I don’t think “generic AI” — as in AI not designed for a single purpose that can be invoked by a button or something — will ever go beyond the chat interaction model. The modes may change: you may speak to it rather than type. But I can’t see anything come up with something other than a user-initiated interaction model, in which the user asks the model to do something, and the model responds. It’s just such a natural fit to how people interact in real life. And it’s clear the models are good enough to handle such an open-ended option scope. Models can still do things for people in the background, but only after they’ve been told to do it. A world where the model just does stuff unbidden is not one I’d like to live in.