Home › How-to guides › Automate bug fixing and code upgrades

How to automate bug fixing and code upgrades with an AI agent

I looked through 59 real cases about bug fixing and code upgrades and pulled out what people actually did, the prompts they used and the places it went sideways. No theory, no invented success stories.

What people actually do

• Set up nightly smart-home routines by voice · Alexa+ · Home & Family
  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…
• Get status match and upgrade certificates, then book flight · Dots · Travel & Booking
  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…
• Extract receipt data for freelancer bookkeeping · Claude Code · Finance
  As described in the source, the builder used Claude Code to implement most of a bookkeeping app for freelancers and small businesses across a…
• Handle hotel booking and directions on a business trip · Gemini · Travel & Booking
  On a messy international work trip, Gemini picked a policy-compliant last-minute Heathrow layover hotel and walked someone from gate through…
• Agents cut each service migration from 3 months to 2 weeks · Multi-agent migration (Google Cloud) · DevOps & Infra
  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.…
• Assemble an immigration document package · ChatGPT Agent · Finance
  Seven years of tax returns, bank PDFs, and immigration paperwork—pulled via ChatGPT browser sessions into a single packet in minutes. The post stops…
• Agents find the exceptions in SAP migration projects · KTern.AI agents (Amazon Bedrock AgentCore) · Business & Startups
  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%…
• Morning test-coverage agent on merged code · Cursor · Personal Assistant
  Every morning a Cursor automation reviews recently merged code, adds tests that match existing conventions, runs them, and opens a PR.

Browse all 1081 cases to see the rest.

Prompts to start from

These are the reusable prompts from real cases. Swap in your own details, run it once on something low-stakes, and read what the agent did before you trust it.

Home/family task: set up nightly smart-home routines by voice. Keep changes reversible and ask before messages, payments, or deletes.

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.

Tools that show up most

Claude Code (15), Cursor (9), Devin (7), Codex (6), Claude (3), Replit Agent (2).

Guardrails

Work on a fresh branch, ask for a plan before edits, and review the diff and test results before merging.

Where it went wrong, or nearly did

• Validation agent that stops after three tries: A guardrail by design: it patches failures up to three times, then leaves a blocker note for a human.

More mess: the AI agent fails page. Related: the prompt library.