Cisco Live 2026 Amsterdam: Why AI Agents Fail Without Infrastructure Ready For Scale

[00:00:04] What does it actually take to run AI at scale once the experimentation phase is over? Yep, AI and agentic AI conversations are everywhere right now. But so much of the attention still holds on models, tools and shiny demos. Far less time is spent on the infrastructure underneath it all. Yep, the networks, the compute, the security layers,

[00:00:29] all these things that determine whether AI agents can operate reliably in the real world. Well, today here on Tech Talks Daily, I'm joined by Rob Lay from Cisco. And together we're going to explore what an AI ready environment really looks like, especially as organizations begin to move from pilots into production. And we dig into why technical debt suddenly matters again, how agentic AI changes the demands on infrastructure,

[00:00:59] but also why security and resilience can no longer be treated as bolt-ons. So if you're responsible for scaling AI across a business or wondering why some projects fail while others move fast, this conversation will get into the foundations that often decide success or failure. So when AI starts making decisions on your behalf, is your infrastructure really ready for it?

[00:01:26] Let's find out more as I get my guest from Cisco onto the podcast now. So thank you for joining me on the podcast, Rob. Can you tell everyone listening a little about who you are and what you do? Certainly. Thank you for having me. So my name is Rob Lay. I'm the CTO and Solutions Engineering Director for Cisco in UK and Ireland. So what that means is that I lead all of how we approach our customers with our technology,

[00:01:53] how we figure out how our technology is going to help them fix their business problems. And I've got a bunch of very, very smart people that work in the organization, all solutions engineers, all those people who build that relationship with the customer and help them figure out how we can really help them with their technology and their business problems. I've been in IT a long time, 25 years, been at Cisco for about four and a half, and worked at various other large sort of systems integrators and that sort of stuff through the years,

[00:02:21] companies like Fujitsu and NTT and that sort of stuff. So I did start life as an engineer, but it's been a little while since I've done anything deeply technical. You say that, but it's still in your veins. I think that stuff never leaves you. And yeah, you must try and get your hands dirty now and again, I'm sure. I break out in sweats when people start asking me to do demos and things now. Now, although we're recording this early in 2026, I've already been to several tech conferences on this side and the other side of the Atlantic.

[00:02:50] And predictably, when people talk about agentic AI, the focus is very typically on the models themselves. But what caught my attention from what you're doing at Cisco is from your perspective. Tell me about why infrastructure is becoming the real consistent on scaling AI agents in production. And what does an AI ready environment actually need to look like? Because there's a lot of confusing and conflicting information out there. Yeah, I think it's a really interesting space. Right.

[00:03:20] We are in the midst of what I believe is the biggest transition that we've seen in decades. In terms of the impact it genuinely has on the way businesses operate. We had things like the cloud transition, but that was quite a technical thing that sort of sat in the background. AI is going to touch every part of businesses, how they operate, how they engage with customers. And the thing is, the infrastructure that supports all this stuff is becoming absolutely critical

[00:03:48] to how companies can actually execute on this promise, if you like, of AI and agentic. And it's the way we sort of frame it is, it's those things that were frustrations in days gone by. So those things like a little bit of technical debt where you didn't upgrade a bit of switching or that little bit of latency here or there. Those things you used to be able to get by with today, they fundamentally undermine those investments

[00:04:15] that customers are making in AI and how they can see the return on value from those. And so what we're driving towards is helping customers really figure out what those use cases need to be. It all needs to start with the use case, because once you understand what the use case is going to deliver, then you can start to figure out what that infrastructure needs to look like. And it's a whole range of things.

