot-email-customers-from-your-own-domain",
   "title": "Let the bot email customers from your own domain",
   "summary": "As described in the source, the poster located Grok Bot’s assembled-in mail too limited for customer-facing use, so they stood up a mailbox on their own domain and connected it to Grok as a custom MCP connector. Because incoming mail does not wake the bot automatically, they also added a webhook that starts a Grok automation on each new message; they note the webhook had to use Grok’s signed whsec_ secret format, not a bearer token. It gave them a customer-facing address on their own domain and an inbox they could open themselves, while still letting the bot react to mail as it arrived.",
   "agent": "Grok Bot",
   "topic": "Email & Inbox",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/GrokBot/comments/1x2iafb/grokbot_email_after_a_day_the_limits_nobody",
   "prompt": "Go through my mail for this goal: let the bot email customers from your own domain. Propose labels, drafts, and unsubscribes as a checklist; change nothing until I approve.",
   "added": "2026-10-15",
   "interest": 11,
   "page": "https://agentsightings.com/case/let-the-bot-email-customers-from-your-own-domain.html"
  },
  {
   "id": "pagerduty-incident-to-proposed-fix-pr",
   "title": "PagerDuty incident to proposed fix PR",
   "summary": "A PagerDuty trigger spins up a Cursor agent that checks Datadog logs and recent code changes, then Slack-posts a summary with a proposed fix PR.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "A PagerDuty incident just fired. Use Datadog to inspect related logs, search the repo for recent suspect changes, draft a fix on a branch, and post a Slack summary with the PR link.",
   "added": "2026-10-15",
   "interest": 10,
   "page": "https://agentsightings.com/case/pagerduty-incident-to-proposed-fix-pr.html"
  },
  {
   "id": "custom-terminal-agent-for-crud-scaffolding",
   "title": "Custom terminal agent for CRUD scaffolding",
   "summary": "An HN user runs a local DeepSeek-backed coding agent with only filesystem tools to scaffold routing, CRUD, auth, themes and admin dashboards in minutes.",
   "agent": "Custom agent",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48337719",
   "prompt": "Using only local file tools, scaffold a web app with auth, CRUD for <resource>, a basic theme and an admin list view. Show files created and how to run it.",
   "added": "2026-10-15",
   "interest": 10,
   "page": "https://agentsightings.com/case/custom-terminal-agent-for-crud-scaffolding.html"
  },
  {
   "id": "set-up-a-home-chat-server-and-manage-accounts-from-your-phone",
   "title": "Set up a home chat server and manage accounts from your phone",
   "summary": "One practitioner notes that hermes Agent shows up in a write-up about set up a home chat server and manage accounts from your phone. The linked post is the primary source for steps and outcomes.",
   "agent": "Hermes Agent",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48650822",
   "prompt": "Infra task — set up a home chat server and manage accounts from your phone. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 9,
   "page": "https://agentsightings.com/case/set-up-a-home-chat-server-and-manage-accounts-from-your-phone.html"
  },
  {
   "id": "run-varied-business-tasks-through-a-general-agent",
   "title": "Run varied business tasks through a general agent",
   "summary": "Community report: the reviewers gave Manus, a cloud-based agent that browses the web, writes and runs code, and manages files, 15 real tasks. They included a competitor spreadsheet, a travel itinerary with booking links, CSV analysis with charts, a landing page, a content calendar and a 50-page PDF summary. The article's table lists 10 tasks, and its stated split is 70% full success, 20% partial, 10% failure.",
   "agent": "Manus",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://justpickai.com/blog/manus-ai-review-2026",
   "prompt": "In this repo, run varied business tasks through a general agent. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 9,
   "page": "https://agentsightings.com/case/run-varied-business-tasks-through-a-general-agent.html"
  },
  {
   "id": "review-linux-kernel-patches",
   "title": "Review Linux kernel patches",
   "summary": "Community report: roman Gushchin's team at Google spent several months building Sashiko, an agentic code review system, and Leaned on it internally before release. It is now open source and covers every submission to the Linux kernel mailing list. On 1,000 recent upstream issues with Fixes: tags, it located about 53% of the bugs, and human reviewers had missed all of them.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47427647",
   "prompt": "In this repo, review Linux kernel patches. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 9,
   "page": "https://agentsightings.com/case/review-linux-kernel-patches.html"
  },
  {
   "id": "investigate-broken-builds-and-merge-fixes",
   "title": "Investigate broken builds and merge fixes",
   "summary": "As described in the source, androidclaw is a persistent OpenClaw agent at OpenAI. It works across company context, Git, GitHub and logging systems. Marsan said it has been well-adopted internally and called tracing issues and publishing fixes \"Kinda game changing.",
   "agent": "OpenClaw",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://openclaw.ai/blog/openclaw-enterprise",
   "prompt": "In this repo, investigate broken builds and merge fixes. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 9,
   "page": "https://agentsightings.com/case/investigate-broken-builds-and-merge-fixes.html"
  },
  {
   "id": "generate-ad-copy-variants-and-scrape-competitor-ads",
   "title": "Generate ad-copy variants and scrape competitor ads",
   "summary": "Linked example of generate ad-copy variants and scrape competitor ads. The commenter said they used the Claude API through Google Apps Script to generate ad-copy variants, with a workflow that produced 10 variants in about 30 seconds instead of a 30-minute brainstorm.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/digital_marketing/comments/1sywh5r/what_ai_tools_are_actually_helping_you_execute",
   "prompt": "In this repo, generate ad-copy variants and scrape competitor ads. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 9,
   "page": "https://agentsightings.com/case/generate-ad-copy-variants-and-scrape-competitor-ads.html"
  },
  {
   "id": "update-a-legacy-telegram-bot-and-refactor-an-app",
   "title": "Update a legacy Telegram bot and refactor an app",
   "summary": "Case study takeaway: for the Telegram bot, The builder located the Agent needed a project stood up first, so they spent $0.25 on a fresh bot and then added their legacy code. They asked it to copy the new API's functionality. The card-app refactor went fairly smoothly.",
   "agent": "Replit Agent",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://aiartisan.substack.com/p/what-i-built-with-replit-in-1-month",
   "prompt": "In this repo, update a legacy Telegram bot and refactor an app. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 7,
   "page": "https://agentsightings.com/case/update-a-legacy-telegram-bot-and-refactor-an-app.html"
  },
  {
   "id": "run-coding-agents-like-a-medical-team-to-fix-bugs",
   "title": "Run coding agents like a medical team to fix bugs",
