Question
You're new to a 200k-line Python monolith and ask an AI agent to map the request lifecycle so you can safely move authentication into middleware. The agent produces a clean call graph and concludes "auth happens only in the decorator @require_login on each view, so centralizing it in middleware is a safe, mechanical move." What do you do before you trust that map and start the refactor?
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.