Developer finds 46 bugs in AI-refactored code, warns against blind trust

A developer used an AI coding agent to refactor a 1,920-line CLI entrypoint into 410 lines, but found three reference errors manually. After adding ESLint's no-undef rule, they discovered 46 undefined references across 7 files, breaking 7 of 14 commands. The fix was not a smarter AI model but a static analysis gate that doesn't trust anyone.