If the default mode is that LLMs generate crap code that you have to fix with even more LLMs, then something is fundamentally wrong.
In terms of over engineering, I wouldn't be surprised if the human tendency for skeuomorphism (combined with an loss of technical skill) will create even weirder code.
The default mode of the whole startup world is humans generate crap code that you have to fix with even more humans. And it has been like that for at least two decades.