2. Building Future-Proofed Workplaces with Jeetu Patel and Anurag Dhingra

Jeetu and Anurag Dhingra explore the transformation of physical infrastructure and intelligent software as the foundation for the workplace. As the nature of work continues to evolve, the demand for environments that are both seamless and secure has never been higher.

Transcript

SPEAKER_01 0:01

Hello, everyone. Welcome again to Cisco's Inside Product series of conversations. And I have with me today my good friend Anurag Dingra, who is the Senior Vice President of Workplace Technologies, which includes our campus and branch networking, which includes collaboration, which includes everything else. We all work for you, man.

SPEAKER_00 0:21

It's nice to be here, Jeet.

SPEAKER_01 0:23

It's great. Congratulations on all the amazing stuff that you folks have been building. So let's let's take a step back. Sure. What is happening in AI and how is that actually impacting the workplace?

SPEAKER_00 0:33

Yeah, so it's a really good question. You know, um we've talked about this before. The AI state of the art has moved from chatbots to agents. And when you think about what these agents are doing, they're starting to take on tasks that you and I would typically do. And then the interesting thing is there are big implications in the workplace. There's implications on the network, there's implication for security, there's implication for tools that we use for collaboration. And uh we have some research that we've recently published around the traffic volumes and how much an AI agent consumes for the same task versus a human. And so you have to think about if you're building infrastructure today, how do you future proof that? How do you put something in place today from an infrastructure point of view? That is going to stand the test of time for the next couple of years as a state-of-the-art default. So so the uh workplace has to account for a blended team of humans and agents working side by side. And that's really where our innovation is focused on right now.

SPEAKER_01 1:25

And that that completely changes traffic patterns compared to just humans working.

SPEAKER_00 1:29

Absolutely. So you know, when you think about what you're delegating to these agents, these are tasks that we used to do. These agents are using the exact same tools, exams, exact same applications. Yep. And they don't take a break, they don't get tired, they're on 24 by 7. And so the network is new.

SPEAKER_01 1:43

No personal days off, no vacation, no sick days.

SPEAKER_00 1:46

Network is never idle. So you have to design for this new world uh and it's much different than how it used to be when they were just humans.

SPEAKER_01 1:53

And and is um that that redesigning the network, is that gonna be um 20% increase in in capacity? Is that gonna be uh 200% increase? Is it gonna be a 10x increase? What do you think is gonna happen in the next five years?

SPEAKER_00 2:10

I think we are in the very, very early phases of this, right? I think you can draw inspiration from what's happening in software engineering as an example, right? When I look at how our engineers are working, they're using coding agents, they are starting to now go from one coding agent to a fleet of agents. Uh and that's really the you know sign of things to come. So imagine every human had dozens, if not hundreds, of these agents that they're delegating tasks to. So it's just that factor, right? If you have 10 agents for every human, there's like 10x increase in traffic volumes. Right. But I think this is going to be much, much better.

SPEAKER_01 2:40

Right, because they're working 7 by 24, so that's 30x, and then they're adding five times the amount of pro you know amount of bandwidth, so that's another that's 150x. Absolutely. So it just keeps compounding.

SPEAKER_00 2:51

It just keeps compounding, right? Exactly. Yeah.

SPEAKER_01 2:53

And so, okay, so that's great. So given that, there's a lot of people that provide networks, even though we are the largest. Yeah. Um and you were um uh given a pretty mighty task of saying don't just grow this business at the rate of inflation. Grow this business at at at very high rates and you've delivered on that. But in fact, what was interesting was this industry, by the way, has been pegged to the rate of inflation for the past, I don't know, 30 years, where you know, three, four percent growth. Right. Uh order growth was twenty-five percent last quarter. Yeah. Um and why why do you uh I mean the answer is obvious. It's because of agents, it's because of new um new kind of implications of uh how people need to upgrade their infrastructure because of methos, all of those things. But walk people through how organizations should be thinking about their networking apparatus and infrastructure.

