So like how should like a company decide what an agent is allowed to do?

Yeah.

The same way that they decide what any software system can do.

I think the big insight that we used to start our company was that agents are not people.

I know even now, people still refer to agents as a peer to a human.

And you'll hear a lot of vendors talk about non-human identity as a peer to a human and have teammates that are AI.

And I think that's a fallacy that can get a lot of teams in a lot of trouble.

So what security work does running one inside a company create?

I mean, you could, there's one coming out every month at this point.

Yeah, it's fascinating to me.

You know, I'll say this, like all those papers saying that there's a lack of enterprise adoption.

I'm not seeing it.

Right.

I wonder, I wonder how time delayed they are.

What I see is exponential.

the company,

I don't want to have to approve every $2 charge that goes into that credit card.

yeah so this is an important point of security so the obviously the answer is you do not want to give an agent your credentials right you would never give another human your credentials why the hell are you giving agent your credentials right and so but the what you do want to do is the way you off the way you interact with other people you want the agent to act as your delegate right? It's not you.

So this is the big problem with all the desktop agents.

So all the agents have access to your file system and your command line interface.

They're wildly unsafe today because they're acting directly as you, not your delegate, they're you.

And so if you can go in on Amazon CLI and drop a database, soak in the agent.

And that's why if you go on Twitter, you see all these horror stories of people losing files and losing emails and losing databases because the agent had direct access to everything the way the human did.

the speed of ai today so i need it to be acting very very quickly and putting that human in the loop takes and opens up additional vulnerability or time for vulnerability that i can't deal with so somewhere in there is a happy medium and the happy medium is actually scaring the crap out of me at this point

Yeah,

absolutely.

And then when you get on those phone calls, it's actually, it is a skill to know how to have those phone calls.

A rookie mistake I see is you show people your product and you ask them if they like it.

And then people give you feedback.

And you get a lot of false positive feedback.

And you also get false negative feedback.

So there's really like, I'll go back, I'll take this to the rubric