Question
You're shipping v3 of an SQL-backed data-access library with several schema and API changes. You have an AI draft the migration guide for users upgrading from v2, covering renamed methods, a changed connection-pool config, and a new required index. The draft is thorough-looking. How do you confirm a migration guide is actually complete and that following it end-to-end leaves a user on a working v3 — without trusting the AI's coverage?
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.