[00:04:39] You know, it's very fast networking to integrate and connect all of the things like the GPUs that these models run on. It's things like one of our platforms that we released back in the last year around unified edge, because we've got customers starting to look at the fact that actually, if they're, for example, retailers, they might want their customers to be using AI as part of their shopping experience,

[00:05:04] where it doesn't work when you need to haul all that data back or that inferencing back to a data center. So we need to be able to help customers put that capability closer to the users. So the infrastructure that's needed, it depends quite a lot on the use cases that our customers are driving towards. But I think that we actually had our AI summit in the US yesterday. So that's a Wednesday, the 3rd.

[00:05:30] And Chuck Robbins, our CEO, said that we're hitting a point of alignment now across the industry, across the technology world, where we're seeing companies move from where they were experimenting with ideas and use cases to where they're starting to scale into the execution of those use cases and really try and drive that business value from them. And that's where that foundational infrastructure with all of those different pieces is really going to come into its own in terms of enabling those organizations

[00:06:00] to help that actually work and get the value from it. Yeah, that's why, for many reasons, it feels such an exciting moment, that going from just the low-hanging fruit, the experimentation coming out of pilot, to scale, a massive scale right across the world there. And I did attend Cisco Live in San Diego last year. And one of the things that stood out to me was that Cisco was beginning to position itself as an essential to agentic AI adoption.

[00:06:27] And we are talking about not just adoption, but scale now. And very often we undertalk or underestimate just the power and the complexity of the technology that we're talking about and how it all fits together in the tech stack. So on that side of things, for people listening that are on this journey, can you tell me a little bit more about how the network, compute, and security layers, how they all have to work together for agents to collaborate reliably with humans and indeed each other at enterprise scale?

[00:06:58] Because it is incredibly complicated if you take a look behind those agentic AI headlines, isn't it? Yeah, look, it is really challenging. And I think that there's a couple of different aspects to it. One is the sheer pace and speed at which these things operate and also the innovation that's going alongside these things at the moment. If you think about the traditional way that people designed, for example, their IT environments and their operational environments, because the two often mirror each other,

[00:07:27] you tended to have quite disparate... You'd have a platforms organization who would have all the compute and that sort of stuff. You'd have a network team. You'd have a security team. And certainly in customers I've dealt with, quite often they don't like each other. That just doesn't work in the world that we see today. It struggles when we're talking about straight AI from a chatbot type LLM perspective. Once you introduce agentic, that becomes a whole different world

[00:07:56] because you start to introduce agentic workflows. You start to introduce that level of automation, which there may still be a bit of human oversight to it, but it's not very clear. You haven't got the humans hands-on, pressing the buttons, changing things. The agents are off doing this stuff. And so what it relies on is a much, much closer linking and joining and integration between those various different pieces of the IT infrastructure,

[00:08:23] the network, the computer security, all of those pieces. And there was another interesting... We had Jensen Huang from NVIDIA at the summit yesterday, and he was talking about how he sees a lot of this. And obviously this guy's GPUs are this guy's bread and butter. But he's talking about it as being the reinvention of that compute stack, that top to bottom app, compute, network security, and also a shift from explicit to implicit programming.

[00:08:53] So organizations moving from where you used to have people actually writing code to create something to a point where you're telling an agent what you want to achieve, the intent behind it, and the agent is off doing this stuff. And all of that relies on these pieces being massively integrated. And that's a big part of where Cisco has been driving, certainly over the last two or three years, in terms of bringing the product portfolios together, bringing the pieces together.

[00:09:22] Nobody comes to us anymore and goes, I want a network. They come to us and they go, right, I need a secure network to deliver AI, and I need to know exactly what's happening with it from an observability and assurance perspective. Can you do all of that for me? As opposed to, I need a few switches, right? It's a very, very different conversation that we're having with customers these days. And I'm glad you mentioned security there, because I've been to a lot of tech conference where there is a big focus on thousands of agents

[00:09:52] all being unleashed, all talking to each other. And the security element of it is either not mentioned or tagged on at the very end. And refreshingly, Cisco have introduced this idea of security actually being fused into the network rather than bolted on at the end as some kind of afterthought. So kudos to you there. But what does that shift mean in practice for organizations that are currently deploying AI workloads, especially when some of these workflows are increasingly autonomous? Yeah, I think it's really interesting.