   "summary": "Community report: cockroach Labs assembled a GitHub Actions pipeline in which labels on an issue trigger role-specific agents. A Fellow tries to reproduce the bug and posts a treatment plan, and a separate Review Attending approves or rejects it before any code is written. Over five months it merged 1,238 PRs and over a million lines of code with 7 reverts.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50021899",
   "prompt": "In this repo, run coding agents like a medical team to fix bugs. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 7,
   "page": "https://agentsightings.com/case/run-coding-agents-like-a-medical-team-to-fix-bugs.html"
  },
  {
   "id": "fix-a-broken-api-website",
   "title": "Fix a broken API website",
   "summary": "A published walkthrough shows the Reddit post says Someone used Claude Cowork on their laptop for technical tasks and let it inspect and control work directly, describing it as like a competent IT person remote-controlling the computer. For this case, they say it went into the code of their API website and fixed issues manually after Replit’s agent kept breaking simple additions and not understanding the site. The stated result was that the API website got fixed after repeated problems with the Replit agent.",
   "agent": "Claude Cowork",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rok3g8/claude_cowork_is_magical",
   "prompt": "In this repo, fix a broken API website. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 7,
   "page": "https://agentsightings.com/case/fix-a-broken-api-website.html"
  },
  {
   "id": "weekly-slack-digest-of-meaningful-repo-changes",
   "title": "Weekly Slack digest of meaningful repo changes",
   "summary": "A scheduled Cursor agent summarizes major merged PRs, bug fixes, debt work and security updates into a weekly Slack post.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://cursor.com/blog/automations",
   "prompt": "Summarize meaningful changes to this repository in the last seven days for Slack: major PRs, bug fixes, tech debt and security or dependency updates. Keep it under 400 words.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/weekly-slack-digest-of-meaningful-repo-changes.html"
  },
  {
   "id": "ship-more-pull-requests",
   "title": "Ship more pull requests",
   "summary": "Documented elsewhere: Devin was pointed at ship more pull requests. devin from iPhone, Slack, browser, GitHub, and Linear are employed by Ryan Carson for his actual work, and he claims to have seen a rise in PR shipping rates.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/ryancarson/status/2035746736803614981",
   "prompt": "In this repo, ship more pull requests. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/ship-more-pull-requests.html"
  },
  {
   "id": "run-terminal-commands",
   "title": "Run terminal commands",
   "summary": "According to the write-up, gitHub Copilot shows up in a write-up about run terminal commands. The linked post is the primary source for steps and outcomes.",
   "agent": "GitHub Copilot",
   "topic": "DevOps & Infra",
   "source": "GitHub",
   "url": "https://github.com/orgs/community/discussions/161238",
   "prompt": "Infra task — run terminal commands. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/run-terminal-commands.html"
  },
  {
   "id": "run-multi-stage-pr-reviews",
   "title": "Run multi-stage PR reviews",
   "summary": "A published walkthrough shows claude Code shows up in a write-up about run multi-stage PR reviews. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48090276",
   "prompt": "In this repo, run multi-stage PR reviews. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/run-multi-stage-pr-reviews.html"
  },
  {
   "id": "run-an-entire-business-from-the-terminal",
   "title": "Run an entire business from the terminal",
   "summary": "As described in the source, from the linked report: constructed a framework that manages their entire operation powered by Claude Code instead of Claude.ai, with specific tasks not specified in the introductory statement.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1rwmj25/i_stopped_using_claudeai_entirely_i_run_my_entire",
   "prompt": "Infra task — run an entire business from the terminal. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/run-an-entire-business-from-the-terminal.html"
  },
  {
   "id": "run-a-fleet-of-engineer-bots-that-fix-ci-and-ship-prs",
   "title": "Run a fleet of engineer bots that fix CI and ship PRs",
   "summary": "Lingxi runs five specialty Grok bots (mobile, desktop, Android, infra, harness) that spawn cloud coding agents, watch CI and security failures every half hour, and leave fresh PRs for morning review—over 200 agents at peak, far past a manual cap of ~15.",
   "agent": "Grok Bot",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/XFreeze/status/2098448967294173482",
   "prompt": "Coordinate a fleet of specialty coding bots on this org: scan for red CI, security issues, and merge conflicts every 30 minutes, send fix agents, and queue draft PRs for human review. No production deploys without approval.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/run-a-fleet-of-engineer-bots-that-fix-ci-and-ship-prs.html"
  },
  {
   "id": "route-linear-tickets-to-prs-across-multiple-repos",
   "title": "Route Linear tickets to PRs across multiple repos",
   "summary": "Real-world thread: the developer sends tasks through Telegram/OpenClaw into a single Linear account that acts as the control plane across several projects. In Cyrus, they set up labels and routing so it knows which git repo to build worktrees from. The setup lets the developer work more hands-off, with PRs arriving across multiple repos for review.",
   "agent": "OpenClaw",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49615716",
   "prompt": "In this repo, route Linear tickets to PRs across multiple repos. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/route-linear-tickets-to-prs-across-multiple-repos.html"
  },
  {
   "id": "refactor-a-million-line-repo-with-1-393-subagents",
   "title": "Refactor a million-line repo with 1,393 subagents",
   "summary": "As described in the source, teknium gave his everyday Hermes agent a standing goal with /goal, plus a dev skill it had built up over time.",
   "agent": "Hermes Agent",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://nousresearch.com/refactoring-hermes-with-1393-agents",
   "prompt": "In this repo, refactor a million-line repo with 1,393 subagents. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/refactor-a-million-line-repo-with-1-393-subagents.html"
  },
  {
   "id": "nightly-roadmap-chipping",
   "title": "Nightly roadmap chipping",
   "summary": "A Codex automation reads a product roadmap over MCP, estimates difficulty from the codebase, and ships one small fix or improvement each run.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://build.ms/2026/2/12/automating-codex/",
   "prompt": "Read the roadmap, pick the smallest item that can be safely finished today, implement it on a branch with tests, and summarize the change.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/nightly-roadmap-chipping.html"
  },
  {
   "id": "migrate-180-microservices-with-dual-cursor-agents",
   "title": "Migrate 180 microservices with dual Cursor agents",
   "summary": "Brex built generation and validation Cursor agents that draft PRs, fix CI, and run integration tests so one engineer migrated 180 Micronaut services in a month.",
   "agent": "Cursor",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://www.brex.com/journal/agent-automations-for-big-migrations",
