So Nate's finally talking about something that the rest of us, at least here on LinkedIn have been saying for the last couple months. People want proactive AI that adapts to you and learns what's important... to you.
The way he lays it out makes sense.
A sort of personal operating system that lives on a sort of heartbeat, and gets triggered by either events like a message coming in, or a scheduler.
Sure. Granted.
A really good personal assistant would be great!
But my complaint with this particular conversation as it's unfolded on LinkedIn and X, is that assistants are boring. Easily understandable, probably easy to fund. But boring.
Especially when you consider that you could probably work up an agent in Openclaw or Nanoclaw or Hermes to do exactly this set of things in a passable but less than perfect way.
Every piece of software you build should have a thesis.
An answer to "What problem does this solve?" If the answer to that question is, "Well, everything..." you're not thinking clearly about the problem.
I don't think consumer facing or dev facing agents have seen their final form yet. But increasingly, the thesis doesn't live in the framework anymore.
We really should be looking at the world that exists, and building more interesting tool sets for our agents.