Most “AI debug me” sessions devolve into the AI guessing at random causes. The structure below gets to root cause faster, with fewer detours.
Who this is for
Anyone debugging code daily who reaches for ChatGPT / Claude / Cursor when stuck.
When to reach for it
A specific failure (stack trace, wrong output, intermittent bug) you can reliably reproduce.
When this is NOT the right tool
Performance regressions that need profiling; flaky tests where the cause is environment; design problems disguised as bugs. For platform-wide issue hunts (re-renders, navigation, native modules), see auditing a React Native project with AI.
Step by step
- Reproduce the bug. AI cannot debug what you cannot reproduce.
- Capture: exact error / stack trace, the input that triggers it, what you expected vs got.
- Paste all three into one message: “Here is the error: [trace]. The trigger: [input]. Expected: X. Actual: Y.”
- Ask: “List 3 most likely causes, ranked by likelihood, with one specific check for each.” DON’T let it write fixes yet.
- Run the suggested checks. Report results back: “Check 1 passes, Check 2 reveals: …”
- Now ask for the minimal fix for the confirmed cause. Apply, re-run reproduction. Confirm.
- After fix: ask the AI to write a regression test for this specific bug.
Recommended workflow
A 500 error: trace + input + expected → 3 hypotheses → run check on hypothesis 2 (env var missing in test env) → minimal fix → regression test → done. 15 minutes vs an hour of random guessing. When the bug only shows up in production on Firebase Hosting, pair this with the AI Firebase deploy checks workflow — most “works locally” bugs trace back to firebase.json rewrites or function region mismatches.
Common mistakes
- Pasting only the error message without the trigger or expected behavior. AI guesses wildly.
- Letting AI write a fix before you confirm the cause. Often fixes the wrong thing.
- Trying 5 fixes in a row when none work. Stop. Re-read the trace. Often the bug is somewhere else.
- Forgetting to add a regression test. Same bug will return.
Advanced tips
- For intermittent bugs, ask: “What ordering or timing assumption could cause this to fail sometimes?”
- For “works on my machine” bugs, paste the environment difference explicitly (Node version, OS, env vars).
- When AI keeps suggesting wrong causes, give it negative evidence: “It is NOT a network issue, NOT a permissions issue.”
Copy-ready prompt
Error / trace:
{paste exact trace}
Trigger: {what input or step}
Expected: {what should happen}
Actual: {what happens}
List the 3 most likely root causes, ranked by likelihood. For each, give one specific diagnostic check I can run. Do not write a fix yet.FAQ
- Should I include code?: Yes — the function around the error, plus any data the bug depends on. AI cannot guess.
- What if AI hallucinates a cause?: It will sometimes. Always run the suggested check before accepting the cause. The diagnostic step catches hallucinations.
Related
Tags: #AI coding #Tutorial #Workflow