[00:10:20] And I've spent a lot of time working in security through my career. So it's been an interesting shift to see happen. But fundamentally, this comes back to that speed thing and this autonomy question or point. When you have agents and workloads operating in a really autonomous way, you don't have the ability to move everything off to a choke point to enforce some security on it.

[00:10:45] The security has to be fused into the network infrastructure that these agents and these workloads are operating on. Because anything else just introduces delay, it introduces latency, it introduces bottlenecks and that sort of stuff. And the way that we've gone about that is to take some of our best practice security capabilities, we start to look at how we can embed them into the infrastructure itself. And that's done through hardware accelerators in things like switches.

[00:11:13] So we can now truly put security into the network device itself. We can do the same in the server, both Cisco servers or third-party servers as well, but also into the workload itself. And with technologies that are now available with things like eBPF, what it gives us is that ability to really fuse it into the environment. But that also leads to a different challenge, which is how do you drive consistent policy across all of those things?

[00:11:41] Because they all enforce, from a technical perspective, the security controls in a slightly different way. So we have to take a little bit of a step back and start to look at how we define policy a little bit differently and how we deliver a security platform that can then take that intent-based policy and translate it into the relevant technical controls for wherever an agent is, wherever a workload is, and deploy it that way.

[00:12:07] And that then gives you the resilience and the consistency in terms of security policy, regardless of whether a workflow changes something or moves a workload or an agent changes. The platform has the security within it so it can adapt on the fly. And we will have many IT leaders listening around the world that currently feel overwhelmed by a tool sprawl and operational complexity. And I would also add into the mix, hey, if we fast forward 12 months,

[00:12:35] you're going to be looking at agent sprawl as well. But rather than put a dampener on the excitement here, Cisco's talking about a problem advantage and operational simplicity all the way across the entire stack. How does that platform approach things like day-to-day operations and changes in that, especially for teams that are running AI-driven businesses? Because, again, there's a lot of exciting opportunities here, isn't there? Absolutely.

[00:13:02] It is a genuinely exciting time to be working in the industry at the moment. It's fascinating the pace of change at which this is all taking place. But Cisco, a couple of years back, we set out on a very, very clear path towards this platform model. And the intent behind this is that it has to be additive. It has to deliver additional value. So it's not just a case of, oh, well, you've got three or four things all with a Cisco badge on them.

[00:13:31] If you've got something which has a Cisco badge on it and you buy another piece of the platform, it's almost a case of, you know, one and one shouldn't equal two, one and one should equal three, because you should get some additional benefit over and above by taking pieces from the platform that work together. And that really helps when it comes to that operational sprawl or operational complexity and tool sprawl piece, because we've put a huge amount of work

[00:13:57] into delivering consolidated management platforms as well, you know, across different bits of the portfolio. So, for example, we have security cloud control now as a single point of policy and management for the security stack. We've brought together, you know, the old Catalyst and Meraki into Cisco networking so that we now have consistency from that perspective. We're driving similar capabilities from a data center networking perspective. And all of those help reduce the number of portals and things

[00:14:27] which our customers have to engage with. But then we've taken it a step further as well with something that we launched last year at Cisco Live, which we call AI Canvas, which is our agentic ops platform. And this is all about releasing the or leveraging the agents in a way that still has human oversight. So we're not just going, let them go. He'll be fine. Don't worry about it. There's still that degree of human oversight. But with the volume of data that we see now,

[00:14:56] we build all of that into the Cisco data platform and that's powered by Splunk, which was the acquisition that we made back in 2024. We've also got our own large language model that we've trained specifically on Cisco data to sort of CCIE level for any of those people who have been in the industry a while. They know the cachet that goes with the CCIE certification, these engineers who are right at the very top of their capability.

