Question
You're using a coding agent to add idempotency to a payment-creation endpoint in a Java service. Twice it has produced code that 'works' in the happy path but reuses the idempotency key incorrectly under concurrent duplicate requests — the second request still creates a charge. Each fix tweaks the surface but misses the race. How do you diagnose why it keeps getting this wrong, and how do you re-steer or take over?
Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it — tests, edge cases, reading critically — and how you’d re-prompt or decompose to get it right.
Vibe coding: describe the solution in plain language (or narrate it) and the coach grades your approach. Generating runnable code from your description is coming next.