Mark Williams-Cook

coded in a way that's very predictable, it's not great for security because you don't know how the system's going to react.

And I'm part of the mail group that talks about the agent-to-agent protocol development.

And someone had mentioned about using AI and immediately someone came back and was like, this creates a huge surface for security risks.

Like, you know, we don't want to do this in this case.

And it's interesting seeing those conversations happen because we always belt forward, right?

With technology.

And you think, and lastly, I'd just say on the like LLM agent stuff,

It was quite late in the day that Windows started having those user access control pop-ups, where if you want to change something on your system that's important, it kind of gave you a pop-up and we're like, oh, the computer's about to do this.

Is that okay?

It's kind of annoying, but it was needed because viruses and Trojans and everything were rampant for a long while in Windows.

And I wonder if we're going to have to have something like this within LLMs and within agents, or otherwise they can just be sent off to go do stuff, buy stuff, and you're unaware.

They need to be, it's about to do this.

Is that okay?

For sure, right?

For sure.

Yeah, yeah.

Yeah.

I guess I haven't enjoyed seeing people talk about writing using semantic triplets for AI.

I don't even know what that is.

Yeah.