8. The Opportunity and Challenge of Agentic Security with Jeetu Patel, DJ Sampath, and Matt Caulfield

Understanding the Opportunity and Challenge of Agent Security Welcome to another episode of Inside Product Podcast. Today I have with me two very special guests, two former CE OS of companies that we have acquired Armor Blocks with DJ Sampath and ORT with Matt Caulfield. And so welcome to both of you. Hey. Speaker 2 Thanks for having. Speaker 1 Us, yeah, we're going to. We're going to just riff on what is going on with agent security. So I'm just going to set some context and I want to make sure that we get our customers very clear and the audience very clear about what is it that we can do that is unique in this domain of agent security, which is a very, very heavily talked about domain right now. So assuming that we are in this kind of next phase of AI with agents and agents are going to be like digital Co workers rather than just tools that are going to have tools access, we need to do a few things with agents, right? One is we have to protect the agents from the world with things that prompt injection attacks, data poisoning, all that stuff. The second area is we got to protect the world from agents because these agents can go rogue and if they go rogue, they can actually do a lot of harm. So what is it that we are doing to provide a level of security so that customers can feel a level of trust when they delegate jobs to agents and they don't feel a high degree of anxiety when that happens? Speaker 3 Yeah. No, I think you've got the framing right, which is we had to think about a lot of organizations are both building AI applications and they need to protect those applications and those agents. And then a whole lot of organizations are starting to employ these agent, you know, workers and giving them access to things and they need to protect their sensitive enterprise applications from the havoc that those agents can cause. And I think Cisco pretty unique in that our portfolio spreads, you know, across both of those and I think we can definitely dive into a bit more about both sides of that coin. Speaker 1 Yeah. And by the way, what's really interesting to note before DJ you jump into this is both of you came with relatively small acquisitions that your companies weren't very, very large at the scale at the time, but both of you have stepped up into very large roles at Cisco. DJ, you run all of the platform and AI initiatives at Cisco. Matt, you run all of the identity initiatives at Cisco. So these are like big jobs. So firstly, you know, I just want to let you know that you folks are doing a great job and you've actually assumed a massive level of responsibility and brought the startup ethos into Cisco, which is great. But talk to us about agent security, DJ from your perspective. Speaker 2 Listen, one of the things about agents, you mentioned that they're part of your workforce. At this point of time, the thing that you do when you hire employees is you do a background check. Speaker 1 Right. Speaker 2 You got to be able to do the same thing with an agent because guess what, when an agent comes into your workforce, you're delegating a lot of responsibilities to the agent. The agents going to start taking actions on your behalf. They're going to start sending emails, they're going to check your calendars, they're going to do work, you know, so we're, we're sort of moving from this paradigm of AI being just chat bots to, you know, them being agents where they're taking actions. And so, so it becomes even more important to do that background check. And the way you have to start to think about these background checks is you have to make sure that you're thinking about the model provenance, you're making sure that the model that the agent uses to be able to do different things where. Speaker 1 Does it come from? What's the origination of it? Speaker 2 Exactly right. You, your supply chain suddenly becomes so much more critical that it now it, it should be using the right models. And, And what if the models have vulnerabilities? You got to make sure that those models are red teamed. You're, you're, you're understanding exactly about the vulnerabilities are hiding inside of these models. You know, something that's really interesting is this for regular software, there's something called a CVEA common vulnerabilities exposure. There's a CVE database that you could go look up these vulnerabilities and see how you could go about patching them. When you start to think about a model, there's no such thing. There's no CV database for a model, and these models are. Speaker 1 And by the way, they change every three months. Speaker 2 They change very rapidly and, and, and these models are trillions of parameters, large, right? 