cmdln, In his latest, I think @pluralistic put his finger right on the problem with adopting LLM's to write code.
https://pluralistic.net/2024/04/01/human-in-the-loop/
Even if LLM's were far better at generating code, he's right that this is totally a reverse centaur situation. The nature of the technology means all generated code has to be reviewed. Code review is a net drag, even in the best of circumstances, meaning LLM's in the long run won't help developers but rather place even more demands on our overloaded faculties.