Hello, everyone, and welcome again to Cisco's Inside Product Podcast. Today I have with me DJ Sampath. DJ is the former CEO of Armorblox, and actually joined us. How long has it been, DJ?
Has it been three years since you've been here?
It's been three years to the day.
That's great. And so DJ is now running all of Cisco's AI and platform efforts, is someone that I spend a lot of time with. And welcome to the Pod.
Thank you so much for having me. I'm super excited to be here.
We're going to talk a lot about a lot of exciting things, including what's happening in AI, what's happening with Cisco Cloud Control. And so let's get started.
Let's start with what is going on in the market with AI right now that customers should keep in mind? And how is that changing infrastructure, specifically, in the world?
Jeetu, one of the biggest things that we need to start to internalize is the fact that we're moving from chatbots to agents. And as we move to agents, agents are starting to take actions on behalf of a user. And it's incredibly important to make sure that there's trusted delegation that's happening where these agents can go out and execute tasks entirely on your behalf. And so I think the shift that we're starting to see that's happening is with agentic coding. We're noticing that these agents are able to do things that we never thought was possible about six months ago, and there's a very common pattern. It starts out with these agents. When you start models, you feel like they're not perfect at the task that they're doing. And then within six months, you suddenly start to realize that those things that you thought was not perfect, it now does those jobs incredibly well. And then it starts to take on things that you never thought was possible. Right now, one of the most interesting things that's happening is in the frontier of science. We're starting to see these models are able to solve mathematical problems that haven't been solved for decades. And so what we're having to do right now as a team, is we start to think about these types of problems where we're having to peer a little bit into the future, and build for model capabilities getting there.
Just assume that some of the challenges that are deficits right now won't be deficits for a long period of time, and assume that those are going to be solved.
And then, because your planning cycle is going to take longer than how quickly it's going to have these models that are recursive self-learning loops are going to go out and solve the problem.
And loop engineering has become the thing that everybody talks about right now. If you go back almost three months ago, it was all about the harness. And it was really important to nail the harness, and the experience of how these models would work within the context of the harness-- the access to the tools that they have, the access to this MCP servers, and endpoints, and so on and so forth. But very quickly, that evolved to creating the right types of loops that you can take your agentic coding to start leveraging those to drive towards an outcome. And so what's happening now, if you take a look at what software engineers are doing today, they're basically refining these loops. And they're making sure that these loops are able to take a task all the way to completion from a ticket being filed that says, here's a specific issue that you need to go out and solve for. And then driving that from that point on to automatically writing code, testing it, pushing the updates, validating it, and then making sure that that ticket gets closed at the end of it all.
If you can capture that in a way that the agent can learn how to do that, then you've got a repeatable engine in place is basically the whole point.
The exhaust from every single one of that stuff, you've got to record that. And that's how you constantly keep improving that loop engineering in itself. And this is really at the core of what we decided to do with Cloud Control.
Right. So hang on before you go into Cloud Control. It seems like the speed and the exponential curve is almost becoming a vertical line. And as that happens, I want to make sure that people understand we wanted to embark on a project which was actually going to make sure that we were not operating like a holding company, but a completely integrated platform. And that integrated platform was loosely coupled with all of our assets, but tightly integrated. So you don't have to buy everything all at once from us. But boy, when you do buy two things together, they work like magic.
You actually lower the marginal cost of ingestion of every new technology that you deliver. Everything that you buy from us has compounding value on things that you might already have from us. And the value that the ecosystem benefits and garners from our platform is greater than the thing that we garner over time. So explain to people what Cisco Cloud Control is, and what we've built for the market that you launched at Cisco Live just a few months ago.
You bet. I want to break it down into three things as I talk about Cisco Cloud Control. I'll start first by talking about how we unified management plans across different domains. We brought together networking, security, data center infrastructure, observability, and collaboration all together--
--and compute into a single pane of glass.
One scaffolding that allows you to log in once. You're not logging into different dashboards to be able to do these things. They call it a swivel chair. You're swiveling from one computer, one screen, to another. There's no swivel chairs. You are basically logging into this portal. And once you're in, we've organized all of these services in three distinct categories. So the first one is the Platform Services that allow you to see things like inventory or topology of your entire estate in a single place. Now, this includes other things like identity, it includes licensing. We built all of these Platform Services that are in a single place that never existed before across all of these disparate places.
All these give customers intelligence about their environment, and how they should be thinking about and making sense of their environment over time. Especially for large enterprises, this is a huge problem.
This is a huge problem in a post-Mythos era. This is even more paramount to have a deep understanding of exactly what kind of assets do you have, what software is running inside of each one of those, and what vulnerabilities might exist in them for you to be able to go out and re-imagine it? So update your operational perspective with respect to the inventory that you have.
So DJ, how does this connect to our services side of the business? I know that we've created some tooling in the services side with Cisco IQ. How does all of this come together?