SPEAKER_00 3:48

So the way I think about this is how can you turn network into a differentiator for you, into an advantage for you. And uh the goal most organizations have these days is how do you scale AI to every single individual in the organization. And so the network can be either a hindrance for that or an enabler for that. And that's the way to think about this. Right. And when you uh want to make it an advantage, when you want to make it a differentiator, then you have to make sure that the traffic volumes that we're talking about, the traffic patterns that are shifting, the network is capable of handling that. So the speeds and feeds matter and then they're gonna matter even more as these agents proliferate. But then the question is how do you deal with security posture now? There's a completely new surface area because of these agents, their patterns are changing. And the interesting thing is everything connects to the network. People connect to the network, machines connect to the network, IoT devices connect to the network, agents connect to the network. So the network becomes a very important control point for policy enforcement. It can monitor everything, it can watch behavior, and it can provide those signals to a policy engine and then enforce the policy. So the network becomes a security enabler. And then the third piece here really is how do you make it all simple to manage? Because at the end of the day, humans and IT teams are struggling to find capable people with the right skill set entering the IT workforce. So how do you use AI as your companion to to make you manage all of that? So when you wrap that up, you know, simplify operations, security that is tightly integrated in the network, and a scalable uh pool of network devices, whether those are switches or routers or access points, that combination is really how you scale AI.

SPEAKER_01 5:25

So I I I like the the three S's, like you know, you have to scale the network, you have to make sure that you're securing it and you've got to simplify it. Um walk us through first, how is Cisco different compared to everyone else? Yeah. Um and two, what would you um how would you envision like if you had a magic wand and you could go out and have a customer say, utilize all the richness of capability that you've built, what would you want them to actually have? And what what does that world look like for them?

SPEAKER_00 5:57

Yeah. So I think the biggest uh way that I think Cisco is different from everybody else is the breadth of our portfolio. Yeah. You know, we have so many things that we do with this company. Obviously, we're a networking company, but we are also a world-class security company. We are a world-class collaboration company, we are a world-class observability and data platform company. And when you bring all of these pieces together, that is actually what you need to scale AI and provide people the tools to work with AI agents. And what we have done at Cisco is built a platform. We brought our portfolio together, we've consolidated, there's a single front door now with Cisco Cloud Control, and that is a natively AI-first agent-first platform. So all of the things I was talking about in terms of simplification of operations, security policy management, and managing your full estate comes together in one front door. And then, you know, you've talked about this before. You know, what does it mean to be a platform? It's you know, when you have one piece of technology, that's great. But when you add the next one, the incremental cost of that ingestion should go down, and then you should compound the impact. And that's really what we're building here. And so I think that's really what sets us apart, where you can do these things piecemeal, but the full stack, co-designed uh stack, I think that delivers value that you just can't get anywhere else. So one very specific example is when something goes wrong.

SPEAKER_01 7:13

Okay.

SPEAKER_00 7:13

You know, so you are trying to use an application, you are on your laptop, and you're not able to access the application. Is it your laptop? Is it your Wi-Fi access point? Is it your internet connection? Is it the application? How do you even debug this? How do you even start? It takes days sometimes to nail this down. And when something goes wrong, it's always the network. It's always the network. And you know, one of our teams has this t-shirt, I really like it. It says go ahead, blame the network. That's because the platform actually brings all of this telemetry across the full product portfolio together, and then you have AI that can reason through that and figure out where things are broken. And that's the platform advantage when you can do that end-to-end.

SPEAKER_01 7:52

So you have this really kind of interesting articulation of the vision that we've been talking about, which is I've got these ambient agents and they are sprinkled all over my network. Walk through that vision on what what what it is that you've actually done to really fundamentally reimagine how an organization manages the infrastructure within their estates.

SPEAKER_00 8:14

So I think that's it's a similar journey going from chatbots to agents, right? The the holy grail, you know, the the thing that we want to solve, that our customers want is can I have agents that can run operations for me, like fully autonomous? Right. That's what we want to build towards. You don't just flip a switch to go there. You can't just say, look, it's a black box, trust me it works. Never works in production like that. People don't trust it. People don't trust it. So you have to give them enough building blocks that they can test themselves. Then there's an audit trail, you can inspect the work, and you get to the autonomy. So what we are trying to do is You want to build confidence over time. You want to build confidence over time. And then when you are uh confident, when you are comfortable with the reasoning that these agents have and the actions that they're suggesting, then you can start to delegate control. So the way to build this in in our mind is first of all, you need to have agents that are always on. This is the ambient agent we were talking about. They're monitoring the telemetry, they're monitoring all things that you care about. Are people able to connect successfully? Are they getting application experience that is amazing? All of those things that you care about from a user experience point of view, they're monitoring that. When they see that something looks off, you know, you've crossed a threshold, errors are uh increasing, they get into reasoning mode. They use the best models. Some that we have created, we have a deep network model and frontier model.