10s of trillions in some cases. How do you go about figuring, like how is this model going to respond when you start to see these, you know, different types of behaviors? You got to validate these things. And you just cannot do this at human scale. You have to do algorithmic red teaming to be able to make sure that these models are safe and secure and these agents are running the right type of, you know, code. As you think about these agentic loop, they're using the right type of tools, they're connecting to the right type of MC servers because you know, model context protocol has become a thing where everybody's using that to be able to talk to different data sources, different APIs, different, different products and, and whatnot. You got to make sure that you're, that you're scanning every single MCP server before you go ahead and deploy that inside of the environment. So there's there are a lot of pieces to this and you just have to make sure that you have a holistic approach to completely understanding your supply chain before you start so. Speaker 1 We should talk about this entire platform approach. I just want to kind of riff off of something that you said, which is, you know, think about this like doing a background check on agents. The way I think about agents is they're like teenagers, right? They have, they're supremely intelligent, but they have no consequence, No Fear of consequence. They have no kind of, you know, if something goes wrong, they don't really have remorse. It's just like they move on to the next thing. And So what we have to do is make sure that before something goes wrong, you can intercept rather than doing it. And that's why runtime interception is so important as we go through this. Speaker 2 That's exactly right. You know, we talked a little bit about what you got to do from a supply chain perspective. But once you have that and a little bit under control, runtime becomes so critical. You got to, you got to see every single prompt, you know, request a response that's going back and forth. You got to make sure that every tool call is, you know, is happening the right way. And inside of all of this, I think one of the more important things that started to become, you know, very obvious to a lot of people is the the identity aspects of agents. You know, how do you make sure that these agents are doing what they're supposed to do? Are they scoped appropriately? And so at one time being able to make sure, you know, being able to confirm that these agents are behaving the way that they are supposed to is, is absolutely, you know, critical. Speaker 1 OK. Deconstructing the Full Stack for Agent Security So let's take a step back because what I want to do is I want to make sure that by the time the spot is over, every customer that's struggling right now with what they need to do to implement agent security has a clear playbook, OK? And so you've got about 18 minutes to make that playbook happen. So let's start with Matt. We'll start with you explain the core components in this full stack of agent security and how should a customer think about deconstructing the body of work which is to make an agent secure. Speaker 3 Yes. So there's sort of an architecture that the industry is converging around to secure agents. And one interesting part is that there's no single silver bullet. There's no single aspect like solving it on the endpoint or solving out the data layer or solving it at the identity layer. It actually requires us to sort of rethink and almost reinvent the past 30 years of cybersecurity. If you think about all the innovation that's happened in identity, networking, observability endpoint over the past 30 years, suddenly we need to take a step back and say, hey, we need all these same controls for agents, but none of the controls that we have, none of the vendors that are there actually have products that work in the agent world. And so when we take a step back, an ideal architecture for securing an agentic workforce involves something on the endpoint or close to the agent, wherever it might be. And it might not be on a laptop, it might be on a server in a data center. It's just as happy there as it is on your desk, right? There's something on the identity side. So we need to not just look at all these transactions at runtime with no context. We need to give identity and provide context around what was the purpose of. Speaker 1 It and when you say me to give identity, you, you want to make sure that every agent, every service, every machine has its own distinct identity. Speaker 3 Absolutely, and agents right now are that blind spot. Not that we were doing a great job for non human identities in the 1st place. I think a lot of organizations have struggled with anything that's not a human identity. We're just now getting to a good. Speaker 1 Point you just acquired a company Asterix for non human identity. They were The Pioneers in non human identity. Speaker 3 That's right, there's a whole bunch of companies trying to tackle this problem. Asterix has been leading the pack for a long time, discovering what are all the non human identities. And IT it's very impression of them because they did this before AI and before agents really became popular, but it positioned them very well to not just discovering non human identities, but all agent identities as well. And so discovery is a big piece of this, not just giving an identity, but discovering what are the identities of the agents that are out there. Speaker 1 Because you started with, I need to make sure that I have an inventory of all the agents that are running in my environment. I need to be able to have a continuous discovery mode that can go out and find new agents that are starting to merge up. What next? Speaker 3 Yes. So once you have that, you're in a better place than you were, you know, before because at least you understand the attack surface, then it comes to putting in place controls. So Cisco is really uniquely positioned because we know how to sit in the middle between