   "prompt": "For ticket <id>, apply the Micronaut 4 migration scripts, fix local build errors using the matching skills, open a draft PR, then stop after three failed self-corrections and comment what blocked you.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/migrate-180-microservices-with-dual-cursor-agents.html"
  },
  {
   "id": "diagnose-why-a-mac-is-running-slowly",
   "title": "Diagnose why a Mac is running slowly",
   "summary": "A user has Claude Code check load averages, memory pressure, disk space, stuck processes and Docker or Time Machine caches, then explain fixes with risk notes.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "My Mac feels slow. Check load averages, memory pressure, disk free space, top CPU processes, Docker disk use and Time Machine local snapshots. Explain what is wrong and suggest fixes with the exact commands and risk level for each.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/diagnose-why-a-mac-is-running-slowly.html"
  },
  {
   "id": "debug-windows-crash-dumps",
   "title": "Debug Windows crash dumps",
   "summary": "Raw minidumps are painful; ForensicDbg predigests native x86/x64 crashes into linked findings and speaks MCP, so an agent can debug post-mortem without spelunking WinDbg by hand.",
   "agent": "Custom agent",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49821086",
   "prompt": "Infra task — debug Windows crash dumps. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/debug-windows-crash-dumps.html"
  },
  {
   "id": "build-data-analysis-reports-and-dashboards",
   "title": "Build data analysis reports and dashboards",
   "summary": "Documented elsewhere: custom agent shows up in a write-up about build data analysis reports and dashboards. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49927295",
   "prompt": "In this repo, build data analysis reports and dashboards. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/build-data-analysis-reports-and-dashboards.html"
  },
  {
   "id": "build-and-deploy-a-meeting-app-from-a-prompt",
   "title": "Build and deploy a meeting app from a prompt",
   "summary": "Linked example of build and deploy a meeting app from a prompt. User entered one prompt into Replit describing a meeting management app that tracks agendas, participants, and action items plus analytics for effectiveness, participation, and follow-up.",
   "agent": "Replit Agent",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/socialwithaayan/status/1986663983777259839",
   "prompt": "In this repo, build and deploy a meeting app from a prompt. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/build-and-deploy-a-meeting-app-from-a-prompt.html"
  },
  {
   "id": "build-a-c-compiler",
   "title": "Build a C compiler",
   "summary": "Observed in the wild: claude shows up in a write-up about build a C compiler, with figures like 4.6. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50068489",
   "prompt": "In this repo, build a C compiler. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/build-a-c-compiler.html"
  },
  {
   "id": "async-research-projects-with-coding-agents",
   "title": "Async research projects with coding agents",
   "summary": "Asynchronous coding agents are given open research questions, write and run experiments in a dedicated repo, and report findings.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://simonwillison.net/2025/Nov/6/async-code-research/",
   "prompt": "Investigate whether <library> can do <task>. Write small experiments, run them, and commit a README with findings, code and what didn't work.",
   "added": "2026-10-15",
   "interest": 6,
   "page": "https://agentsightings.com/case/async-research-projects-with-coding-agents.html"
  },
  {
   "id": "review-contracts-like-ndas-and-saas-agreements-in-claude-code",
   "title": "Review contracts like NDAs and SaaS agreements in Claude Code",
   "summary": "Case study takeaway: claude Code shows up in a write-up about review contracts like NDAs and SaaS agreements in Claude Code. The linked post is the primary source for steps and outcomes.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1r6tse1/i_built_a_free_contract_review_skill_for_claude",
   "prompt": "In this repo, review contracts like NDAs and SaaS agreements in Claude Code. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 4,
   "page": "https://agentsightings.com/case/review-contracts-like-ndas-and-saas-agreements-in-claude-code.html"
  },
  {
   "id": "make-auditable-rule-based-decisions",
   "title": "Make auditable rule-based decisions",
   "summary": "As described in the source, the Rete rule engine is responsible for determining loans, fraud, and triage cases, while an LLM and an MCP server are utilized for RAG over policy documents.",
   "agent": "Claude",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49803683",
   "prompt": "Infra task — make auditable rule-based decisions. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 4,
   "page": "https://agentsightings.com/case/make-auditable-rule-based-decisions.html"
  },
  {
   "id": "let-agents-edit-word-documents-via-mcp",
   "title": "Let agents edit Word documents via MCP",
   "summary": "According to the write-up, the agent edits a clean HTML version of the Word file instead of raw OOXML. A small fine-tuned model (3–8B, LoRA) turns each changed block's HTML into OOXML. On the vendor's own 279-task benchmark, it was 2.",
   "agent": "Custom agent",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49881505",
   "prompt": "Infra task — let agents edit Word documents via MCP. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 4,
   "page": "https://agentsightings.com/case/let-agents-edit-word-documents-via-mcp.html"
  },
  {
   "id": "turn-slack-feedback-into-prs-and-debug-infra",
   "title": "Turn Slack feedback into PRs and debug infra",
   "summary": "From the linked report, at Sid's last company a dedicated Slack feedback channel fed every user issue or request to Devin. Given read-only access, Devin triaged messages, let users chat directly with it to turn issues into PRs, and debugged async flows or infra on its own. Sid ranked top 200 global Devin users.",
   "agent": "Devin",
   "topic": "DevOps & Infra",
   "source": "X",
   "url": "https://x.com/sidstronaut/status/2105417679691485630",
   "prompt": "Infra task — turn Slack feedback into PRs and debug infra. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 3,
   "page": "https://agentsightings.com/case/turn-slack-feedback-into-prs-and-debug-infra.html"
  },
  {
   "id": "surface-unanswered-emails-from-the-past-week",
   "title": "Surface unanswered emails from the past week",
   "summary": "One practitioner notes that the builder wanted to recreate Affinity's Unanswered Emails tab powered by the Gmail API. Devin struggled with authentication, so he wrote that himself in Cursor. It produced a working function, a front end, and a public deployment in one late-night session.",
   "agent": "Devin",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://jamesin.substack.com/p/hello-world-meet-devin",
   "prompt": "In this repo, surface unanswered emails from the past week. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 3,
   "page": "https://agentsightings.com/case/surface-unanswered-emails-from-the-past-week.html"
  },
  {
   "id": "read-and-edit-word-excel-and-powerpoint-files",
   "title": "Read and edit Word, Excel, and PowerPoint files",
