Cisco: The AI Chatbot Era Is Dead – Here's What Comes Next

At Cisco Live 2026, Cisco pulled back the curtain on a suite of AI-native tools designed to help organizations orchestrate, secure, and manage a blended workforce of human and AI agents.

To get the inside track, CX Today’s Rhys Fisher sat down with Vinod Muthukrishnan, VP and GM of Webex CX at Cisco, for a wide-ranging conversation on what the shift to AI-native CX actually demands in practice.

Muthukrishnan is clear that the industry has moved past the experimentation phase, and that the bar for what comes next is significantly higher than most vendors are letting on:

“It’s never been easier to build an AI agent. It’s, however, never been harder to make it enterprise grade. And that’s the dichotomy we’ve got to deal with.”

The conversation covers a lot of ground. Muthukrishnan pushes back on the industry’s fixation with deflection and containment rates, arguing that no single metric can capture the full customer journey.

He walks through Cisco’s AI Agent 360, a framework combining custom guardrailing, runtime monitoring, and Splunk-powered observability, and makes the case that security and observability are foundational requirements for any serious agentic deployment, not optional extras bolted on afterward.

He also introduces the concept of “one experience”: the idea that regardless of channel, timing, or how many agents were involved, every customer interaction should feel like a continuation of the last.

Looking further ahead, Muthukrishnan raises a question that will resonate with anyone thinking seriously about where this technology leads. If agentic AI can cut across every organizational silo, does the enterprise as we know it still make sense?