the agent and the resource. And if you look at some of the identity providers, just doing things at authentication time is not enough and the agent could get a credential, go take access. But once you hand over an API key or a credential to an agent, you don't really know what it's doing with it. So being in the path between agents and resources becomes important. So we've got discovery, we've got the inventory, and then we have the control, the ability to look at every single MCP tool call, every single API call, every single LLM, you know, thought process. Because we can sit in the path as a gateway between the agent and all of those things it depends upon. We can start putting in really smart controls around what is that agent doing? Can we restrict what it's doing? Can we understand its intent when it does something to measure it against what was the original intent of the user or the organization? And that's where behaviour and observability start to come in. We start to need to be able to apply guard rails and even deeper behavioural controls that prevent an agent from going off the. Speaker 1 Rail. So having a full observability for AI stack that actually gives you is the agent drifting from the behaviour that you expect it to have? And if it does, can we go out and dynamically intercept that behaviour? One observe it, but two, dynamically intercept the behaviors so that we can stop from harm being done. Speaker 3 Exactly. And just observing, it's not enough. Being in the path is so important 'cause that's where we can actually stop it. So it's really all these pieces working together as a solution that becomes necessary. And so if you just put on your endpoint blinders or your identity blinders, it doesn't give organizations the complete solution architecture. Unifying Agent Security Management with Cloud Control Thing so this is this is and I want to come to you DJ, about something that we've that you've built in cloud control. But the way that this when when Matt starts to describe this whole thing, it seems like a lot of peace parts, right? And each one of those peace parts might be a product. And now it's, you know, customers like, well, I, I this seems overwhelming. I just want to go out and solve agent security. You've got a way that you've actually solved that within cloud control, which is the overall Uber scaffolding for managing any technology from Cisco, right? Walk everyone through that concept of apps that you've built. Speaker 2 Listen, the, the, the whole core premise of cloud control is to fundamentally reimagine the control plane for a lot of these types of, you know, solutions that we need to take the market. Our customers really don't want to sit down and like, you know, start putting all of these piece parts together. And, you know, the, the most interesting thing about, you know, what's really happening with agents is these things are moving so fast. You know, you might go ahead and set up a very specific type of control for say, your open claw deployment. And the next week there's a new agent at runtime called Hermes that that pops up. And now that becomes a zeitgeist. And so this game of whack a mole is not something that's sustainable when you try to keep putting piece parts together and point solutions that don't talk to each other. So what Cloud Control does is something really different. You know, we recognize that you might have to bring in different products into your environment. Speaker 1 Some of those might not even be Cisco products. They might be a third party. Speaker 2 Exactly. You know, in fact when we launched Cloud Control at Cisco Live, we announced 50 plus partners that are joining us in making sure that their products work seamlessly with Cisco's, you know, cloud. Speaker 1 Control that gets to 5000 pretty fast, by the way. Speaker 2 We're moving fast. The number of people that have raised a hand since then is absolutely staggering. So you know, we're, we're slowly starting to bring each and every single, you know, partner on board. Some of these are even competitors that are starting to connect into our solutions to be able to offer a holistic solution for our customers right now. Now, one of the key things big unlocks over here for us was we said the way we can start to manifest these solutions is using this concept of applications or apps inside of cloud control. Now what these apps do is they allow you to stitch things from different products together, ultimately solving a core pain point for the customer. When a customer thinks about agentic security, there's an app for that right, that's being built that shows you, hey, exactly what's happening. You know, with respect to all of the agents that are inside of the environment, Matt was talking about discovery and inventory. Imagine a single place where you can see all of that stuff, you know, start to come together. And then you have the ability to drill into specifically the actions that each and every single agents taking. You have the ability to apply a policy to the actions and even the the tasks and the sub tasks that those agents are taking in a single place. You don't have to worry about like trying to go to like different dashboards and, and trying to figure out how do I enforce this over here on my network, on my endpoint, on my, you know, you don't have to do that. You now have a single place to be able to express your policy and