I'd be a
little less reflexively hostile if the prompting used to generate it was prominently included, so that it was at least possible to see which properties of the output were deliberately asked for.
When I read a piece of code I don't understand, ideally I'm able to assume that the author intended what was written, and learn something when I figure out why. With AI generated code, I can't do that: the original author is more or less destroyed once the context is gone; at best I can ask someone who was kinda-sorta around when the author wrote it (i.e. the person who prompted it). At worst I'm really just asking someone (or something) else to guess what was intended.
Even granting that the generated code is implementing a coherent design (and that's a big if!), what we're left with is a very quick way to generate legacy code who's author is no longer around to answer questions. Such legacy code, if it's been proven to work, I might just stick a “don't touch this” sign on it, but if it starts causing any problems, I'd end up doing a ground-up rewrite if only to (potentially) understand what problems and concerns influenced the design to make it what it is.
AI can certainly be useful for this later task, but it'd be far preferable to just start there, as a form of pair-programming. And at that point, I doubt it'd be so interesting to say “built with ‹whatever model is popular today›” in the title.