Completely agree. Saying this as someone who loves handcrafting code and has been doing it professionally for decades, it looks like a dead end at this point. LLM generated code might not be pretty, but these damn models are already so good at handling complexity, that they can actually work fine with a byzantine code base (which honestly would also describe most human-derived codebases I have worked with). I am still reeling from this realization and trying to figure out what it means.