We're reaching the top end of the chatbot UX modality.
We've reached the point where the model intelligence matters less than the scaffolding it's activated in.
For a long time, the intelligence of the model wasn't good enough, so every incremental gain unlocked new use cases.
We're now to the point where for a lot of use cases, it's hard for a normal user to even distinguish the quality improvements.
We've reached the "good enough" level you can take for granted, at least in the default chatbot modality.
The models have gotten so good that the limitations come from the chatbot modality itself.
Now the main question is what is the right UX paradigm to unlock its value and integrate into our lives, and how to get the right context into the right interactions.
That's a very different kind of problem to solve!