   "summary": "Documented elsewhere: officeCLI is a single binary with no Office install needed. Running its install command adds a skill to detected agents such as Claude Code, Cursor, Windsurf and GitHub Copilot. The README shows sample decks, Word documents and spreadsheets that it says agents made with no templates or manual editing.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48807225",
   "prompt": "In this repo, read and edit Word, Excel, and PowerPoint files. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 3,
   "page": "https://agentsightings.com/case/read-and-edit-word-excel-and-powerpoint-files.html"
  },
  {
   "id": "patch-annoying-software-behavior-by-pointing-an-agent-at-it",
   "title": "Patch annoying software behavior by pointing an agent at it",
   "summary": "Community report: the builder pointed coding agents at software that annoyed them. For vmconnect.exe not going fullscreen after login, the agent debugged the problem and assembled a hook system that leaves Microsoft's files untouched. Both fixes reportedly worked, though a few rounds of telling the model to do better were sometimes needed.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50030987",
   "prompt": "In this repo, patch annoying software behavior by pointing an agent at it. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 3,
   "page": "https://agentsightings.com/case/patch-annoying-software-behavior-by-pointing-an-agent-at-it.html"
  },
  {
   "id": "give-coding-agents-git-native-persistent-memory",
   "title": "Give coding agents git-native persistent memory",
   "summary": "Documented elsewhere: you build a Go binary, then run a bootstrap command on a project. It creates a knowledge/ folder of Markdown files with YAML frontmatter, an agent skill, and an AGENTS.md. The README claims search under 300µs, about 4ms to validate 50+ concepts, and an 80% token reduction.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49581240",
   "prompt": "In this repo, give coding agents git-native persistent memory. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 3,
   "page": "https://agentsightings.com/case/give-coding-agents-git-native-persistent-memory.html"
  },
  {
   "id": "find-exploitable-bugs-in-smart-contracts",
   "title": "Find exploitable bugs in smart contracts",
   "summary": "A published walkthrough shows researchers assembled SCONE-bench from 405 historically exploited contracts. Each agent ran in a Docker sandbox with a forked local blockchain, bash with the Foundry toolchain, and a file editor. Across the 405 contracts, the models exploited 207 in simulation.",
   "agent": "Claude",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46115214",
   "prompt": "Infra task — find exploitable bugs in smart contracts. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 3,
   "page": "https://agentsightings.com/case/find-exploitable-bugs-in-smart-contracts.html"
  },
  {
   "id": "triage-failed-github-actions-runs-and-self-heal-flaky-tests",
   "title": "Triage failed GitHub Actions runs and self-heal flaky tests",
   "summary": "As described in the source, the builder assembled an OpenClaw agent over a weekend that watches GitHub Actions for failed runs and reads the logs and test output. It classifies each failure as flaky, a real regression or an environment issue. Time-to-diagnosis fell from 45 to 6 minutes, and flakiness dropped from 8% to 0.",
   "agent": "OpenClaw",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://suneetmalhotra.com/blog/openclaw-production-ai-agent-runtime-2026",
   "prompt": "In this repo, triage failed GitHub Actions runs and self-heal flaky tests. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 1,
   "page": "https://agentsightings.com/case/triage-failed-github-actions-runs-and-self-heal-flaky-tests.html"
  },
  {
   "id": "source-and-screen-candidates-and-reply-over-whatsapp",
   "title": "Source and screen candidates and reply over WhatsApp",
   "summary": "From the linked report, openClaw runs on a dedicated machine such as a Mac Mini. A 6 AM booked task uses the browser skill to visit chosen GitHub repos, pull contributors, filter for recent activity and send a Telegram digest. The guide says API costs run roughly $10 to $30 a month for light use and $100 to $200 for heavy use.",
   "agent": "OpenClaw",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://www.glozo.com/blog/openclaw-for-recruiters",
   "prompt": "In this repo, source and screen candidates and reply over WhatsApp. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 1,
   "page": "https://agentsightings.com/case/source-and-screen-candidates-and-reply-over-whatsapp.html"
  },
  {
   "id": "vibe-code-a-pull-request-dashboard-across-azure-devops-repos",
   "title": "Vibe-code a pull request dashboard across Azure DevOps repos",
   "summary": "Architecture came from ChatGPT; Junie inside WebStorm then built an Azure DevOps PR dashboard across four repos—approvals, conflicts, comments—in roughly three hours, with YouTrack wired in and PATs kept local. The team adopted it the same day.",
   "agent": "ChatGPT",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46721215",
   "prompt": "Build a small dashboard that aggregates open PRs across these Azure DevOps repos, highlighting approvals, conflicts, and comment threads. Keep secrets local and show a working read-only view before any write APIs.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/vibe-code-a-pull-request-dashboard-across-azure-devops-repos.html"
  },
  {
   "id": "verify-spice-simulations-against-a-real-oscilloscope",
   "title": "Verify SPICE simulations against a real oscilloscope",
   "summary": "Documented elsewhere: lucas Gerads gave Claude Code two MCP servers: one for a LeCroy oscilloscope and one wrapping spicelib for SPICE simulation. The demo uses a trivial RC filter driven by a simple microcontroller. The post says the setup became very valuable for validating SPICE circuits and models, and it removed tedious data analysis such as time-axis normalization and alignment.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47801255",
   "prompt": "Infra task — verify SPICE simulations against a real oscilloscope. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/verify-spice-simulations-against-a-real-oscilloscope.html"
  },
  {
   "id": "validation-agent-that-stops-after-three-tries",
   "title": "Validation agent that stops after three tries",
   "summary": "Brex's PR validation agent deploys services, runs integration tests, patches failures up to three times, then leaves a human-readable blocker comment.",
   "agent": "Cursor",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://www.brex.com/journal/agent-automations-for-big-migrations",
   "prompt": "Monitor this PR's CI and deploy. If tests fail, apply the matching skill and retry up to three times. If still blocked, comment with root cause and stop. Never merge.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/validation-agent-that-stops-after-three-tries.html"
  },
  {
   "id": "upgrade-remix-to-react-router-overnight",
   "title": "Upgrade Remix to React Router overnight",
   "summary": "Kent C. Dodds pointed a Cursor background agent at migration docs, deleted old patches, fixed routing against route output, and woke up to a working upgrade PR.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks",
   "prompt": "Upgrade this project from Remix v2 to React Router v7. Look up the official migration guides first, delete obsolete patch files, verify with `npx react-router routes` against main, and open a PR.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/upgrade-remix-to-react-router-overnight.html"
