Andy Halliday

So it's scrolling around and it's moving around and and then it's taking snapshots of the actual preview UI and say it knows.

I mean, it's very good.

You know, if it's using pixel examination, it's doing it in brilliantly because it can see details of the UI that even I didn't notice.

Yeah.

Correct those.

So, yeah, but that's the way it works right at the moment.

Now you can imagine if WebMCP happens.

Yeah, if you're looking at the UI of a preview that you're working on as a code or an application developer, then you'll want to have a visual interface for your agent that way.

But if you're asking an agent to go out and search for and then fill up a cart and shop for you, right now that is very compute intensive and it's going to get much, much easier with this WebMCP example.

I'm with you.

Yeah, so we're touching on a subject that is at the core of a very interesting paper that I came across today.

And it's by a couple of professors, I believe, at the University of Pennsylvania Wharton School.

So this is the business side of the school.

And it's about...

thinking fast and slow.

There are these two systems of thinking that happen in the human mind.

One of them is fast, quick, like Malcolm Gladwell's blink.

You have a very immediate and intuitive kind of capacity to arrive at a conclusion.

And then there's thinking slow, which is deliberative reasoning.

And we're training AI at this point to be very good at this deliberative reasoning.