Suspension of Disbelief
In theory, an app is a better surface area for an AI assistant to have compared to existing channels. The app experience can be fully customized and controlled.
Yet people still love iMessage assistants. If good enough, I have no doubt that today, people would prefer to call their agents over the phone (despite the 8kHz limitation). Why is this?
My theory is that it evokes the suspension of disbelief. You're able to convince yourself that what you're interacting with is just a human, using tools and interacting with computers the same way that humans do. I think people really like this, it tickles some part of the brain. I think that's a big part of why the Sesame research preview was successful, for instance. I would bet that the anthropomorphization of AI will continue indefinitely, because humans are simply wired this way.
I don't think this is the endgame though. I think the endgame is definitely mixed interaction. Imagine if Tony Stark only talked to Jarvis the way he talked to any other human. That would be limiting and slow. It's because Jarvis is a computer that he is able to communicate through augmented modes of interaction, through GUIs and holograms.