That's a great question, Jeetu. So we've built this in such a way that you can use Cloud Control to be able to seamlessly access Cisco IQ with the same login, with the same sets of experiences. In fact, the data connectors exist between these two tools in such a way that when you're looking at an inventory or a topology, you get information that are enriched from all of the services contracts that you have already in place. And so these two things are working hand-in-hand to be able to give the right types of experiences for our customers. Now, in addition to the Platform services, the second thing that we introduced was the dashboards that you know and love. All of the different products that we have inside of Cisco are all accessible and available from Cloud Control. You can just click into Meraki, and then you're, bam, inside of the Meraki dashboard. You click into Intersight, you're right there. Security Cloud Control, seamlessly you're hopping into each one of these dashboards. So for users that have learned how to operate these dashboards over many, many years, they don't have to unlearn something. They can seamlessly just click into those things and start using them. Now, the third thing that we did was we said, hey, what if we created applications within Cloud Control that can seamlessly start stitching different products together, but ultimately drive--
And Platform Services, basically.
--and Platform Services to offer an outcome that the customers are looking for. And so that is really what we did with that third category. Because a lot of times, what we're seeing increasingly, especially with agents coming into your infrastructure, your personas are starting to blur. A networking persona needs to do security stuff, a security persona needs to look at observability data. And so as much as the Conway's law has remained in a lot of cases, we've built teams around the styles of communications.
You ship your org chart as Conway's law.
Exactly. You end up shipping that as Conway's law. And that starts reflecting in the organizations that are buying based off of that, as well.
But what we're now starting to see with agentic operations starting to become a reality, is these things are starting to blur. And so it now becomes even more important to have apps that stitch across these domains. So if you are thinking about, say, a situation where you're trying to connect a campus and a data center together.
Exactly. You now need a single fabric. You can now have an app that allows you to configure the networking inside of your campus, and configure the network inside of your data center in a way that you are not copy-pasting these across different dashboards. Because that's where errors creep in, that's where our configuration mismatch shows up.
So this just becomes a single click EVPN-VXLAN fabric creation.
Without any effort-- no coding, no scripting, just natural language. Give instructions, and before you know it, it understands your intent and it just creates a fabric.
Exactly. It copies all of the virtual routing fabrics on one side. It copies the routing table information. It starts to look at it, and says, hey, wait a minute. The configs don't look quite right. Here are the changes that you need to make. Here are the VLAN tags that you need to put in place. All of that complexity is now abstracted away so that you are focusing on connecting those two endpoints, environments, and then seamlessly running your workloads on top.
The other example I really like of your apps that you have is the agent security app. Because you need to have five or six different backend services sometimes. I might need identity. I might need Secure Access. I might need AI defense. I might need Tokenomics. I might need observability for AI-- all of these things. But what you've done is you've said, I'm just building an app for it.
And that app abstracts away all of that complexity and just creates one app that you can just say, OK, Mr. Customer, you can just go out and deploy this.
And what's even cooler about that? A customer may not always have all Cisco products.
That's right. But you don't have to do any kind of rip and replace on any of that stuff.
You don't. You now have the ability to integrate those apps seamlessly into those products, as well. But guess what? You provide a lot more value in stitching that, so customers start to use these apps more than they're using any one of those individual piece parts.
And that is a game changer, when you think about day two and day three operations for a lot of these folks.
So you said that you want to have a unified management plane. That's the first thing that you want to do-- a single pane of glass for all of Cisco's products across any domain.
The second big thing is, when you start to think about day two and day three operations--
So what's the second thing? What is the second thing that Cloud Control does? The is the first one, yeah?
Right. So the second big topic that I want to hit with Cloud Control is, the way you start to operate an infrastructure like this needs to be reimagined with agents that are now in place. So agentic operations is what we're calling it, and I think agentic ops requires us to reframe how we think about debugging and troubleshooting.
It's a successor to AIOps.
Correct. It's a better way to do this, because the agents have the ability to take actions. They have a certain degree of autonomy, and they're trained to be able to do this in ways that are completely auditable and are secure. So that's really at the core of agentic ops. But the way we reimagined this was we introduced a capability called Canvas. And what Canvas does is it provides you with a surface that brings together generative UIs, where the UI gets generated on the fly. It connects all of the multiple domains together, and it also unlocks multi-user and multiplayer capabilities. Because when you think about debugging and troubleshooting, it's never a single person that's cutting across the board. It's always a team that works together to be able to solve this.
But all this tooling is for single user and single persona.
That's right. So far, that's how we've trained our teams. It's sort of back to the Conway's law. I think we have to break that, and that's really what we're doing with Canvas. We're re-imagining how multiplayers, multiple folks can come in and start working in debugging and troubleshooting. It gives them a system of record. You don't have to constantly go back to your tickets and copy-paste those things. You're now starting to integrate each and every single one of that stuff-- using an MCTP server, using agents that can do both immediate responses to the questions that you're asking, all the way from that mode to doing deep reasoning where it might take maybe several hours for it to be able to go out and figure something out. But it has the long running capability to reflect on your configuration files, spin up a--