then I love it to be enforced in your entire state with a single click. That is the power of bringing these apps inside of cloud control. Speaker 1 So let me now try to see if I I can help pull this all together. Between the two two parts that you're talking about and a third part which is around agent observability. You will have identity around agents and non human identity. You will have the discovery of that non human identity so that you can constantly keep it refreshed and updated. You will then be able to make sure that you can observe the behavior of agents, observe the infrastructure that an agent is running on, and also be able to keep an eye on the tokenomics and how much token is, how many tokens are being consumed by those agents. You will then be able to say these agents have tools access and I will be able to make sure that that tools access is securely granted to the agents. And it might be a very different class of access than what you would grant to a human because you might not need to grant it very long duration access. You might need to grant it access just for a certain period of time. Get the access. Granted the action is done. You take away the access. Speaker 2 Just in time, just enough, just long. Speaker 1 Enough, just long enough. Applying Zero Trust and Open Source to Agent Security So one area that I wanted to really bring up is one of our fastest growing products we've had is secure access. And it's been, it's been something that's fastest growing from not just a top line perspective, but also from the number of seats that we've actually been able to go out and deploy in the very, very compressed amount of time. And the reason for that is this notion of zero trust for humans being able to access applications has been a pretty important problem to solve with least privileged access given to people so that you don't over provision permission. But as you get into agents, that kind of notion of access control becoming much more around even action control, because you have to be far more granular and scoped like you talked about earlier. And the permissions that you grant to the agents are super important. Talk to us a little bit about that, Matt. And how are you thinking about zero trust being applied to agents rather than just humans? Right. Speaker 3 So zero trust, the principles kind of transcend humans or non humans or, or agents. They're they're generic and they they work well for everybody. Secure access is interesting in that it's the foundation of a lot of our zero trust architecture. And today it deals with human and even devices on the network. We're extending that for agents, right? Rather than saying, hey, we need a whole new set of capabilities and a whole new product set in order to deal with giving agents access to SAS, apps, the Internet and these other things. We say, look, secure Access already does those things. And it's one of the, you know, it is the fastest growing product that we have. And therefore, can we extend it to not just be humans and devices, but also agents running on endpoints or running in data centers or cloud, getting access to the same sets of capabilities? And so when we think about zero trust for agents, you're right. It's a lot of, it's like, how do we go from access control to action control? It's not just, hey, throw the existing controls at agents and hope for the best. We need to go a bit of a step further. But this is a really good foundation for organizations to give agents secure access, no pun intended, to the resources that they need to do their job and everything around it, around identity, around behavior monitoring, around granular access control and least privilege. All that now gets inherited in this zero trust free agents architecture as well. Speaker 1 Because the way I think about this is one of the big differences between agents and chat bots is that agents have tools access. And if you can provide that tools access in a secure fashion with the level of scope down permissions that you need for only the job that needs to get done for just enough amount of time, just enough permission, just in time, I think you're going to have a whole different kind of unlock that happens with agents. And right now that over provisioning can become a huge issue from a security and safety perspective. Speaker 3 Solving that tools access problem alone would be a huge win for most organizations. Speaker 1 Awesome. So DJI had one other question that I wanted to make sure I asked you about which was this notion of open source. We have actually now over time invested and our contributors and some of the largest open source projects, whether it be ice surveillance, whether it be hotel, we've also started open sourcing our models. You just took over AI research within the company with a mean running that charter for us. Walk us through your perspective on open source and why it's so important. Speaker 2 Listen, what are the things that we have seen in the past, in the recent past is the adoption of AI tooling happens first in the open source community. We saw this happen with Openclaw where as soon as the runtime was put out there, it caught on like wildfire because it solved a very meaningful pain point because it provided A harness that I loved you to be able to run things inside of a Mac mini. And that's all very meaningful problems. And you saw that it just took off. There are so