SPEAKER_01 9:34

And describe the deep quick sidebar, describe the deep network model.

SPEAKER_00 9:38

So when you think about models today, people immediately think of the latest and greatest frontier models, you know, models from OpenAI, models from Anthropic, Google. And those are general purpose, large models. They are really good at lots and lots of different things. What we were trying to do is look, for network operations, you want a model that understands networks intimately. Right. But you're not going to ask this model to write poetry for you. And so what you can do is build a specific, deep model that is not trillions of parameters, it's billions of parameters. It's domain specific. It's trained on all of the learnings that we have as a networking company over the last 40 years. And so it's really good at networking operations. So that's a model we have built. But you know, sometimes you need to look at things beyond just the network, you have to collect telemetry from lots of different places. So frontier models have a place uh there as well. So use an ensemble of models, R models, bespoke models, frontier models all working together. So that's the reasoning part of it. Then there's a question of, okay, based on reasoning, you came up with a remediation of how to fix it. How are you going to make that safe in production? So this is where the idea of a digital twin comes in. You want to be able to test that in a digital virtual copy of your network. And once those things look good, those tests pass, then you deploy that in production. And all of those blocks, you can inspect, you can ask the agents to ask you for approval at every single step. But then if you feel comfortable for a certain class of problems, you can just say next time when the agent runs into this, it can just fix it itself. That's the dial. That's so it's not a flip uh switch, it's actually move the dial. And then you do that slowly, and when you're comfortable, you have an autonomous network.

unknown 11:19

Trevor Burrus, Jr.

SPEAKER_01 11:19

And you've you've built outcomes that are largely deterministic on a foundation that is non-deterministic.

SPEAKER_00 11:26

Right. So that's that's the key thing, right? So look here is if you take a step back and you know just think about what makes an agent different from a chatbot. I really like this story that I heard Peter Steinberger tell. I mean he's the he's the inventor of the open claw, right? So he he coded up the very first version of open claw. He was communicating on WhatsApp, sending messages, and the agent would do something with the come back with a response. He went on vacation, and while he was traveling, just by chance he recorded a voice note and sent it to his agent on WhatsApp. And the agent responded back, and when the response came back, then Peter was like, I never coded voice capabilities. How is this working? So he asked the agent, How is this working? And the agent said, Well, I received a file, I couldn't figure out what it is, I inspected the file, it was an audio file, I converted it into a format that I understand. I wanted to then convert it into text. I couldn't find anything on your machine, but I found your open AI credentials that you have given me. I uploaded the file to OpenAI, turned that into text, and now I understand what you're asking me to do. And so the point is if it was a chatbot, it would have said, sorry, I don't know what you're telling me. The agent improvised. Yeah. The agent had a goal to serve Peter. The agent had never seen this problem before. And so the agent got creative and came up with an answer. And so the agents will be persistent in pursuing a goal. They'll be very creative. Now imagine an agent in the workplace. You give it a goal, and it's like, oh, G2's asking me for this HR data. I'm gonna go get this out of that ecosystem. I'm gonna let him feel like a pen test uh on your network. And it's not like somebody's trying to hack into the network. The agent is trying to serve you. Yeah. And so how do you deal with that type of creativity? You need a harness. You need a harness that can control this, that can give it guardrails, and then you need controls so you can inspect the work. And that's what cloud control, by the way, is it's uh it's a harness for these agents. And so that's the way to kind of go across that uh autonomy dial. Trevor Burrus, Jr.

SPEAKER_01 13:29

And so as you've as you've started building this, and you've actually the one thing that's been great is you were among the first people in the company that um you know jumped on the bandwagon and said, we're building this unified platform. Yeah. I'm gonna make sure that Meraki and Catalyst and everything is one unified thing. You unified the hardware, you unified the software, made it available in cloud control. It's actually you were one of the first ones out there. What made you so convicted on the fact that this is the way to go? And what have you delivered so far? What should people if people know the um the campus and branch networking from a year ago, what do they not know about workplace networking today?

