Question
You ask an AI to fix a Java service throwing intermittent `OutOfMemoryError` under load. It suggests bumping `-Xmx` from 2g to 4g; OOM comes back. You re-prompt, it adds `System.gc()` calls 'to free memory sooner'; no change. Third try it switches the GC to G1 and bumps heap to 8g; OOMs still recur, just less often. The model is treating every OOM as 'not enough heap.' How do you recognize it's stuck and reset?
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.