many of these AI projects that start out in the open source community and it, it's a very strong indicator for what the community's expecting with respect to these solutions. And it's very useful. And it's, it's, it's, you know, it's, it's, it's a really high signal way of understanding what the market needs. But more importantly, you're also moving the community forward. When you start to think about, you know, contributing in open source, you're, you're fundamentally changing how people start to perceive these tools. Let me give you a quick example. We put out a a skill scanner that came out as soon as Open Claw was launched. And what it did was it basically looked at any skill that you would add to your open claw setup. And there was a marketplace called Claw Hub where lots of skills were being, you know, put out there. We decided we're going to scan all of those skills because we noticed that some of these skills were actually vulnerable. They were, they were actually malicious. You could use a skill that was meant to read your inbox. Guess what? It actually would look for password resets or two facts and would send that out to, you know, exfiltrate that e-mail out to somebody else. And so we built a scanner and within 48 hours, this open source repository just took off right Today. We have, like with all the open source projects that we've contributed, we have an MCP scanner, we have a skill scanner. We put out defense Claw as a project that allows you to like protect your runtimes, agentic runtimes. It protects open claw claw code codecs. We have over 4500 GitHub stars right now. Speaker 1 Which is huge. It's huge. Speaker 2 In a very short amount of time and it tells you that the there was a need in the community and and we're fundamentally moving the needle forward for the community to start understanding and adopting AI tools safely and securely. Speaker 1 That's awesome. Then the models that these agents run on will actually constantly be validated. You'll actually have full visibility in the data flowing through the models. They'll be getting validated on an ongoing basis. You'll do algorithmic red teaming on the models, and if you need to apply runtime enforcement guard rails, you'll be able to do that automatically through AI defense. And all of this stuff will be stitched together behind the scenes, will be available through cloud control in a simple app called Agent Security. And you won't have to worry about all the behind the scenes plumbing that's happening because Agent Security will be able to give you everything in one consolidated interface. Have I got that correct? Speaker 2 That's exactly right. How Cisco Differentiates Agent Security with AI Gateways So how is this differentiated when Cisco provides this versus a bunch of start-ups providing this or other companies providing it map? Speaker 3 Yeah, there's so many pieces to it. There's certain pieces that only Cisco is really uniquely suited to deliver. One of those is discovery. We have a better view into everything. I don't want to say just in the network, but in the organization, both in and off. Speaker 1 The network? Why do we have a better view? Speaker 3 It starts with networking, right? We can see the traffic on the network and we can understand that. Hey, it looks like there's an endpoint over here that's talking to api.anthropic.com. Well, I wonder if there's something interesting going on on that endpoint, And of course there is. Having that level of DNS visibility and network visibility helps us find things that are not in the systems of record. There's no other way to find them. So network is really important for discovery. Plus, we already have all this great discovery capability through identity intelligence into the identity systems. We also have an endpoint presence. Not a lot of people know that, but Cisco's Secure client is widely deployed on millions of endpoints around the world. A lot of people started using it for VPN, but it's grown a ton since then. That's an amazing sensor. We can see we're actually running on the endpoint terms of agents. And then we have spunk and all this observability data. There's a treasure trove of discovery information hidden away in all those logs that can tell us like, hey, it looks like there are agents here because we see that something looks like an agent is getting on the network or sending a log of some kind. So between all those data sources, Cisco's very uniquely suited to provide differentiated and very broad discovery capabilities. So I think that's that's the first thing. And the second thing is really because we are a networking company at our core, being in the middle and beyond the scale as an AI gateway is really critical. We've seen that oftentimes organizations will deploy AI gateways for specific applications. Speaker 1 What about the importance of AI gateways again? Speaker 3 Yeah. So if if folks who are listening don't know like the AI gateway is really the cornerstone of what sits between the agent and the rest of the world. It's used both for the LLM traffic as well As for tool call traffic. And it's a really interesting kind of choke point in the architecture where because we have that in place, we can actually insert controls there and do deeper inspection of everything that the agents doing. And so it's really important that no agent ever has direct access to the Internet. It's important that