It's not about solving puzzles for us, craft-focused developers. It's about caring about the clarity of our writing, for the same reason a writer wouldn't want an AI agent writing their book.
A writer wouldn't want an AI agent writing their book because the end product is their writing. Readers actually care about the words and prose. That's what they are paying for after all. Users of a software program meanwhile want it to work as promised, that's it. The syntax, language, design patterns used, how elegant the code is etc. are all irrelevant. If an AI agent can write "better" code (in terms of meeting that promise) than a human programmer than that is objectively the right way forward.
This can apply to anything. Why care about how well a shovel is built if all you care about is a hole in the ground. Why care about the quality of the hole in the ground if all you need is laying foundation that can hold a house… Why care about a neat kitchen, if all you care about is a cooked dish. At every layer there are multiple sets of end users. Code is used by programmers and businesses as the source of truth and automation. Just like shovels are used by people who dig with them, not just customers who need holes.
If your point is nobody will ever need to read the code, there's a reason why truly self-driving cars aren't happening yet. We will need human intervention as a failsafe, probably for a while. And humans have been known to care deeply about way lesser things than reducing friction for handling a failure contingency.
The code being written really is the end product. Code is unreasonably effective because it is an efficient representation of an idea.
Dismissing all the other concerns it addresses except what the user thinks and how much money it makes is not only irresponsible and short-sighted, but an objectively inferior product. In one word: fraud.