[00:15:24] And what that's done is it's given us an ability to very effectively manage and troubleshoot, secure up our platforms from a networking and a security and an AI perspective. And the thing about AI Canvas is when you look at it, it'll look blank. And it's got a little AI agent inside that you engage with. And what it does is it builds you context-relevant widgets as you go through whatever activity it is that you may want to do.

[00:15:53] It's also multiplayer, so we can bring in different stakeholders, different personas into the same view so that you can end up with collaboration. But all the stuff in the background, you know, say I need to troubleshoot that catalyst switch over there. There's an agent that spins up and off it goes. It does the troubleshooting for you. So we're making it much, much more simple by leveraging agents to reduce the workload on the IT ops teams, the sec ops teams, the network operations teams, and bring all of that together as part of the platform.

[00:16:23] Wow. Sounds incredibly cool. And I'm curious, just listening to you there, as AI agents continue to make decisions and take actions, I would imagine resilience will become a business issue rather than just a technical one. So how are you at Cisco thinking about digital resilience in this new agentic AI era? And what do you think leaders listening should be prioritizing to avoid fragile systems further on down the line? Yeah, digital resilience is absolutely foundational to everything that we do.

[00:16:52] You know, we have great focuses on things like workplaces and security and data centers and that sort of stuff, but all of it is underpinned by digital resilience. And I think that, you know, the world has moved from the point, especially with agentic and the way that, again, we come back to that speed point, right? It's the way that things are operating and adapting so quickly that we can't have a world where you had a business continuity plan and you'd test it once a year.

[00:17:19] And if something happened, then you'd pick some people up, you'd move them somewhere else, you'd give them some different tools. That world doesn't exist anymore if you look at the way businesses operate, especially when we're talking about AI and agentic. So what we're moving towards now is helping customers have that assurance, that visibility across the entire stack of what they're doing so that they can identify things in a much more proactive way. You know, they can identify that little niggle before it becomes an incident

[00:17:48] so that actually we're reducing the volume of incidents that customers are having to deal with because we can do things like identify the choke points, identify the bit of an unowned network, say, that's causing a problem and change some routing or something in order to avoid it. So what it does is it puts a much more proactive view into our customers and into their hands. And it's also about us understanding that customers, especially a lot of our enterprise customers,

[00:18:15] they have very large disparate states of all sorts of things. Data particularly, you know, people will have data in S3 buckets, they'll have some data in their own data center in databases, things like that. We can't turn around and go, well, we've got this great way of working, but you've got to put it all in Splunk. That doesn't work. So one of the things that, again, we launched last year was federated data analytics. So we can now go and speak to you, engage with Snowflake or engage with your S3 buckets

[00:18:43] and start to look at the analytics that we can drive off that data to give you a much broader view. And in an ideal world, we actually get to the point where the customer actually comes out of some sort of incident or challenge better prepared for the next time as well, so they don't face the same thing the next time around because we've enabled those learning mechanisms, we've given them the insight and the awareness of what's happened and what's going on,

[00:19:10] so they're better prepared the next time around. And we're recording this episode during the time where Cisco has got a huge focus on European audiences and we're also at a time where data sovereignty will remain a critical concern across Europe for some business leaders. So how are you at Cisco addressing customer demands for choice, control and autonomy over data while still enabling global connectivity and AI innovation?

[00:19:40] It's, again, quite a tricky balance, I would imagine. It is a tricky balance to strike. And Cisco is a US firm. And so we've always had that global challenge of different requirements in different areas. But you're right, that data residency, data sovereignty challenge is something that's really right at the forefront of a lot of our customers' minds at the moment. And it's been there for a while, but with the prevalence of cloud, the prevalence of cloud management,

[00:20:09] the prevalence of the hybrid world in which we're seeing customers, some stuff deployed on-prem, some stuff in cloud, it's becoming much more of a challenge for our customers. And so at Cisco, and this was driven actually within the European part of the business, we've driven a true sort of sovereign capability into a lot of our platforms and products. So that includes all sorts of stuff like enabling customers to define where they want their data to be resident. And we've got a whole range of places across Europe where they can do that.

