Coding agents have moved well beyond autocomplete. Tools that can read a whole repository, plan a change, edit many files and run the tests are now part of everyday development for a lot of people. The question is no longer whether they work, but which tasks they handle well.
This guide collects real, sourced examples from developers using coding agents on actual projects: framework migrations, bug hunts, refactors, test writing, prototypes and side projects built mostly by an agent. Each case names the tool used and links to the original write-up.
Reading through them, a few habits come up again and again. Developers who get good results ask for a plan before any edits, point the agent at official documentation, keep changes on a separate branch, and review the diff and test output before merging. The agent does the typing and searching; the developer stays responsible for the design.
Use the cases below as a menu of tasks worth delegating. Open one that matches your work, copy its reusable prompt and adapt it to your codebase. Starting with a well-tested, low-risk area is the quickest way to learn where an agent fits in your workflow. The list is ordered by interest.
What is in this guide: 60 sourced cases, added between 2026-10-15 and 2026-10-15. The tools that come up most often are Claude Code (17), Cursor (10), Codex (7), and the cases mostly sit under Software Development (59), Developer Tools (1). They were reported on Hacker News (20), Blog (20), X (10). If you only have a few minutes, start with the top three: “Fix app bugs, harden security and ship fixes autonomously”; “Plan meals and track symptoms”; “Edit site design visually and have an agent write the code”.
Each case page gives a short summary in our own words, a note on why it matters, a reusable prompt broken into steps you can follow, and a link to the original first-hand account. Nothing here is a paid placement, and we only list cases with a working public source.
60 real cases, most interesting first
Fix app bugs, harden security and ship fixes autonomously· Hermes Agent · Software Development 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…
Plan meals and track symptoms· Claude Code · Software Development 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…
Edit site design visually and have an agent write the code· Claude Code · Software Development 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.…
Prototype product features as a non-coding PM· Cursor · Software Development 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…
Plan in Claude Code, implement in Codex· Codex · Software Development Calvin keeps Claude Code open for planning, git and exploration, then flips to Codex when ready to implement, with separate-family review on PRs.
Build a deals page step by step· Devin · Software Development 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…
Build a cardiac health tracking app· Claude · Software Development Community report: after a November 2025 hospitalization for dehydration, the poster started with a spreadsheet and then used months of…
Run parallel local coding agent sessions· Claude Code · Software Development 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…
Give a coding agent a paper corpus to find new techniques· Claude Code · Software Development 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…
Local multi-session agent with parallel task graphs· Captain Claw · Developer Tools A local agent runtime splits a large task into parallel sub-sessions across different model providers and resumes them from SQLite.
Cross-family AI review before human PR review· Codex · Software Development After six months of agent-only coding, a commenter routes every PR through a different-model-family reviewer plus deterministic gates before looking…
Ship an iOS app in 17 days with agents· Cursor · Software Development 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…
Let an agent start a business from scratch· Devin · Software Development Toolbox for Devin: Ramp spend card, ElevenLabs voice number, AgentMail, Stripe, GitHub memory. Mandate: invent a business. Result after ~four weeks…
Test features on screen and merge fixed PRs· Devin · Software Development 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…
Hunt business-logic flaws and open fix PRs· Devin · Software Development As described in the source, parallel agents each take a segment of the codebase and reason across files for business-logic flaws, chained auth…
Write, review and document code across a product's PRs· Devin · Software Development One practitioner notes that according to founders, Devin is responsible for handling all ChatPRD PRs and is the top contributor to Gumroad's merged…
Write a Linux kernel improvement· Claude Code · Software Development A published walkthrough shows according to the headline, Linus Torvalds is working on a large patch for the Linux kernel powered by Claude Code.
Use Codex for non-coding chores· Codex · Software Development Documented elsewhere: people in the thread describe powered by Codex directly for ad hoc computer work rather than building full automations first.…
Test how well coding agents can onboard to your product· Custom agent · Software Development 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…
Team playbook for Claude Code and Cursor· Claude Code · Software Development An engineering team shares a playbook: research before coding, plan mode, phased commits, and 3–5 parallel Claude Code sessions in git worktrees.
Take ideas to implementation· Claude Code · Software Development A published walkthrough shows Claude Code covered take ideas to implementation The linked post has the narrative detail.
Several coding agents in parallel· Codex · Software Development A developer runs multiple coding agents at once on separate low-risk tasks and reviews their output as it lands.
