Question
An incident is open: a Python ETL job is failing and the on-call AI assistant proposes a one-line fix — wrapping the failing parse in a broad `try/except: pass` so the pipeline stops crashing. It would stop the pages immediately, and everyone's tired. As the engineer holding the pager, decide whether to apply it under pressure, and where the line is between a legitimate fast mitigation and a dangerous AI band-aid.
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.