SPEAKER_00 14:15

Yeah, so first thing is, you know, like you said, we had two completely parallel product portfolios in networking. Right. We had Catalyst, amazing product, and we had Meraki, also amazing product. Yeah. And two different products. Two diff two different products. There was definitely a lot of overlap. You have switches here, you have switches there, you have routers and routers, Wi-Fi. And we were putting our customers through a lot of difficulty in making the decision. And we had all these caveats. If you want to manage this from the cloud, these are the devices you pick. If you want to manage them on-prem, this is what you pick. I mean, that's not the way to buy, that's not the way to operate. So what we've done is we should have done this many years ago, but you know, I'm glad we did we did this now. We brought it together and said, we're just gonna build a a switch is a switch. A router is a router, we're gonna have the same piece of hardware, we're gonna standardize on a single operating system, and that operating system is cloud native, so you can manage it from the cloud, but it preserves all the sophistication of catalyst, so you can manage it on premise through CLI, you don't have to give that up. Right. And by the way, a single licensing model, you decide you want to manage it on-premise or in the cloud. That choice you don't have to make upfront, you just buy a switch, you just buy a router, buy a wireless access point, and then it works the way you want it to work. And so we've done that. In addition, when we were going through this, we were like, well, AI is happening. You know, there's gonna be traffic pattern change, there's gonna be security change. What does the future of a networking device look like? And I mentioned this before. What is gonna stand the test of time? So we refreshed the whole portfolio. We built switches that have 8x the capacity of our previous generation switches. We have built core fixed uh switches for campus so your backbone can carry the traffic that these agents are gonna put on that. We've put Wi-Fi 7 access points out there, the latest and greatest generation, and then we've wrapped all of that with the amazing software, connected that through a through a platform. So that is a very different product portfolio than we had 18 months ago. It is ready for AI and it is built with simplification in mind, but maintaining all the sophistication that we that we had in our portfolio.

SPEAKER_01 16:25

And so this this notion of like I'm I'm gonna inject a ton of simplification without losing the sophistication, which is what you just talked about, is is what you've delivered. And it's in a fully vertically integrated stack. So we make our own silicon, that silicon goes into your system's hardware. Right. Um that system's hardware is running on our OS. Right. Though that OS is managed in um a unified management plane in the cloud. Right. You've got security baked into the fabric of the network, and then you've got agents that are going out and managing that entire management plane.

SPEAKER_00 17:02

Exactly. So that's a very different stack than we've ever had in the past. And this is the type of stack that you're fully observable, of course. Fully observable. And this is the type of stack that you need for the modern workplace. And then by the way, let's not forget our collaboration technology. So network obviously is the foundation, connectivity is foundational. But then on top of that, the tools that we use to communicate with each other, whether that is a a meeting tool or a messaging tool or a video conferencing device that goes in a conference room, that device also understands networking intimately. Of course, it works on any network that you have, right? It's not like it's just tied to the Cisco network. But when it's combined with a Cisco network, you plug it in, all the media, audio, video, control channel, everything flows over a single cable, power is delivered over that, the network on the other side understands this is a Cisco device, it can automatically configure it, automatically secure it, and all of that shows up in cloud control from a manageability point of view. So it's it's a really, really, you know, when you combine all of these pieces, you know, going back to what you were saying, the incremental cost of ingestion is very low, but then the compounding effect is uh where the value is.

SPEAKER_01 18:11

So walk through like because there's a there's a cultural change and shift that we have gone through in the past few years, past couple years, I'd say. Right. Um and you know, it used to be that we were very incremental in our innovation. And now we've actually got step functions of innovation that are coming about. What did you do with your team to make sure that this was actually instilled?

SPEAKER_00 18:37

Yeah, so I think look, culture change is not easy. Yeah. And culture change is hard when only one person is trying to do that. So I think the good thing that we had uh is once you stepped in as the CPO, the team that you have assembled, uh, my peers and all of us, I think collectively as a leadership team are very well aligned in terms of where we're going. And I think once you have that level of alignment, all the other things become easy. All of us are pushing uh very much in the same direction, and that's what the teams are seeing. But the other thing that is happening is we are living and breathing AI ourselves. That's right. All of our workflows, the way you work, the way I work, is constantly pushing the envelope, right? And there are two parts to this. You know, I was sharing the story earlier with a customer. I'm an engineer by training. You know, I started at Cisco, I was writing software for a while.