  },
  {
   "id": "turn-tickets-into-merged-pull-requests",
   "title": "Turn tickets into merged pull requests",
   "summary": "Community report: optio is a self-hosted tool you run on Kubernetes. You assign it a GitHub Issue, Linear ticket or Jira card. The README says the pipeline runs end to end, squash-merging when CI is green and the review is done, then closing the linked issue and recording cost.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47520220",
   "prompt": "In this repo, turn tickets into merged pull requests. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/turn-tickets-into-merged-pull-requests.html"
  },
  {
   "id": "trade-stocks-via-robinhood-mcp",
   "title": "Trade stocks via Robinhood MCP",
   "summary": "Community report: the setup was a Robinhood MCP for an agentic trading challenge against $VOO. The ChatGPT agent sold TSM shares and bought HOOD shares. No results or shortfalls are described.",
   "agent": "ChatGPT Agent",
   "topic": "DevOps & Infra",
   "source": "X",
   "url": "https://x.com/dankulous/status/2085026696764227820",
   "prompt": "Infra task — trade stocks via Robinhood MCP. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/trade-stocks-via-robinhood-mcp.html"
  },
  {
   "id": "track-investments-over-time-via-claude-code-and-robinhood-mcp",
   "title": "Track investments over time via Claude Code and Robinhood MCP",
   "summary": "Documented elsewhere: claude Code covered track investments over time via Claude Code and Robinhood MCP. requests information on obtaining Claude Code's connection to Robinhood through MCP to continue monitoring investments following researching or trading; a question, not a conclusive conclusion.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/RobinhoodApp/comments/1vu1olv/tips_for_tracking_investments_over_time_with_rh",
   "prompt": "Infra task — track investments over time via Claude Code and Robinhood MCP. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/track-investments-over-time-via-claude-code-and-robinhood-mcp.html"
  },
  {
   "id": "test-devin-on-integrations-migration-and-security-review",
   "title": "Test Devin on integrations, migration and security review",
   "summary": "Observed in the wild: answer.AI gave Devin, which works through Slack in its own Docker environment with a browser, editor and shell, about 20 real tasks. These included generating and uploading synthetic data to Braintrust, migrating a Python project to nbdev, and a security review of a repo under 700 lines. The Braintrust integration was abandoned and redone in Cursor.",
   "agent": "Devin",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://www.answer.ai/posts/2025-01-08-devin",
   "prompt": "Infra task — test Devin on integrations, migration and security review. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/test-devin-on-integrations-migration-and-security-review.html"
  },
  {
   "id": "test-api-endpoints-in-postman-and-report-results",
   "title": "Test API endpoints in Postman and report results",
   "summary": "Real-world thread: user supplied Comet with context on the API payloads and details, then let it handle testing in Postman. Comet generated multiple test payloads, executed the endpoint tests itself, and produced a final report after the run. Comet completed the tests and delivered the report with no further input required.",
   "agent": "Perplexity Comet",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/ai_for_success/status/1995164144509784334",
   "prompt": "In this repo, test API endpoints in Postman and report results. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/test-api-endpoints-in-postman-and-report-results.html"
  },
  {
   "id": "split-ui-and-backend-across-desktop-agents",
   "title": "Split UI and backend across desktop agents",
   "summary": "A developer runs Codex desktop for backend and architecture while Claude Code desktop handles UI and Figma-to-code via MCP.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://dev.to/taruntomar122/my-actual-ai-coding-setup-in-2026-and-how-i-decide-what-to-use-for-what-44oh",
   "prompt": "Backend agent: design the API and data model for this feature. Frontend agent: implement the UI from the Figma MCP once the API contract is in api.md. Do not share a working tree.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/split-ui-and-backend-across-desktop-agents.html"
  },
  {
   "id": "ship-a-product-in-72-hours",
   "title": "Ship a product in 72 hours",
   "summary": "Seventy-two-hour crunch: three SpaceX engineers + Grok Bot fleet. Scoreboard read 433 pull requests from an empty repo; bug repro and merges were agent-led, while migrations/deploys stayed human-only.",
   "agent": "Grok Bot",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/1920web1080/status/2101658383958876623",
   "prompt": "In this empty repo, drive toward a shippable product over the next 72 hours. Prefer tiny PRs, reproduce failures before fixing, and stop for approval on migrations, deploys, or anything public.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/ship-a-product-in-72-hours.html"
  },
  {
   "id": "self-host-services-on-a-home-server",
   "title": "Self-host services on a home server",
   "summary": "Real-world thread: jordan Fulghum put Ubuntu Server and Tailscale on a Beelink mini PC and SSHed in from his laptop. He installed Claude Code on the server and described outcomes in plain language. Fourteen containers run on under 10% CPU and about a third of the memory.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46580326",
   "prompt": "Infra task — self-host services on a home server. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/self-host-services-on-a-home-server.html"
  },
  {
   "id": "run-sandboxed-claude-code-fleet-sharing-files-via-mcp",
   "title": "Run sandboxed Claude Code fleet sharing files via MCP",
   "summary": "Claude Code farm, each worker jailed. File I/O only via MCP to an agentfiles:// fake disk; the orchestrator alone talks to AWS. Credentials stay outside the sandboxes; every call is logged.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49782293",
   "prompt": "Design a sandboxed agent fleet that shares files only via MCP on a virtual filesystem. Keep cloud credentials in the orchestrator, log every tool call, and do not expose raw AWS access to the agents.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/run-sandboxed-claude-code-fleet-sharing-files-via-mcp.html"
  },
  {
   "id": "run-multi-agent-llm-financial-trading-analysis",
   "title": "Run multi-agent LLM financial trading analysis",
   "summary": "Real-world thread: you install the repo, set an API key for an LLM provider, and pick a ticker and date in the CLI or through the Python TradingAgentsGraph call. Analyst agents cover fundamentals, sentiment, news and technicals. The source reports no benchmark results.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49605822",
   "prompt": "In this repo, run multi-agent LLM financial trading analysis. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/run-multi-agent-llm-financial-trading-analysis.html"
  },
  {
   "id": "run-openclaw-bots-for-repo-upkeep-note-taking-and-team-slack-help",
   "title": "Run OpenClaw bots for repo upkeep, note-taking and team Slack help",