[00:20:39] Air gap deployment capabilities. So, you know, some of our customers, obviously, are some of the slightly more sensitive ones. You don't want any sort of connectivity in certain scenarios to cloud or the internet. So we've created air gap deployment models where you can do things like offline updates and that sort of stuff, but also the ability to switch between those for the customer so that they can operate perhaps with their data resident in one country. They can have the cloud management that gives them that flexibility. But if something happens,

[00:21:09] then they can turn that off and they can run in an air gap operating model. Lots of focus on making sure that our technology is certified to some of the European regulations and some of the country-specific regulations within Europe, again, right? A lot of our technology has some quite US-centric certifications. Lots of focus has been put on making sure we have European-relevant certifications as well from that perspective. So quite a lot of pieces going on

[00:21:38] that all feed into giving customers that level of control back over what they're doing from a data and a sovereignty perspective. And another big focus for business leaders around the world right now is an increased importance of measurable impact and ROI from every tech project. And we've all seen and heard about examples of businesses caught in pilot purgatory or they're just struggling to find ROI from their tech projects. But I'd like to focus on the more positive side here.

[00:22:07] And you must get to hear a lot of stories from different customers around the world. So from what you're seeing with customers today, where are organizations actually getting the most immediate value from Cisco's platform as they adapt to more AI-driven ways of working? And where are expectations still ahead of reality? I'm sure you've seen both sides of this, but what's working best? Yeah, I think it's interesting. And I think that the sheer pace of innovation, which is that hallmark of this AI era that we're in,

[00:22:36] is the innovation pace. It's sort of outstripping where the use case pace is at at the moment. Look, we've got lots of customers doing some really cool stuff. And lots of them are on different stages in that journey. Some of them are still trying to figure out what some of those look like. Some of them, you know, things like fraud detection in financials, drug development in pharmaceuticals. Some of these are really well-defined use cases where we've got customers who have got

[00:23:03] significant deployments of AI infrastructure, networking, compute stacks, all of the software mid-layer and stuff that we work with partners around. And they're driving massive value from those. And on the flip side, we've got a bunch of customers in the middle who are still working out how they get those pilots to prove themselves. And we've got some great capabilities in that space as well. Things like our secure AI factory where your complete stack of capability,

[00:23:30] you're predefined with Cisco validated designs and reference architectures from the likes of NVIDIA and some of our other partners to give customers that sort of assurance that what they're going to go and get is going to work. It's going to deliver what they need from an infrastructure perspective. Because I think a lot of that frees them up to then focus on the other aspects that are critical to this. Things like the data, the governance, making sure that they understand what the use case needs to be.

[00:23:59] Because I think we've moved on from that, you know, oh, Copilot can summarize my emails from the week I was on holiday. You know, I think how much value there is for the business is perhaps debatable in that. But there's some fantastic use cases in a lot of our customers and retailers that are starting to look at what they can do in terms of that customer experience. Experience is a really big thing at the moment, actually, across lots of different ways that we work with customers, whether it be enabling them from that AI perspective in terms of a retailer, for example,

[00:24:29] doing something for their customers all the way through to some of the stuff we do, for example, in our future-proof workplaces platform aspects where it's about devices. It's about giving customers the experience of being in the room, even if they're virtual, and some of those sorts of things. And we've had quite a strategic tie-up, for example, with NVIDIA for a long time within our collaboration portfolio to help drive and help deliver some of those use cases to our customers from that perspective as well.

