This has been my gut feeling about the new AI technologies we've seen recently. We'll see what happens
(NoSQL is still useful, but it didn't "replace SQL" as promised, and the others have completely failed in spite of promising much).
It's always hard to predict the future; often the first iteration of something new isn't very good or even downright bad, and it's not easy to predict if it will ever be any good. I remember trying Rust somewhere in the early 2010s "before it was cool" and I hit some bug in the installation (would eat up all my memory). Then I tried the new version that fixed it and I ran in to problems actually compiling programs due to another bug. This doesn't reflect the status of Rust today at all, but back then it was essentially impossible for me to tell if it would always remain the buggy mess it was then, or if it would be able to put itself together and become stable. There's been plenty of buggy things that never became stable (Itanium, Hurd).
When in reality there's,
- People who aren't convinced yet
- People who don't want to be convinced (because of comfort, hubris, or livelihood)
- Human processes and skills that have to be adapted to a new way of doing things
- Material cost of making those transitions, which has to be justified to those making decisions
- Ecosystem support
Etc etc