Yeah I also keep thinking this. I don't see LLMs reliably producing code that is up to my standards. Granted I have high standards because I do take pride in producing high quality code (in all manner of metrics). A lot of the time the code works, unfortunately only for the most naive, mechanical definition of "works".