[00:24:57] So there's always that bit of, this is the futures type stuff and things like AI Canvas and things like that. There's a lot of development still to go in that. We've got customers starting to work with it now, but the future is really exciting. It's a really exciting time to be in this space. And I think there's some massive things that are coming down the line as well. Fantastic. And a lot of people will be listening to this episode, our conversation during Cisco Live EMEA, which is in Amsterdam.

[00:25:27] And there'll be a lot of press releases coming out, a lot of noise. It'll be very difficult to keep up to speed without the scale of announcements that will be coming out of that. But if enterprises or business leaders listening today, if they were to take away one lesson from that event, what is that message that you're delivering at the event? What should it be about preparing their infrastructure, security, and teams for an AI-powered future that is moving faster than most roadmaps anticipated? Because there's a lot of talk over the last few years

[00:25:56] around the speed of technological change, but there's also a waking realization that it might never move this slow again. That pace is ramping up. So what's that key message that you're delivering this year? The key message is that the infrastructure, and Cisco is a big part of this, is absolutely essential to enabling the customers to adopt that AI and agentic capability they're looking at. They need to refine the use cases. They've got to start with the use cases, because otherwise what we see is that

[00:26:25] these pilots just fall by the wayside, because there is an improvement benefit from it. So refine the use cases, and then talk to us about what that infrastructure needs to look like in order to support that use case and really deliver the value for your organization. Because we're in the great position at Cisco, right? We can help you do almost anything when it comes to what you want to do from an AI perspective in terms of the infrastructure and the capability. We just need to figure out what that's going to be for any given customer so they really get value from it. Awesome.

[00:26:54] Problem first, use case first, then the technology. That's my kind of language. Love that. And as I said, there will be a lot of announcements coming out of the Cisco Live event. So anyone listening wanting to check out some of that, there'll be a lot of videos, keynotes, et cetera, and press releases. Where should they go? Cisco.com is the main sort of website. At Cisco is the handle across all of our social media. You'll see plenty on LinkedIn, on Twitter or X, on YouTube. And then if anybody wants to contact me, then they can do via LinkedIn as well. It's just Rob Lay, all one word. Awesome.

[00:27:23] Well, I will add links to everything you mentioned there. So for everybody listening, you want to find out more information, either check out the show notes. There'll be a useful link section there and I'll put all the links and equally go over to my website, techtalksnetwork.com and there'll be a blog post and links and videos and things there. But more than anything, Rob, just thank you for coming on here today and bringing this topic to life. Everybody's talking about it right now, but talking about it in a language that everyone can understand and solving real business problems, I think that is the magic here. So thanks again.

[00:27:53] Thanks very much, Jitai. So as AI agents become more autonomous and deeply embedded into business operations, the conversation is now shifting away from experimentation and we're heading towards trust, resilience and measurable outcomes. Refreshing to hear, isn't it? And this discussion today, I think highlighted a reality that many leaders are now facing. Yeah, AI doesn't fail because models are weak. It fails when the underlying infrastructure

[00:28:23] cannot keep up with the speed, with the scale or complexity because networks, compute, observability and security are all rising to the level of business priorities rather than just being seen as the background plumbing that IT worry about. So a big thank you to my guests and indeed Cisco for setting up this interview and allowing me to share how they're thinking about platform design, digital resilience and data sovereignty, especially at a time

[00:28:51] where organisations are preparing for an agentic AI-driven future. And everything we covered today, they're not theoretical challenges. They are decisions being made right now in boardrooms and IT teams around the world. So as this adoption accelerates and agents move closer to the core of how we get things done, will your system bend under the pressure or is it built to hold? Over to you. www.techtalksnetwork.com Let me know your thoughts. If you're attending Cisco Live in Amsterdam,

[00:29:21] please let me know everything that you saw and everything that you heard, anything that inspired you or got you thinking differently. I'd love to hear from you. So pop over to my website and let me know your thoughts there. Other than that, I will return again tomorrow. I'll be waiting in your podcast feed. Same time, same place. Speak with you then. Bye for now. Thank you.