Transcript

0:08

Hello and welcome to TX Today. I'm Reese Fisher, Associate Editor, and today I'm delighted to be joined by Vinod Mutokrishnan, the VP and GM of WebEx TX at Cisco. Vinod, thank you for joining me today. How are you doing? Please, I'm doing great and thanks for having us. No, absolutely. I'm uh I'm really looking forward to the chat. I know it's uh an exciting time for you guys at Cisco. We've got Cisco live kind of just around the corner. Now we've got some big announcements drop in. I guess before we get into the really meat of that, what props are you most excited about and why does this feel, I guess, like a significant moment for the contact center space? Well, thanks for starting us off with that. I think the conversation around AI has been there long before the GPT moment, but I think the GPT moment, as we call it, is when it became mainstream, where everyone sort of experienced AI. And us individuals, enterprises have experimented with AI, AI agents in different ways. I think the reason I'm most excited about this is we've reached a point where the experiments have reached maturity, and we have a definitive view on what it takes to make AI agents, to make them safely, scalably, securably. And if you have to truly enterprise grade your agent tech systems, what are all the systems and measures you need embedded in that experience to make it happen? And a lot of our announcements you'll see are really towards that. It's not just another AI agent, but all the things you need to do to run a truly agent tech system safely, scalably, securely. And that's what I'm really most excited about. It's sort of a coming-of-age moment for agent tech, the term agent itself. Yeah, I like I like that phrase in the coming of age idea of it. I think I think that that kind of shines through in what we've seen. Obviously, I guess a big part of this release is this idea of moving past the AI chatbot phase into what you guys are calling the AI native contact center phase. Could you just tell me a little bit more about that, please? No, again, thanks for that. Look, again, the experiments rightly started with point problems, what AI could take off, like deflecting a certain call type. As I've said it fairly publicly before, terms like deflection and containment, I think, are uh counterintuitive to what you're trying to do, because no customer wants to be deflected or contained. You want to actually offer always on highly preemptive, highly empathetic, human-like service, irrespective of whether the human's there or not, because it's like three in the morning, um, or because there's like 2x queue up on your call cues. So for us, really, agent take an AI-native CX stack looks at a concept we have evangelized called connected intelligence, which is humans need to work with humans. You and I are here on this call because this is the highest value conversation. We all to communicate, collaborate, talk about something. But increasingly, humans are working with AI. And most importantly and scalably, AI is going to work with AI to do things that unassisted autonomously over a long sort of uh task execution window. How do you account for this new world where all of these three modalities need to happen on one platform? This means you need to rewire the foundational architecture of your contacts and a stack, or I dare say customer engagement stack. And that for us is what an AI-native CX stack looks like. If you just throw a lot of AI on top of the existing contacts and a stacks or existing workforce stacks, you'll get what I call the.ai approach to building things. So as you'll see as we go through a lot of our announcements, we've foundationally reimagined what a CX stack would look like in the age of AI. We've assumed a world where the vision of connected intelligence is a reality and basically rebuilt from the ground up all the systems, all the architecture, all the products. And that is what really, really excites me about this AI Native Contact Center phase we are in now. Yeah, yeah, I bet. I was wondering, you you mentioned the kind of deflection and containment, perhaps not being the metrics you guys are looking at with this AI Native Contact Center. Are there any in particular that you are prioritizing all the uh significant wins that you're seeing with it then? So one of the things we've always said is that look, customer experience is not a point in time. No one single metric captures CX because CX is a continuum, it's a journey. You're going about your journey. Let's say, for example, you went to the shopping mall, then you're looking around stores, you discover something, you decide to buy it, size is not available, you go to e-com, you purchase, it gets delivered. You were told it get delivered in two days, it takes six, you know, and then it shows up, it's the wrong size. Now you ask them for a refund. That's a journey. You didn't sign up for any of these steps, right? We impose those steps on you because we're trying to sell you the trousers or the shirt or what have you. So no one metric at no one point in time, when seen in isolation, gives you a holistic view of customer experience. So the reason I, even if you take the term containment without judgment on what it sounds like, but in isolation is not good enough because did you get the resolution you seek? Did you get it in the shortest amount of time you sought? Was your CSAT at worst neutral to what it would have been if I had been personally speaking with you, or ideally better because the resolution was faster, what have you. When you look at these together, now you are cognizant of the journey and the outcomes the customer seeks. Right? So again, we always talk about what are the deflection rates. What if 100% of the calls that were deflected ended with the words, man, this was a terrible experience, or I don't want to talk to you again, or next time just get me to a human, you know? So that's why we always say that we should not get metric obsessed. Metrics are the best way to track, measure, and improve any process. But metrics in isolation can give us a very, very lopsided view of actual customer experience. And we've said this, and I know it sounds very counterintuitive. Like the purpose of AI and automation is actually to make customer experience more human. And humanness is around being, if if you view your best friends, you expect me to be always available to you, know your history and context. Um, might not always tell you what you want to hear, but tell you with clarity, I should know if you're a morning or a night person, are you a texter, are you a phone person? That's the humanization of customer experience. And AI and automation can help us do that at scale for every single customer. And hence the whole standalone metric uh business, I find it very dangerous, if you ask me. Yeah, yeah, I really agree. I think that's uh that's a really nice way to look at metrics and uh a quite refreshing take from a vendor like yourself. So yeah, thank you for that. Um I wanted to talk about what is you know naturally kind of one of the hottest conversations in in AICX space right now, and that's this idea of trust, I guess. In particular, trust in letting these AI agents handle real customer interactions. How does Cisco think about kind of maintaining that human element? And what do you think a well-orchestrated blend of human and AI actually looks like in practice? So we've said this. There's two