> This happens with any sufficiently big/old codebase. We can never remember everything, even if we wrote it ourselves The important thing about a codebase was

> This happens with any sufficiently big/old codebase. We can never remember everything, even if we wrote it ourselves The important thing about a codebase wasn't ever really size or age, but whether it was a planned architecture or grown organically. The same is true post-LLM. Want to put AI in charge of tool-smithing inconsequential little widgets that are blocking you? Fine. Want to put AI in charge of deciding your overall approach and structure? Maybe fine. Worst of all is to put the AI in charge of the former, only to find later that you handed over architectural decisions at some point and without really intending to.