Users and AI adaptive paths - any early automation research going on?

Bouncing around some discussion on how we use the internet and webapps are already changing and the topic of adaptive paths came up.

Extrapolating quite a bit from the increased use of AI chat to navigate the internet to the potential idea of moving away from defined structured user flows with clicks, field entry and element interaction to a different model.

Simple conversations and dynamic real time adaptations to get a user from A to B.

Multiple wrong turns alongside multiple successful paths of varying degree to the destination.

Now doing exploratory test sessions around this would continue to follow the same usual testing practices, discovery, experiment, identify and explore risks.

What about UI automation though, are there any happy paths any more, can entry points to flows now come from anywhere, will data entry bypass the UI.

Different AI’s will approach A to B differently and even the same AI will often do things differently every single time and then you have the users where every conversation (journey from A to B) is unique.

Can your UI automation handle this non-determinism, how does it approach this?

Any early thoughts on this? I realise we are not quite there yet but might be an interesting thought discussion that perhaps could significantly impact UI automation.

*Footnote that is likely a distraction but did enter our offline discussion- the whole vibium train talks a little on adaptive paths in automation but that may be more for a solution to existing problems rather than the consideration of users embracing adaptive paths themselves.