And Claire is asking, "In a world of rapidly evolving agents, what does a trustworthy AI runtime look like when it needs to support both internal and external users across an ex- enterprise with external integrations as well?"

Uh, yeah, sure.

Um, so, uh, so there's a lot of components that go into a trustworthy agentic runtime or harness.

Um, again, the basis of my presentation was really at that lowest level authentication, authorization and security, right? So the, the, the runtime, um, harness needs to expose controls such that the, the, the, the most risky component of the system, which is the LLM, the, the most non-deterministic, isn't the one you rely most on for your safety net, right? Your system is, um, authorized and secure.

And so that safety net is, is probably the base case, lowest level, and the most tried and true, right? We've, we've, we've had those controls in place for ages, eons.

Um, the runtime harness and involvement, um, for running your agents also should have a very rich human-in-the-loop control system.

So very often, human-in-the-loop is, is maybe underplayed as, as a nice usability feature, but for system, um, or for, or for items and actions that are consequential that may not be able to undo, um, you wanna be able to check, uh, uh, with a human in, in order to inform, uh, the decision-making process.

Yep, sorry about that.