The evidence isn't in that it failed but why it failed. It's not that the problem was more difficult than they anticipated, it's that they didn't comprehend the nature of the problem they were solving from the outset. Software development is an incremental creative activity that involves good taste, constantly shifting and vague requirements, and keen understanding of human factors and ergonomics. LLMs fail at all three. And in fact the shifting and ambiguous requirement element is fatal to the notion of automating the process.
Outsourcing and oop and uml and LLMs and vibecoding --- they're all the same futile attempt at the same thing: abstracting the human out of a loop built for the benefit of humans. It's nonsensical, and the only people who want to do it are capitalists, so it's doomed to fail yet again.