every agent never has direct access to an MCP server or direct access to an API because it's just going to cause havoc, right? Having something in between is what really differentiates agent security versus human security. We don't need all these things in between. Typically agents do because we need to scrutinize every action. So AI gateways have become a really important kind of industry standard way in order to get control over what agents are doing and because you know and. Speaker 1 That's also where you granted permission. So just in time permission for just enough permission for just long enough. Speaker 3 That's where that gets enacted is through that gateway, because you can imagine like I might have an old application, it doesn't really understand just in time permissions. It just understands, here's your API key, you can use it for the next 50 years. You need something in the middle that understands like, OK, there's an API key here. I need to only provide a very thin slice of that. Speaker 1 Give me an example of that, because this is such a powerful concept of the Just in time, just enough, just long enough. Walk us through a real example of that. Yeah, so. Speaker 3 We, we've got a say Salesforce, very popular SAS application. People might want to start using AI to help them, you know, create opportunities to update opportunities in Salesforce. I know that's not always something that people enjoy doing, but if you had an agent that could do it, talking to Salesforce directly, then great. But Can you imagine if that same agent started, you know, deleting records and creating thousands of opportunities? You know, your manager would be pretty upset about that. Having something in the middle that says, OK, I'm running my agent to go and update my opportunity based on my last meeting with that customer. The gateway gives it just the permissions it needs to touch that single opportunity. Speaker 1 Scoping the permissions for just the very, very narrow kind of window that you wanted to have. Speaker 3 Exactly right. And that's what agents need. They need that level of control and scrutiny. So because Cisco is networking company, we know how to put things in the middle without disrupting the experience with by doing it at scale, by doing it with performance, we can sit in the middle. Speaker 1 So DJ, when all of this comes together in Cloud Control, what is the end user's experience look like? Speaker 2 The. Speaker 1 You know when I say end user I mean the IT administrator or the security person within within our customers and then also the the eventual end user that's dealing with the agent. Give me both the experience. Speaker 2 You got to have, you know, you got to make sure that you're making this experience amazing for both of those personas, right? Because the IT admin has this constant challenge of like, you know, calling right now, you know, agents are going to cut across your network estate. You're going to have to enforce policies and the identity side of the house. You're going to have to enforce actions and guardrails on a, on a, on a class of agents that are going to do very different business tasks. So which means, you know, the, the job of an IT admin now becomes a lot harder if you don't give them a single place to be able to manage all of these things. They're going to have to play whack A mole and chase a lot of people down and, and then file tickets and takes forever to be able to make a change. What what cloud control really does is it it sort of brings this all together in a single place for the IT admin that's. Speaker 1 Huge. Speaker 2 Change in terms of, you know, day 2 and day three operations that that these admins have to perform now on the same, you know, the same note when you take, when you think about how people are starting to use agents, they're starting to use agents from wherever they are. You know, it could be over a slack thread or it could be over a Webex message or it could be a text message over Discord. People are starting to communicate with agents in like in, in different shapes and forms. So it becomes really important to be able to have a user experience that notifies them when they do something, you know, that is not quite right. Or when the agent goes out of control or does something wrong, the ability to inspect and respond needs to be available to those users. And this is where the AI gate based only becomes incredibly important, right? You now have the ability to inspect something and, and, and, and instantly inform the user saying, hang on, this thing is not the right thing. Agent's not doing the right thing. I'm going to block this action. I'll give you a quick example. Right now, we've had guardrails in the AI, you know, for the longest time and and as part of the gateway that we deployed, what the gateway does is it basically tells you when an application decides to start sending emails to your board members, you know, it'll go ahead and send out a quick, you know, notification saying, hang on a second, this is not a behavior that is expected this. Speaker 1 Goes back to the scoping of permissions. Like you know, you can give people you can give an agent access to your e-mail. Just not enough to go start sending random emails to the board members. Speaker 2 Correct, but you've got to inform them why it's going to not take action on something, right? The end user needs to be notified