None of them feel like tools, at best you have to force them to be that, but are still really bad at it, see vibe coding and “AI-assisted” or higher effort AI images.
It’s also like they skipped a few steps too. I think an actual synonym search could be way more useful than a sycophantic chatbot built into my text editor (wonder if something like that already exist). At least in case of animation I heard some frame interpolation tool actually made for artists got trashed due to their makers wanted to hop onto the still in-infancy full-on generative AI bandwagon.
However I have to give it to them when it comes to amazing outsiders. They often already thought digital art or composing on the PC was something like what the AI was, only they didn’t know what button to press.
I suspect there are a lot of niche scientific fields that benefit and will continue to benefit from the enormous LLMs being built. Problem is, those will be used by a small number of scientists and technologists, applied to very specific problems. And for most scientific problems the existing open-source LLMs are good enough. But the AI companies want mass-market subscribable services so they can leech money off us forever.