When I Reject AI Code Even If It Works

In a personal blog, Vini Brasil discusses criteria for rejecting AI-generated code even when it passes tests. He cites issues like overly complicated logic, lack of comments, unnecessary abstractions, and inconsistent style. AI often produces 'clever' code that is hard for humans to read and modify. Brasil emphasizes that code is read far more often than written, and that AI lacks a sense of simplicity. He also notes that AI can introduce subtle bugs that tests miss, especially in concurrent or stateful systems. His rule of thumb: if the code is not immediately understandable to a colleague, reject it and ask for a simpler version.