SPEAKER_01 19:27

Yeah.

SPEAKER_00 19:27

And as I started to take on leadership roles, that I stopped writing production code. For several years I did not touch code. Late last year, I saw how my engineers are using coding agents, and I was, you know, coding for fun at home. Yeah. So I code-writing again. I I I called up my head of engineering for Meraki, for ThousandEyes, for iOS X and said, hook me up with code access. And they started laughing in my face. They're like, we're not gonna let you anywhere near uh production systems. But they did eventually, and I was able to be productive within hours with that. And and that's the, you know, when you're now starting to contribute code.

SPEAKER_01 20:04

Everyone is a builder. Yeah.

SPEAKER_00 20:06

Everyone has transformed their work uh the work style and workflow. Yeah. And I think that is really what's pushing the envelope in in how we are operating now. Aaron Ross Powell, Jr.

SPEAKER_01 20:13

That's awesome. Well you should you should feel very proud. What has surprised you in the um in the take rate of some of the technology you're building with customers?

SPEAKER_00 20:22

Yeah. Uh look, last year at Cisco Live, almost a year ago, we had a hypothesis that agents are gonna proliferate in the workplace. And so we came out with this architecture, an AI ready secure network. Yeah. Simplified operations with agents, security baked into the network, and scalable new generate next generation devices.

SPEAKER_01 20:41

The conversation I was having back then, this is Which by the way, I don't think you mentioned this yet, but your entire fleet of devices you have refreshed at the moment. Everything, yeah. Everything is refreshed.

SPEAKER_00 20:49

New switches, new routers, new wireless access point. And we have, by the way, a whole line of industrial networking devices as well. All of that is refreshed, right? So last year when we were having this conversation, I saw customers really resonate with the security angle of AI, really resonate with the operational simplification. But when you talk about scale and when you talk about agents, everyone's like, I don't know, man, I'll s I'll see if that happens. That changed with OpenClaw. Now, when I talk to a customer, nobody's even questioning that. They're all asking the question, how do I scale this and how fast do I scale that? And so that time window was so short where people went from disbelief to how can I get this? That was the surprising. Look, we had a hypothesis this is gonna happen. It happened even faster than we anticipated. The good news we were building for that world. And so we were ready, we are ready when our customers come to us now, and then we can show them how to scale that out.

SPEAKER_01 21:44

So talk a little bit about this notion of token costs rising and new local inferencing starting to actually also take off and what that has as an implication for the network, right?

SPEAKER_00 21:57

Aaron Powell So that's sort of the next frontier that That is again, these frontiers are happening so fast right now, this could be in within the next few months, right? So as you start to use agents more to do work, they are all consuming tokens, right? And the tokens are getting more and more expensive. The tokens are really expensive. And so if you scale it out to everyone in your organization, you're looking at a huge bill. And so now customers are starting to think about how do I make it economical and affordable. And that is where we were talking about the smaller bespoke purpose-build model that we have built for networking. Those type of models are very useful for purposeful tasks. And then you can start to do that inferencing for that model locally, on the edge, on a computer that is sitting on your desk or by your s by your desk. So this is where the whole category of desk side computing is emerging now. And if you think that's the way to scale AI, and if everyone's going to have a computer that is powerful enough to run these models locally and agents in them, then that has a completely different uh angle again for traffic and the speeds that you need in the campus. Because ultimately, even to connect to applications on the cloud, even the in the models, you have to traverse the campus network. And so I think as that desk side computing takes over, we're probably going to have to lean in even more on making sure the network is ready to handle that, uh handle that traffic pattern.

SPEAKER_01 23:20

And it's it's it's going to be I think it's going to be at least an order of magnitude larger from a bandwidth perspective. Yeah. Um and as a result, you might need to also start thinking about efficiency and power in a different way than what we thought about before. Now the good news is because we've had so much experience in serving hyperscalers and neo clouds and service providers and sovereign clouds, a lot of that technology from the silicon onwards can now start to get to make its way to the campus, which is so exciting for campuses. It is very exciting. Yes.

