Question
Two teams integrate over a new internal API. Your side's client and the other team's server stub were both largely generated by AI agents working from a shared OpenAPI spec, and each side's agent filled in ambiguous parts of the spec with its own plausible assumptions — your client retries on 409, their server treats 409 as terminal; your client assumes pagination is cursor-based, theirs implemented offset. Integration testing surfaces mismatches. How do you resolve this, and what does it reveal about keeping shared understanding when both sides let AI fill the gaps?
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.