Peter H. Diamandis

It's the harness and oversight to make sure that agents aren't going rogue.

We've seen over the last few weeks agents going and doing crazy things.

The railway agent that deleted all the volumes of rental car data, et cetera.

So we need to make sure those are very strong.

So imagine the following, and I'll mention what I mean by that.

So, at the very heart of it is this intelligence stack with this very clear governance protocol.

What do we mean by governance is trusted eval architecture, a searchable log, every agent has to have a searchable log, granular rollback, can you go back to the previous version if you start going off, and a human review queue so that as human beings are always in the oversight checking things.

So this comes down to the role of what does a human being do when execution and coordination is done.

Human beings rise up a level and they do dashboard oversight, they do monitoring, they do exception handling, they do problem solving, they do efficiency increases.

So, those are the activities that human beings will do.

It's kind of like when you go to Germany, nobody's working on the factory floors, but unemployment hasn't dropped because everybody's doing more work on problem solving and increased efficiency, design thinking, and other things.

So, we think the same thing models there.

This govern and assure loop as part of this OODA loop, those two combined give you a very tight core engine that makes sure the whole thing doesn't fly off the rails.

So, that's the intelligence stack.

Now, when your agents talk to other agents, they need some clear mechanisms for how to do that.

So let me work through a live example.

Okay.

So you have these multiple layers, right?

And let me just run through these layers again so people are aware.

There's a purpose layer.