   "summary": "As described in the source, from the linked report: and in Slack summarize meeting notes and answer team questions, following runs OpenClaw bots that maintain a personal repo, take notes and send news summaries.",
   "agent": "OpenClaw",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47795202",
   "prompt": "In this repo, run OpenClaw bots for repo upkeep, note-taking and team Slack help. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/run-openclaw-bots-for-repo-upkeep-note-taking-and-team-slack-help.html"
  },
  {
   "id": "run-openclaw-agents-for-terraform-infra-and-gitops-from-telegram",
   "title": "Run OpenClaw agents for Terraform infra and GitOps from Telegram",
   "summary": "One practitioner notes that openClaw shows up in a write-up about run OpenClaw agents for Terraform infra and GitOps from Telegram. The linked post is the primary source for steps and outcomes.",
   "agent": "OpenClaw",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/openclaw/comments/1spb1jz/my_openclaw_agents_were_great_until",
   "prompt": "Infra task — run OpenClaw agents for Terraform infra and GitOps from Telegram. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/run-openclaw-agents-for-terraform-infra-and-gitops-from-telegram.html"
  },
  {
   "id": "run-llm-assistants-over-company-data",
   "title": "Run LLM assistants over company data",
   "summary": "From the linked report, n8n shows up in a write-up about run LLM assistants over company data, with figures like 8. The linked post is the primary source for steps and outcomes.",
   "agent": "n8n",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/n8n/comments/1r3qalu/most_ai_agent_workflows_are_just_glorified_ifelse",
   "prompt": "In this repo, run LLM assistants over company data. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/run-llm-assistants-over-company-data.html"
  },
  {
   "id": "review-github-pull-requests-for-bugs-and-security",
   "title": "Review GitHub pull requests for bugs and security",
   "summary": "Case study takeaway: with the gh CLI authenticated and the Hermes gateway running, you first ask Hermes in chat to fetch a PR diff and review it. You then add a code-review skill with a severity-based output format, and teach it team conventions through memory. The source gives no review results.",
   "agent": "Hermes Agent",
   "topic": "Software Development",
   "source": "X",
   "url": "https://x.com/AgentSparko/status/2106704313657569409",
   "prompt": "In this repo, review GitHub pull requests for bugs and security. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/review-github-pull-requests-for-bugs-and-security.html"
  },
  {
   "id": "reverse-engineer-binaries-in-ghidra",
   "title": "Reverse-engineer binaries in Ghidra",
   "summary": "Community report: A Ghidra plugin starts an HTTP server on localhost, and a Python bridge exposes it to MCP clients such as Claude or Cursor. The agent decompiles functions, follows call graphs and cross-references, and renames, retypes and comments in batches. The README now lists 253 tools, not 110.",
   "agent": "Claude",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50058850",
   "prompt": "Infra task — reverse-engineer binaries in Ghidra. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/reverse-engineer-binaries-in-ghidra.html"
  },
  {
   "id": "repair-formal-proofs-for-a-verified-garbage-collector",
   "title": "Repair formal proofs for a verified garbage collector",
   "summary": "Real-world thread: codex had repository access, edited Rocq proof files, ran targeted make builds, and Leaned on a rocq-mcp server to inspect proof states. It repaired the VST proofs first, then the graph-isomorphism theorem. The branch compiled and merged upstream as PR 30.",
   "agent": "Codex",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://arxiv.org/pdf/2606.21167",
   "prompt": "Infra task — repair formal proofs for a verified garbage collector. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/repair-formal-proofs-for-a-verified-garbage-collector.html"
  },
  {
   "id": "rename-a-core-type-across-a-monorepo",
   "title": "Rename a core type across a monorepo",
   "summary": "A Builder.io engineer handed Claude Code a TypeScript type rename across a monorepo and let it finish in one session while they context-switched.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://www.builder.io/blog/cursor-vs-claude-code",
   "prompt": "Rename type <OldName> to <NewName> across this monorepo. Update imports, tests and docs. Run the test suite after each major package and summarize remaining references.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/rename-a-core-type-across-a-monorepo.html"
  },
  {
   "id": "queue-maintenance-and-bug-fixes-for-cursor-background-agents",
   "title": "Queue maintenance and bug fixes for Cursor background agents",
   "summary": "From the linked report, zachary Proser authorized Cursor's GitHub app for his organizations. Cursor then cloned the repos into sandboxed containers, where the agents ran commands and stood up branches. By the time his coffee was done, most tasks were finished or ready for review.",
   "agent": "Cursor",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://zackproser.com/blog/cursor-agents-review",
   "prompt": "In this repo, queue maintenance and bug fixes for Cursor background agents. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/queue-maintenance-and-bug-fixes-for-cursor-background-agents.html"
  },
  {
   "id": "produce-a-track-in-ableton-live-by-talking",
   "title": "Produce a track in Ableton Live by talking",
   "summary": "One practitioner notes that the MCP server lets Codex run Python inside Ableton Live's object model, plus assembled-in tools for common tasks. In the demo, The builder sent chat prompts asking for an EDM song with macOS \"say\" vocals, chiptune sounds and 80s drum machines. Complete live sets were assembled from scratch in a few minutes, and follow-up prompts refined them.",
   "agent": "Codex",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47999656",
   "prompt": "Infra task — produce a track in Ableton Live by talking. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/produce-a-track-in-ableton-live-by-talking.html"
  },
  {
   "id": "point-a-computer-using-bot-at-any-repo",
   "title": "Point a computer-using bot at any repo",
   "summary": "Linked example of point a computer-using bot at any repo. In the thread, commenters describe Grok Bot as a general computer-powered by agent rather than a repo-tied cloud coder.",
   "agent": "Grok Bot",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/cursor/comments/1vsf5de/cursor_cloud_vs_grok_bots",
   "prompt": "In this repo, point a computer-using bot at any repo. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/point-a-computer-using-bot-at-any-repo.html"
  },
  {
   "id": "play-mayor-in-simcity-through-an-api",
   "title": "Play mayor in SimCity through an API",
   "summary": "From the linked report, you connect your own agent to a headless Micropolis city simulator, either through an MCP server or a REST API. You give it a short brief, for example a small city with reliable power and room to grow, and the agent plays mayor. The site reports 406 mayors, 1,902 cities assembled and about 28 million total population.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=46946593",
   "prompt": "In this repo, play mayor in SimCity through an API. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/play-mayor-in-simcity-through-an-api.html"
  },
  {