saying hang on this particular task. Speaker 1 That's a really important piece is you have to go back to the end user saying this was something that the agent was not granted permission for. Speaker 2 For these reasons, correct, and that closes the loop. At that point, the user recognizes and then reinforces that behavior and, and we can in a as part of a right, you know, strategy from a security vendor perspective, which is sort of what we do. We we make note of that and then we constantly keep improving how we behave the next time you know that happens. So you got to have these dynamic guard rails that are evolving policy that is constantly evolving it. None of these are static, you know, solutions that are going to work because your agents are also evolving just in terms of the the nature of the task that they perform. So if you were thinking, hey, I'm a security admin, I'm going to use a regular expression that I'm going to drop in there that's going to take care of all of my problems, boy, you couldn't be more wrong. Any solution that just gives you fixed rules to be able to go out and solve these problems or not? Speaker 1 You have to be dynamic in Azure. Speaker 2 Exactly. Inside Cisco: Culture, Innovation, and Future Vision So DJ, let me actually just give you folks an idea of what I particularly like about what you've built because I think it's not that obvious when people see it first time. Like if you see a demo for the first time and you see this app that was built, that was an agent security app that was using all the platform services from maybe 5 or 7 different back end areas. You know, identity and secure access and AI defence and Splunk and all those. It might not be that obvious that what we have done is completely taken away any seams from any of those and you haven't burdened people with a very large set of platform capabilities that you might have. You have only gotten them focused on an app that solves a use case. And today you might have agent security as the really, you know, sought after use case. Tomorrow there might be another use case that comes up. Hey, tomorrow you might even have a use case that needs a third party. And the app that you build can completely incorporate our platform services and other people's platform services and put it into that app. And that app is all that the customer needs to think about because that is the use case that they're deploying. That's right. That in my mind is something that is superbly unique and very powerful that I don't think is that readily available by any other player in the market that I've seen. Speaker 2 It isn't, and we're making it even easier to be able to build these apps. One of the advantages of using agentic coding is to be able to turn around. You know, these asks from our customers when they come in and say, hey, here's a use case that I have that uniquely cuts across these products and these platform services that you offer as Cisco, We now have the ability to go back to them and say, yeah, you know what? We built an app for that that solves that pain point in a unique fashion for these classes of customers that care about that. Speaker 1 So in reality, our platform services for all practical purposes could be headless platform services and the agent is the one that's actually controlling those those platform services to the degree that they want for that use case. Speaker 3 I think a lot of customers would rather not deal with UIS anymore because I go through agents. Speaker 1 And that's exactly right. So let me ask you a question. Let me just zoom out again on you folks have both been here for a couple years now. Each one of us have spent a lot of time together. I'm I appreciate you guys putting up with me. What has surprised you about being at Cisco Map? Speaker 3 Yeah, so I'm a I'm a Cisco boomerang. Speaker 1 Yeah, you are. You're the youngest distinguished engineer in the history of Cisco before you came back. Speaker 3 That's what I've been talking. Speaker 1 About no longer the. Speaker 3 Youngest though check the HR Reg No very true yeah what surprised me I I think Cisco coming back so it's an interesting perspective right cuz I got to see Cisco back you know five years ago 15 years ago even and the the speed of change and the organization around shared outcomes now I think it helps to have a single chief product officer that's. Speaker 1 I beg you to say that. Speaker 3 I think to be honest, like that, that's driven a lot of amazing convergence around like how do we get identity working with networking, working with AI, working with observability? That's been pretty, pretty great. And then the second piece is really Cisco's relevance, I think at this point in time is more so than I think I, I've seen it at any point in time over the past 15 years just between the, in the silicon AI networking pieces, everything that we've invested into security, the startups that we've acquired to go accelerate that. It's a really interesting moment for Cisco in particular, because if you look at US versus every other company that's out there, it's totally unique in that we are the critical infrastructure company and that networking company kind of the, the glue of a lot of organizations. They, they rely on Cisco and it, it, it puts us in a really interesting kind of unique vantage point within the. Speaker 1 Industry, I think of us as this kind of vertically