This all comes down to execs looking for a reason to be relevant. Just like every single "revolution" in software development, the executive set thinks that this is going to be the innovation that allows them to finally hire fewer programmers and deliver better software than ever before, and it's never been true, and won't be true this time.
AI has become an incredibly useful tool for competent engineers to build more, better software. Is it orders of magnitude more impactful than, say, the widespread introduction of OOP? yeah, probably. Is it still the same kinda thing? Yeah, absolutely. You can do some _incredible_ stuff as a solo developer with a good coding agent and good domain knowledge these days; as we speak I'm working on a complete astrophotography toolkit that can run off an android smartphone, and this is possible mostly because 1) I already knew enough android to reason about the ecosystem and 2) I know enough about astrophotography to reason about the work that needs doing. Domain knowledge is what separates AI slop from good software that happens to be written using AI, and you can't just magick that out of existence, no matter how much the pointy haired's want it gone.
I don't think we're particularly close to an AI system that can fully substitute for human taste, and if we stop pretending we can already have "fully autonomous AI employees" and instead see AI as a new paradigm for building software in the hands of professionals, we have a good chance of actually changing the industry. But stupid hype cycles are gonna stupid hype cycle, and it'll probably take a few more MSFT-level crash outs for us to get to a level of sanity around it.