   "id": "migrate-a-saas-from-aws-to-digitalocean",
   "title": "Migrate a SaaS from AWS to DigitalOcean",
   "summary": "From the linked report, the developer asked Claude Code to build GitHub Actions workflows that bootstrap DigitalOcean: droplet, managed Postgres, Spaces, Docker setup. Claude read failed run logs with the gh CLI, located the cause, and patched the workflows. The builder says the migration took four hours with zero downtime, moving 29 tables and 3000+ rows.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://dev.to/anand_rathnas_d5b608cc3de/how-claude-code-helped-me-migrate-from-aws-to-digitalocean-in-4-hours-with-live-traffic-hmg",
   "prompt": "Infra task — migrate a SaaS from AWS to DigitalOcean. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/migrate-a-saas-from-aws-to-digitalocean.html"
  },
  {
   "id": "manage-nixos-server-configs",
   "title": "Manage NixOS server configs",
   "summary": "Someone keeps their self-hosted NixOS machines in a config repo with Claude Code, but blocks nixos-rebuild switch and applies each reviewed diff by hand.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=50017667",
   "prompt": "In this NixOS config repo, propose the smallest change for my request, show the diff, and never run nixos-rebuild switch. I will apply approved changes myself.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/manage-nixos-server-configs.html"
  },
  {
   "id": "let-an-agent-extend-its-own-code-inside-a-docker-sandbox",
   "title": "Let an agent extend its own code inside a Docker sandbox",
   "summary": "From the linked report, miguel, assembled on Claude and the Agno framework, runs inside Docker with the project mounted read-only and only its own code directory writable. A host-side runner snapshots git, gives the agent its source plus the next capability, then validates changes with AST, JSON schema and import checks. It finished all 10 seed capabilities, wrote its own checklist, and reached 21+ capabilities.",
   "agent": "Claude",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47329087",
   "prompt": "Infra task — let an agent extend its own code inside a Docker sandbox. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/let-an-agent-extend-its-own-code-inside-a-docker-sandbox.html"
  },
  {
   "id": "let-a-voice-receptionist-schedule-appointments-via-mcp",
   "title": "Let a voice receptionist schedule appointments via MCP",
   "summary": "As described in the source, the commenter's team runs a custom MCP server exposing tools such as list_appointments, schedule_appointment and cancel_appointment against their CRM. An ElevenLabs voice agent, which covers speech-to-text, the LLM and text-to-speech, connects to it over a Twilio phone line. The source is a short explanatory comment and reports no results, numbers or shortcomings.",
   "agent": "Custom agent",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49560285",
   "prompt": "Infra task — let a voice receptionist schedule appointments via MCP. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/let-a-voice-receptionist-schedule-appointments-via-mcp.html"
  },
  {
   "id": "let-claude-read-apple-health-and-fitbit-data-via-mcp",
   "title": "Let Claude read Apple Health and Fitbit data via MCP",
   "summary": "Claude Code covered let Claude read Apple Health and Fitbit data via MCP. The poster used a hosted MCP connector instead of a self-hosted server.",
   "agent": "Claude Code",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeHomies/comments/1uiey1q/if_you_want_claude_to_actually_read_your_apple",
   "prompt": "Infra task — let Claude read Apple Health and Fitbit data via MCP. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/let-claude-read-apple-health-and-fitbit-data-via-mcp.html"
  },
  {
   "id": "hunt-github-repos-that-leak-secrets-for-leads",
   "title": "Hunt GitHub repos that leak secrets for leads",
   "summary": "A privacy-product founder used Claude Code to search public GitHub for repos that expose sensitive values, then scored them as leads for a data-masking tool.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Lenny's Newsletter",
   "url": "https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code",
   "prompt": "Search public GitHub for repos that appear to commit API keys or tokens and also show evidence of coding-agent use. Rank the top 20 by priority and include a LinkedIn URL when you can find one.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/hunt-github-repos-that-leak-secrets-for-leads.html"
  },
  {
   "id": "have-copilot-coding-agent-write-and-push-code",
   "title": "Have Copilot coding agent write and push code",
   "summary": "From the linked report, this is a GitHub community feedback thread on the Copilot coding agent, not a single walkthrough. The original poster assigned issues to the agent and watched it work. The original poster located the agent good at context and code, but its first PRs were weak without instructions, and it couldn't read attachments or org PR links.",
   "agent": "GitHub Copilot",
   "topic": "Software Development",
   "source": "GitHub",
   "url": "https://github.com/orgs/community/discussions/170528",
   "prompt": "In this repo, have Copilot coding agent write and push code. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/have-copilot-coding-agent-write-and-push-code.html"
  },
  {
   "id": "have-claude-collect-names-from-a-wikipedia-page-into-a-file",
   "title": "Have Claude collect names from a Wikipedia page into a file",
   "summary": "Community report: the poster Leaned on Claude Computer Use to control Firefox, go to Wikipedia, search for a topic, then identify all the names on the page and save them into a text file. From the comments, they ran Anthropic’s provided Docker image locally with their API key, and the task completed in about a minute with no manual editing described. It seemed to work for this simple extraction task.",
   "agent": "Claude computer use",
   "topic": "DevOps & Infra",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeAI/comments/1gat8i7",
   "prompt": "Infra task — have Claude collect names from a Wikipedia page into a file. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/have-claude-collect-names-from-a-wikipedia-page-into-a-file.html"
  },
  {
   "id": "give-coding-agents-compiler-accurate-find-usages-for-c",
   "title": "Give coding agents compiler-accurate Find Usages for C#",
   "summary": "Community report: you install a .NET global tool and run it on a .sln or .csproj. It loads the project through MSBuild and asks Roslyn to resolve every symbol. The README's example is an internal method whose single compiler-resolved caller was a test, which shows how test-only usage can be spotted.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49667188",
   "prompt": "In this repo, give coding agents compiler-accurate Find Usages for C#. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/give-coding-agents-compiler-accurate-find-usages-for-c.html"
  },
  {
   "id": "give-agents-a-multi-engine-math-kernel-via-mcp",
   "title": "Give agents a multi-engine math kernel via MCP",
   "summary": "Real-world thread: mathKernel runs as an MCP server (mathkernel-mcp, stdio, 162 math_ tools). The model parses and plans, and the kernel does the computing across SymPy, Z3, Lean, mpmath and numba/CUDA. The source gives no benchmark or outcome numbers.",
   "agent": "Custom agent",
   "topic": "DevOps & Infra",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49592366",
