I'm addressing your particular point about not seeing any use for this in the OS. I disagree. If there is a flexible layer that can consume and correlate context, there are tons of use cases
Then indeed there might be. I'm not saying AI is useless in general, even though it might be totally irrelevant to my personal workflow. But my question is still what Copilot brings. Does it do what you describe? If not, what does it do?
Yeah, I doubt it does anything close to what I'm talking about. If it did, it would be a massive leap forward in integration. More likely, the button opens up Copilot side-bar and only takes text input. But we shall see. Maybe some RAG going on, maybe a bit capable of multimodal, but I'd be a little surprised if there's more than the usual "prompt LLM here".