Exactly. Quality of code is a programming invention to make it easier to write and maintain correctly functioning applications.

That is the entire purpose of "quality of code".

If the end user experiences a correctly performing application, now, and in the future, they don't care at all what the code looks like.

AIs could resort to a single global array of primitives and forget all about functions, and just use gotos if it helped them (it probably doesn't).

That is only true for one-shot applications, though, whether written by human or machine. The reason we care about code quality is because rarely we don't have to look at code again after we first wrote it. Poor code quality makes maintenance and extension more difficult and expensive -- again, regardless of the degree of LLM support.

At least for human-written code, there's usually a thought and concept to be discovered underneath. For LLMs, one-shotting is all they know, and getting them to consider months or years of expanding and changing requirements will quickly turn into an impossible game of Twister.