Self-driving docs via Playwright exploration· Claude Code · Software Development A team lets Claude Code drive Playwright to explore their product, find documentation gaps, and draft the missing pages.
Run real estate transactions· Claude · Software Development 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…
Pair-program· Claude · Software Development Case study takeaway: claude shows up in a write-up about pair-program. The linked post is the primary source for steps and outcomes.
Group outdated packages by upgrade difficulty· Cursor · Software Development Before a big dependency upgrade, Kent asked Cursor to list every outdated package grouped into easy, medium and hard buckets.
Generate live-coding music· Claude Code · Software Development 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…
Fix and resubmit a rejected app remotely· Dots · Software Development Dots covered fix and resubmit a rejected app remotely. User asked their dot to investigate the rejected app submission.
Debug overnight· Codex · Software Development 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.
Control robots· Gemini · Software Development 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…
Build a mental models framework· Claude · Software Development 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…
Boring chores for a coding agent· Claude Code · Software Development A how-to collects mundane tasks, like file renames and conversions, that a coding agent handles well.
Automate web configuration from coding sessions· Claude · Software Development Linked example of automate web configuration from coding sessions. In the thread, one commenter says they use Claude in Chrome during coding to…
Voice memo to PRD to shipped feature· Claude Code · Software Development 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.
Turn one goal into an ordered plan of coding-agent tasks· Custom agent · Software Development 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…
Slack bug report to Linear ticket and fix· Cursor · Software Development When a bug lands in Slack, an automation checks for duplicates, opens a Linear issue, investigates the codebase, attempts a fix, and replies…
Run parallel coding agents from Markdown specs in tmux· Claude Code · Software Development According to the write-up, manuel Schipper runs vanilla Claude Code agents in separate tmux windows, labeled Planner, Worker or PM. Planners write…
Find a Linux vulnerability hidden for 23 years· Claude Code · Software Development 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…
Coordinator agent that hands off to coding agents· Hermes · Software Development A conversational agent runs todos, cron jobs, research and Obsidian notes, delegating repo work to IDE and CLI coding agents.
Coordinate sub-agents to monitor PRs and fix prod bugs· Cursor · Software Development Observed in the wild: the Cursor Projects coordinator agent received overall project context and intent. On its own it monitored fresh PRs and…
Build and publish a BaZi chart reading agent· Codex · Software Development A published walkthrough shows the developer used AI coding agents to build the full stack solo. The agent received a deterministic BaZi calendar…
Work on mainframes and COBOL· Custom agent · Software Development 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…
Use a bot as a QA intern on a live app· Grok Bot · Software Development 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…
XState migration with looked-up docs· Cursor · Software Development 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…
Surgical Cursor fixes after Claude Code draft· Cursor · Software Development Drew switches to Cursor with tight prompts when Claude Code stalls on odd codebase patterns, then returns to Claude Code for polish and Playwright…
Run a multi-bot team that opens draft PRs· Grok Bot · Software Development Think assembly line, not single chat: research bot, classifier, builder, tester, reviewer, security. Each stage narrows context, then a draft PR…
Plan-then-execute coding with stacked review passes· Claude Code · Software Development A developer plans features into markdown files, has the agent execute them, and runs several review skills on every diff before the PR.
Finish abandoned side projects with agents· Claude Code · Software Development HN readers describe finally finishing long-backlog personal tools—music players, trackers, niche utilities—because agents remove the boring glue work.
Find local pilot customers from the codebase· Claude Code · Software Development 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.
Delegate tiny chores to Devin via Slack· Devin · Software Development A developer fires Devin at small, well-scoped chores from Slack and reviews the resulting GitHub PRs at lunch and end of day.
Claude Code for git, plans and codebase tours· Claude Code · Software Development Calvin uses Claude Code daily to create plans, script local tasks, manage git/GitHub, and explain unfamiliar parts of the codebase via fast explore…
Agentic codeowners that assign reviewers· Cursor · Software Development An automation classifies PR risk, auto-approves low-risk changes, and assigns up to two reviewers for higher-risk work based on contribution history.
Spend leftover AI quota on reviewable PRs· Claude · Software Development One practitioner notes that the founder points RemainderBot at a goals file and task-list sources. In cron mode it checks remaining subscription…
PagerDuty incident to proposed fix PR· Cursor · Software Development 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.
Run varied business tasks through a general agent· Manus · Software Development 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…
Review Linux kernel patches· Custom agent · Software Development Community report: roman Gushchin's team at Google spent several months building Sashiko, an agentic code review system, and Leaned on it internally…