SPEAKER_00 23:51

Yeah. So I think that's the advantage of building our own silicon and working with hyperscalers, working with frontier labs, because we have a front row seat into where things are going. Right. And then what's happening in the data center today, we can learn from that and prepare ourselves for what is going to happen in the workplaces. And then we can have the consistency of silicon across all of our products that make us move really, really fast. And we can pivot much faster than we would be if we're dependent on merchant silicon for a lot of this. Let's talk about mythos a little bit. Let's talk about methos a bit. Um so I think uh one of the things that is happening with these frontier cybersecurity models is they're very good at finding exploits and not just one off targeted bugs in products, but they can actually connect the dots across many smaller vulnerabilities and then cr turn that into a bigger exploit. And so I think what we have to do as defenders is first of all make sure our products are secure. And so we are using these frontier models with hardnesses that we have created to make sure our products are are good. But then we have to help our customers do assessments in their environments because this is not just a networking concern, this is not just a collaboration or this is like every product out there. Right. So we have to make sure our customers have those tools, we have to uh give them that. And then we have to make sure that our mental model for how you patch infrastructure is gonna be very different going forward. And you know, in some customer scenarios, I hear them talk about, hey, look, I heck I cannot even apply a patch if there's gonna be a moment momentary glitch in the network, and I do that around 4th of July and Christmas because those are my two holidays when the network is idle. I mean that just doesn't work anymore, right? You have to be able to apply these patches much differently. And we have an obligation to make it easier to upgrade infrastructure, but then we are also doing innovation like Live Protect, which is an instant shield that you can deploy to your gear, and that can give you some protection while you prepare for a patch.

SPEAKER_01 25:49

So I think this just changes the thing to explain to the audience, which is from the time that a vulnerability is announced to when a patch is applied is typically 45 days. Am I right?

SPEAKER_00 26:00

Well, so I actually saw a chart recently uh which basically said if you go back in time four years ago, the time that the vulnerability was announced and a well-known in the wild exploit was 10 months. This is four years ago.

SPEAKER_01 26:13

The exploit was ten months. The exploit was ten months.

SPEAKER_00 26:15

From vulnerability being discovered to expert. Now, even in 2026, before Mythos and these models was hours. Yeah. And now it's minutes, and that it's only going to get worse, right? So even before these models, that time had shrunk so much. You just couldn't afford to wait. Trevor Burrus, Jr.

SPEAKER_01 26:30

So that's actually the point, which is if the patch takes 45 days, but the exploit happens in nine minutes, you don't have 45 days. You've got nine minutes to make sure that you come up with a solution. You have to do something. You have to do something. But you're not going to be able to do a patch realistically in nine minutes. But what you can do is apply some kind of shield or compensating control until the patch arrives. Exactly. And so that's what Life Protect does is it's actually got the stacking of shields where you can have up to like I think 20 shields that can be applied while the patch is still uh in progress. Right.

SPEAKER_00 27:03

So that's a very uh I think useful innovation that we are bringing to market right now, and that's going to be across our whole portfolio. So we'll be able to protect everything. One other thing I'll say again, going back to security, is you know, one of the questions that people ask me is like, aren't these agents running in the cloud? Like, why do I have to worry about it in the in in the campus uh network? And the way to think about this is the brain of the agent is in the cloud. The hands of the agents are here. This is where all the tools are on your campus network. And so the damage will actually happen here where the tool execution is happening. And so when you think about it this way, it is extremely critical to protect uh your your environment in the campus and branch network. And that's why we are doing all of the work that we are doing with our security team and bringing that as closely uh together with networking as possible, because like I said, everything ultimately connects to the network, and that's a really important enforcement point for protection.

SPEAKER_01 27:58

Yeah, yeah. Um this is great. This is great stuff. So um any question I didn't ask?

SPEAKER_00 28:05

Well, I think uh we covered a lot. The future is bright, the innovation machinery has never been uh this fast, and I'm very excited. And this is an exciting time to be in tech. Exciting time to be at this company.

SPEAKER_01 28:15

And and uh you know, congratulations to you and your team. I think you're doing a fantastic job. Uh and my only ask is you know, let's keep the momentum going and let's keep the acceleration um uh going. But I I I couldn't be happier with what we've been able to build. And um the beauty is that we're just getting warmed up. We are just knowing. I can't talk about all of the pipeline, but it's like that's gonna be exciting. Thank you, Interrupt. Thank you, G2.