# AI Code Fixes Need Review Loops (Example) A junior developer used AI to fix a failing test. The test passed. Two days later another integration broke because the fix introduced a different pattern for error handling. Nobody had done anything malicious. The system had made local repair easier than shared coherence. ## What Owlery would notice The visible issue was a code defect. The deeper issue was judgement under pressure. Signals: - fixes arrived without a short explanation of the trade-off - reviewers checked whether the change worked, not whether it fit - local patches created new styles, exceptions and naming patterns - architecture became harder to explain after each small fix ## Better move The team added a lightweight review loop for AI-assisted fixes: - what problem did the prompt target - what alternatives were considered - what existing pattern does the fix follow - what new risk might this introduce - what would make us revert or refactor this The goal was not to ban AI. The goal was to keep human judgement visible. ## Weak answer Tell juniors to stop using AI for important code. ## Better Owlery answer Treat AI output as a proposal, not a decision. Make the reasoning visible, protect shared standards, and review for fit with the system, not only local correctness. ## Watch out This can become policing if seniors use it to shame juniors. The review loop should increase learning and coherence, not create fear. ## See Also - [[AI Judgement Lens]] - [[Decision Hygiene Map]] - [[AI Code Fixes Need Review Loops (Example)]] <!-- owlery-external-loop --> ## Use this example with Owlery GPT Use this prompt when the pattern feels relevant to your context. > Use this Owlery example as a lens for my situation: AI Code Fixes Need Review Loops (Example). Here is what is happening: [context]. What is similar, what is different, and what should we try first? [Open Owlery GPT](https://chatgpt.com/g/g-6a0a198491d48191be9a00a665e43c9a-owlery-observe-frame-decide-learn) ## Feedback Did this page help, miss something, or use the wrong lens? [Share feedback on this page](https://tally.so/r/xXvP6J?source=obsidian-publish&page=ai-code-fixes-need-review-loops-%28example%29)