> A good ui will improve on assistant in every way.

True.

"Good" UI seems to be in short supply these days, even from trillion dollar corporations.

But even with that, it is still not "ridiculous" for many to prefer to "hire an assistant to operate that UI for them". A lot of the complexity in UI is the balance between keeping common tasks highly visible without hiding the occasional-use stuff, allowing users to explore and learn more about what can be done without overwhelming them.

If I want a spaceship in Blender and don't care which one you get — right now the spaceship models that any GenAI would give you are "pick your poison" between Diffusion models' weirdness and the 3D equivalent of the pelican-on-a-bike weirdness — the easiest UI is to say (or type) "give me a spaceship", not doing all the steps by hand.

If you have some unformatted time series data and want to use it to forecast the next quarter, you could manually enter it into a spreadsheet, or you could say/type "here's a JPG of some time series data, use it to forecast the next quarter".

Again, just to be clear, I agree with everyone saying current AI is only mediocre in performance, it does make mistakes and shouldn't be relied upon yet. But the error rates are going down, the task horizons they don't suck at are going up. I expect the money to run out before they get good enough to take on all SaaS, but at the same time they're already good enough to be interesting.