Can you go a little bit deeper into some of those challenges that you were talking about with the xproc and developing those learning pains?

Right.

So agentic applications had a very different policy, and moreover, the biggest difference is that there was a lot more policies.

And very often policies are done per agent, right? Whereas before we were defining policies per, like, an entire application and, and while the policy can be complicated, i- it's actually, you know, the same policy applies to a lot, of a lot of different things traditionally.

But in agents in, in agentic application, uh, you very often have a... each agent have their own policy.

And so that presented a significant challenge for us at cloud scale because our proxies then would need to have millions and millions of different, you know, bits of information loaded into them just so they can on the fly pick the right policy for the right agent.

And even with our resources, it was just not feasible.

Can you differentiate between the two, and then talk to me more about this agentic proxy problems that you're solving?