AI agents are 'remarkable': OpenAI chairman talks compute & AI spending

00:00 Speaker A

I want to talk about the role of personal agents because it has just taken off like crazy this year. If you used an agent a year ago, it looks completely different, right? But it also requires a ton more of compute. So, what the market's worried about is just the spending increasing so much and maybe those outcomes not keeping up. How do you guys see that?

00:26 Speaker B

Well, I'll start with the big picture, which is, is there value? You know, because sometimes if you're spending and you're not getting value, that's one thing. If you're spending and maybe it's inefficient, you know, you know, maybe that can change over time. I think these personal agents are pretty remarkable. You know, Muse and Instinct. I don't know if you I I think you've tried some.

00:46 Speaker A

Tried them all.

00:47 Speaker B

They're pretty remarkable. And and I think a lot of people have had aha moments with them, and I love my dot at my Open AI dot. They're amazing. And I think, you know, over time, I think the interesting thing is the reason why they're so compute intensive is if they're essentially using a web browser on your behalf. Um and one of the things we've tried to do is say, is that really the way these things are going to work in the future? We've worked on a a new protocol called Personal Agent protocol that not only I think will help the relationship between companies and these personal agents, but also make it a lot more efficient. Uh you know, it doesn't it turns out you don't need to spin up a web browser to make a simple API request. And we've put a little lot of thought into if this is a use case that's useful for consumers, what's the right way to do it? And we put a lot of thought into that over the past couple of months.