integrated platform that is a Co designed full stack. And I think I don't think that's actually fully rocked by people on how difficult it is to go out and do everything from the silicon and to photonics to systems, hardware, system, software platform for observability, security models, applications, agents all tied together. DJ what surprised you the most at Cisco? Speaker 2 You know the culture of the company. You know, there's a popular perception from the outside that large companies are dinosaurs. They're slow, you know. Speaker 1 Specifically, Cisco is a legacy company with how people. Speaker 2 Use exactly, you know that that's the perception from the outside. One of the most interesting things that happened, you know, was the speed with which we pivoted. Like pivoting a, a giant ship like this takes a tremendous amount of energy when AI started moving the way it is and you know, the, the cultural norm that was established inside of the company to say, no, we're going to be part of this movement all the way from, you know, from you to like, you know, to Chuck to like everybody that just stood up and said, hey, you know what we, we missed the. Speaker 1 The Cloud Wave. Speaker 2 The Cloud Wave, we're damn sure we're not going to miss the AI wave. And and that the, that, that sort of humility, that understanding, like, hey, listen, we're not going to miss this thing. And we're going to do right by, you know, our customers, we're going to do right by our employees, our shareholders. I think that was a phenomenal feeling. And when you walked into that thing and you realized that, wait a minute, you know, if you say that something needs to change, it happens. You know, if you look at what's happening with respect to AI, you bring a new perspective, people were willing and, you know, ready to accept that and, and move forward fast. So I think, I think, you know, anybody that's out there thinking Cisco's a legacy company is underestimating. And I'm, I'm fine with them underestimating because you know, it's, it's such a wonderful feeling when you, when you exceed expectations and that, and that's what we've been doing consistently for the past three years that I've been here. I've noticed that we've been systematically out executing and and it feels really good to be part of that. Speaker 1 What would you guys say would be things that you continue to remain dissatisfied about that we could do better in? Speaker 3 Hey. Speaker 2 It's a here's the thing, right? There's a natural amount of inertia that exists when a company this large as well, right? There's, there's and it's necessary, like there's necessary friction in place so that people don't go rogue in doing some of these things. And like, you got to have a check and balance in place to be able to do these things. And that's really how you build an organization at scale. Sometimes that that, you know, as a startup entrepreneur, you know, you kind of feel like that's too slow and, and you feel like you want to move faster. And, and I think I, I think we do a lot of things right. And at, at some point we have to sort of take a step back and be like, hey, listen, there are certain places where you have to move faster and, and, and certain places a little bit of friction is OK. Having that understanding, you know, sometimes that equation is not perfect. And, and you have to break glass to be able to make people realize that, no, this is something that we're going to move fast on. And this is something that it's OK, you know, us taking a little bit of time to figure things out and getting certain things right. And, and I think that's something that, you know, we're constantly having to assess and reassess and, and we should be, we should be challenging ourselves and it to say that, hey, here are places where we can do better. Speaker 3 Yeah, I think for me like, I think striking a nice balance between being that kind of hardware networking company as well as software and relevant at all levels of the market. I think oftentimes we'll get pitching holed into selling to service providers or enterprise Oregon government and those are really, really freaking important. I think there's a large swath of folks who grew up learning Cisco and had massive, massive appeal across the board. I think it still does in a lot of places, but I'd like to get back to that kind of mass appeal kind of bottoms up motion of making sure that the the the industry is learning on Cisco even at the smallest scale and then graduating up through that. I think there's more that we can do there to to do well by that. Speaker 1 OK. I always tell people stay, stay constantly dissatisfied and keep making sure that you, you move, you move at a pace because I think speed and being early to market is actually a cultural Moat that every company needs to have in this era. So keep pushing us the way that you are. I think it's been fantastic what you folks have done. The fact that you built Cisco Cloud Control in nine months when the original scoping of that project was about 7 to 12 years is nothing short of remarkable. I think the fact that you've done that, all of the pieces that you've done in agent security in the window of time that you've done it has been amazing. Keep doing what you folks are doing. Thank you again for being here, and I'm sure we'll have many, many more of these conversations. Speaker 2 Thanks for having us. Thank you.