Is that kind of what you're seeing?

Yeah.

I mean, so there's, you know, there's three categories of things frontier models.

small language or very tuned specific models, and then local open weights models that are more generic in nature.

All these three things exist, right? What we'll see is, the example you gave is, some kind of routing or judgment that says, what task are you performing? If this task is complicated reasoning, sophisticated context of the world at large, and it needs connectivity to various kinds of sources, et cetera, you might push it off to the frontier model, because they are the latest and greatest, the most context, et cetera.

If it is something extremely specialized, For example, we talked about our agent monitoring solution.

And we use a very specific small language model that is 97% cheaper, 95% faster than any frontier model.

When both sides run at machine speeds, what is the human job description?

than that.

It's not about data points.

It's the fact that you are trying to match non-deterministic actions with non-deterministic reactions.

To give an example, we've got our, you know, we talked about agent observability, which is basically, you know, this notion of having these evaluations that evals that run on agents, measure them at all time.

The problem is people start doing this and they use what's called an LLM as a judge, right? Large language model, this thing, because it's non-deterministic.

So you can judge non-deterministic behavior.

You got both these guys trying to say, oh, you're going to go off, I'm going to go off, right? The problem with that is it's A, it's slow and expensive.

Yeah.

So what's that tipping point where enterprises actually trust AI agents to take action on production systems?

Yeah, it's a great question.

I mean, all of us have gone, you know, in the last few years become dependent on these chatbots.

We ask questions all the time, everything replies.

And the productivity is gated by the speed of our typing, essentially, or speed of our talking, depends how we use it.

But these agents, these AIs can go at much, much machine speed, essentially.

And companies are fast moving to a point where they delegate work and let these agents to go do things.

I guess the question would be, why is eliminating the need to move and duplicate petabytes of data such a game changer for enterprise data economics? Yeah.

And

there's a couple of implications.

One is it's more attack surface.

The other one is, even if you don't think about malicious actors, it just creates more defects, more problems in your systems.

And then you've got the external threats of bad actors using AI to come in.

And these things are all happening at a speed that is AI speed, machine speed.

So you need to respond at machine speed.

What are the things that you've launched? What do customers buy? And more specifically, what kind of outcomes should they be expecting when they actually get those things? What does success look like?

And

there's a couple of implications.

One is it's more attack surface.

The other one is, even if you don't think about malicious actors, it just creates more defects, more problems in your systems.

And then you've got the external threats of bad actors using AI to come in.

And these things are all happening at a speed that is AI speed, machine speed.

So you need to respond at machine speed.

What are the things that you've launched? What do customers buy? And more specifically, what kind of outcomes should they be expecting when they actually get those things? What does success look like?

Tell us a little bit about that.

The Cisco Data Fabric is a turnkey solution that's really important.

And the reason it's important is that when you deal with data at this massive scale, you have to worry about not just the fact that you could handle this data, but how do you process it? Some of the data has to be processed super fast.

in indexes that you can search very, very rapidly, interactively.

Other data, on the other hand, is much more towards governance.

It can be kept in a little bit less immediate storages and processed differently.

to worry about what is signal, what is noise, what needs to be examined immediately, what can be done afterwards.

Tell us a little bit more about what does it mean for the enterprises out there and what should they be, you know, ready when they're kind of using agent builder?