Owning AI code in review
You used an AI agent to write a Go function that retries a flaky downstream HTTP call with exponential backoff and jitter. It works, you've shipped it before, and now a reviewer comments on your PR: "Walk me through why the jitter is computed this way and what happens if the context is already cancelled when we enter the loop." You wrote the prompt, not the code. How do you respond, and what does this exchange reveal about owning AI-written code in review?
backoff_delay_bounds(base_ms: int, factor: int, max_delay_ms: int, attempts: int, context_cancelled: bool) → list[int][100,2,2000,5,false]out[100,200,400,800][100,2,2000,5,true]out[][100,2,300,4,false]out[100,200,300]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 & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.