   "prompt": "Infra task — give agents a multi-engine math kernel via MCP. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/give-agents-a-multi-engine-math-kernel-via-mcp.html"
  },
  {
   "id": "fix-a-broken-delete-button-in-a-github-repo",
   "title": "Fix a broken Delete button in a GitHub repo",
   "summary": "Community report: after giving Jules access to the GitHub repo, The builder picked the repo and branch and asked it to fix the broken Delete button. The requirements were to match the Delete button in another module and to disable deletion after two hours. Jules produced the code changes and the branch appeared in GitHub, and it wrote test cases that nobody asked for.",
   "agent": "Gemini",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://calvin.my/posts/fixing-a-bug-using-ai-with-google-jules",
   "prompt": "In this repo, fix a broken Delete button in a GitHub repo. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/fix-a-broken-delete-button-in-a-github-repo.html"
  },
  {
   "id": "find-and-fix-flaky-tests-in-a-broken-ci-pipeline-overnight",
   "title": "Find and fix flaky tests in a broken CI pipeline overnight",
   "summary": "With CI logs in reach, Claude Code tracked a rolling 30-day pass rate, chased the noisiest flakes, built minimal repros, and tried fixes overnight. Over a few weeks the pipeline shrank by about 20 minutes and master pass rate climbed ~35% from roughly 40%, though some suggested patches were thrown out on review.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47970174",
   "prompt": "In this repo, measure flaky tests from CI history, prioritize the worst offenders, attempt minimal repros and fixes overnight, and open draft PRs. I will review before anything merges.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/find-and-fix-flaky-tests-in-a-broken-ci-pipeline-overnight.html"
  },
  {
   "id": "draft-pull-request-reviews-for-you-to-edit-and-send",
   "title": "Draft pull request reviews for you to edit and send",
   "summary": "According to the write-up, the poster assembled Cerber to watch GitHub for pull requests requesting their review, run Claude Code through Someone’s existing CLI login, and prepare draft reviews in a local web UI. The agent produces a summary, change walkthrough, and proposed inline comments; Someone can discuss findings with it, rewrite or delete comments, then manually send the review from their own account. It helped with a queue that had grown to 15 PRs and avoided unreviewed AI comments being posted straight to GitHub.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Reddit",
   "url": "https://www.reddit.com/r/ClaudeCode/comments/1wp0zkk/i_built_a_claude_code_review_inbox_for_the_prs",
   "prompt": "In this repo, draft pull request reviews for you to edit and send. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/draft-pull-request-reviews-for-you-to-edit-and-send.html"
  },
  {
   "id": "diagnose-kubernetes-alerts",
   "title": "Diagnose Kubernetes alerts",
   "summary": "A published walkthrough shows in a Mattermost thread, a diagnostic report is posted by an n8n agent that is triggered by Alertmanager and queries Kubernetes, Grafana, and other MCP servers.",
   "agent": "n8n",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://community.n8n.io/t/ai-alert-assistant-automated-incident-diagnostics-with-5-mcp-servers-alertmanager-ai-agent-mattermost/285369",
   "prompt": "Infra task — diagnose Kubernetes alerts. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/diagnose-kubernetes-alerts.html"
  },
  {
   "id": "create-accounts-buy-domains-and-deploy-without-a-human",
   "title": "Create accounts, buy domains, and deploy without a human",
   "summary": "According to the write-up, someone installs the Stripe CLI with the Stripe Projects plugin, logs in to Stripe, and runs the project init command. Then they prompt their agent to build something and deploy it to a new domain. The post describes the flow as taking an agent from nothing to a deployed production app in one shot, with no dashboard visits, copied API tokens or card entry.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=48031684",
   "prompt": "In this repo, create accounts, buy domains, and deploy without a human. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/create-accounts-buy-domains-and-deploy-without-a-human.html"
  },
  {
   "id": "configure-a-local-ai-lab-and-build-a-knowledge-graph-by-chat",
   "title": "Configure a local AI lab and build a knowledge graph by chat",
   "summary": "As described in the source, per this overview, Ivan Fioravanti showed Hermes Agent setting up a local AI lab (liteLLM, Postgres, Prometheus, Grafana) through conversation. Separately, Alvaro Cintas had Hermes install Gbrain, an open-source project from Garry Tan, add it as an MCP server, and query it for entity-aware recall as a personal knowledge graph. The source names both setups as working demos.",
   "agent": "Hermes Agent",
   "topic": "DevOps & Infra",
   "source": "Blog",
   "url": "https://kie.ai/blog/what-is-hermes-agent",
   "prompt": "Infra task — configure a local AI lab and build a knowledge graph by chat. Inspect current config, propose a diff, and wait for approval before applying or restarting anything.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/configure-a-local-ai-lab-and-build-a-knowledge-graph-by-chat.html"
  },
  {
   "id": "codex-mobile-pr-for-a-tiny-lint-fix",
   "title": "Codex mobile PR for a tiny lint fix",
   "summary": "A reviewer notes OpenAI Codex in the mobile app can turn a Slack-style message into a small pull request, handy for trivial fixes away from a desk.",
   "agent": "Codex",
   "topic": "Software Development",
   "source": "Blog",
   "url": "https://www.asad.pw/ai-agents-experiences-with-manus-operator-and-more/",
   "prompt": "From my phone: fix the failing lint on main for file <path>, open a PR, and keep the diff under 20 lines.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/codex-mobile-pr-for-a-tiny-lint-fix.html"
  },
  {
   "id": "catch-intent-conflicts-between-parallel-coding-agents",
   "title": "Catch intent conflicts between parallel coding agents",
   "summary": "As described in the source, foremerge is a local CLI and MCP server that Claude Code, Codex or Cursor agents are wired to. Each agent registers and publishes its intent with semantic scopes such as symbol:PaymentService before editing, then claims that scope. The second agent's published intent immediately raised a HIGH replace_vs_extend advisory, with no files changed.",
   "agent": "Claude Code",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=49789356",
   "prompt": "In this repo, catch intent conflicts between parallel coding agents. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/catch-intent-conflicts-between-parallel-coding-agents.html"
  },
  {
   "id": "build-sqlite",
   "title": "Build SQLite",
   "summary": "From the linked report, custom agent shows up in a write-up about build SQLite. The linked post is the primary source for steps and outcomes.",
   "agent": "Custom agent",
   "topic": "Software Development",
   "source": "Hacker News",
   "url": "https://news.ycombinator.com/item?id=47031268",
   "prompt": "In this repo, build SQLite. Read relevant code, apply a minimal patch, run available checks, and summarize the diff. Stop before merge or deploy.",
   "added": "2026-10-15",
   "interest": 0,
   "page": "https://agentsightings.com/case/build-sqlite.html"
  }
 ]
}