AI is rotting pretty much everything that was carefully curated over the last couple of decades. If it can be expressed in bits it will be degraded. This is one of the most annoying side effects of AI, the general decrease in quality. Quality of discourse, quality in literature, quality of information, quality of software. It's everywhere.
The problem I'm already having with LLMs (in miniature) is code/comments/docs where I don't know whether it's someone's intent or whether its un-reviewed LLM babble. That creates a continual stumbling block for figuring out whether the thing should be the way it is, or whether I'm looking at an unusually-coherent error.
Take the exact same messy pile of bytes but promise: "Everything in this made sense to a human at some point in the process", and that changes how one approaches the problem.
Whether AI written code is going to be better in the longer term remains to be seen. With what I've looked at so far I'm not super optimistic about this.
To me it has become clear that LLMs will, on average, improve the code quality by a lot.
Which model did you use? I hold Opus 5.5 in quite high regard, finding it very reliable in terms of cider quality.