You're restating their point. Normies' output will still be bad compared to the output that good engineers would produce, with or without AI. You wouldn't want to use poorly built software regardless of origin.
Hell, half of my use of LLMs these days is bypassing around crap in software written by people[0].
The metric in question is not how it's built. It's whether it solves a problem. Particularly for single or sub-100-users software, what matters is whether it does the job the user(s) like, the exact way they like it.
Speculating here, but maybe I'm also hitting at a disconnect between perspectives: I look at a product as tool to solve a problem. But some look at it as something to sell for money or Internet fame. For the former, distribution doesn't matter. For the latter, distribution is everything, and then, perceived code quality/effort signals actually start making sense.
TL;DR: I don't mind using shoddy software, but I agree it's a bit harder to sell obviously shoddy software.
--
[0] - https://news.ycombinator.com/item?id=49875994 I started calling this "Bypass-Driven Development".