Question
A PM asks you to explain why a recent latency regression happened, and you use an AI to draft a plain-English explanation of the root cause (an N+1 query introduced by a lazy-loaded ORM relation). The AI produces a tidy analogy: 'It's like the database had to make a separate trip to the store for every item instead of one big shopping trip.' How do you verify an AI's simplification of a technical issue isn't misleading before you send it to a non-engineer stakeholder?
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.