Yeah, so far coding agents haven't been good at reducing entropy at a large scale. That's the area where we need to push hardest in my opinion. Writing code is cheap now, but that makes it easier than ever to inflate codebases to millions of lines of unnecessary complexity that are impossible to work with and maintain. I want to see an LLM reduce a complex multi-million line codebase to 1/10th of its size or less without breaking functionality. That would be impressive!