things we've said. Again, these are not slogans, these are we believe in these. You know, the first is it's never been easier to build an AI agent. You can build an AI agent in the next five minutes. You could do it by running an agent on the side, asking it to build an agent, right? So it's never been easier to build an AI agent. It's however never been harder to make it enterprise grey. And that's the dichotomy we've got to deal with. These tools are available to us, they're available to everybody. And so everyone's pinning up these agents. And it's easy to demo, it's easy to stand up, it looks slick. Um, it's easy. If I did a startup, I might the first day I'd probably build the agent-building interface, right? Make it as slick as you could. Now, what we're seeing around trust is the journey from building a technology demonstrator AI agent to making it enterprise grid is the real journey. And we said this somewhere else and and I stand by it. I almost shudder at the thought of anyone being a CX AI applications company, the front-end company, without being a full-time security and observability company. Is, and I'll tell you why I say that with such vehemence. When you build an AI agent, we need to first figure out what do we need to do to protect the world from these agents. So there may be other bad actor agents coming in, you connect to an MCP server, what's on the other side of it? You're interfacing with a third-party agent, have you validated the identity of that agent, right? So you've got that whole paradigm to think about. You're then thinking about protecting your world from agents. So, you know, you you you you look at um um are we red teaming this? What what what's gonna happen when someone tries to front inject into your agents, when someone looks at a multi-turn conversation and says, okay, I know I can keep going at it, and the 17th turn of the conversation, I know I will get through. The threat surface with AI is so vast that when you think about what do you need to protect your world or your agents from the world, it's a massive escape. And it's easy to build an agent, very, very hard to figure out how to do all of those. But the third is how do you observe your infrastructure, data, agents, network at machine scale? And if something happens, how do you respond at machine speed? And to bring these three attributes into any agentic system is critical. Now you can imagine you and I can build an AI agent in the next five minutes. We don't have the skills, competence, expertise, or the investment into creating these structures. And now we'll have to cobble together seven products to observe these agents, to monitor these agents, to red team and test these agents pre-production, measure them in runtime. So for us, bringing together the attributes we have with AI defense, where you look at a very sophisticated custom guardrailing engine, which allows you to look at it, take your privacy security policies, and actually write custom guardrails for multi-turn, highly context-a-wired conversations to vertically integrate our Splunk portfolio to give observability and AI agent performance at scale, and then to have them all packaged together with the AI agent builder, which is what we're calling the AI agent 360, I think is critical. So the the thing I would tell any practitioner is if you're building an AI agent that is going to access your sensitive data, act autonomously, and talk to your customers, if you thread that together in a sentence, you can imagine the immense opportunity and the immense risk. Either do all of these things around security, observability, and what have you, or don't. But there is no half measures to building these systems. And that is why AI Agent 360 is probably one of our most ambitious uh developments around this front and something we're genuinely excited about. Yeah, yeah, I can imagine. I think that's some some great advice. And I guess on that uh topic of providing guidance, really, I'm wondering if, you know, there's a CX or contact center leader perhaps watching this right now. What is it about the AI Native Contact Center or the main thing that you think that's really going to change things for them day to day? So one of the things we're also launching at Cisco Live is what we call the agentic context engine. Um we're moving from a place where I've got an agent that does appointment rescheduling, one agent that does prescription gifts, one agent that does, you know, check if what's my position in Q, what have you. And it's great. That's exactly how AI experiments should start. Small, contained, yeah, ROI measure what we do. As you start scaling it, the real vision is to offer one experience to the customer, which is no matter when I messaged you or called you, what the context was, what the modality was, um, how async that conversation was, how it doesn't matter when I last contacted you, much like calling your favorite human, you pick up and feel like it's it's like we never stop talking. And that context preservation is a very important part of, as I said, humanizing the customer conversation, uh, keeping it context aware, and picking up where we left off, you know, to say that. And for us, that's an area of massive investment for us, the context engine. And that will allow us to truly deliver upon the promise of what we call one experience, which is for the customer, it feels like one continuous experience, no matter which touch point they reached out to, when last they pinged you, what the context switch was, it feels like you're speaking to the same person. Um, and you'll see us really go deep into that concept with our AI concierge, which is sort of the one front-end agent to handle all customer conversation. But the real secret source is not the one agent to handle the conversation. It's the memory and context layer that makes this agent smarter. Uh so that's again something else that I'm genuinely excited about. Yeah, yeah, absolutely. I was wondering, is the obviously by the time by the time we release this, Cisco Live will will be, will have happened or will be in the middle of happening. I was just wondering, is the what do you maybe predict will be the feature or the capability that will perhaps generate the most buzz or have people most excited if you uh if you were to dust off the crystal ball and have a guess? Uh can I pick two? Yeah, absolutely. Okay. All right. Um, so I'll come to AI Agent 360 in a second because we've spoken about it. Um we are obviously announcing full suite uh workforce experience management. And the reason why I'm excited about that is one of the great benefits of coming late to the party is you get to you get to see all the trends play out and then you do the right thing once. The foundational shift that has happened is that AI went from being a toy uh to a tool to a coworker. And all of this has happened in the last two years, right? Three maybe. And so when you look at that fast evolution and you look at workforce management from a quality management perspective, from a scheduling perspective, from a forecasting perspective, training, onboarding perspective, if you had to build workforce afresh today, it would not look like the workforce from 10 years ago. Because you've got to account for this new coworker whose constraints are different. Capacity is theoretically not a constraint, budget, however, is, right? Um, they can take all call types, but there are call types you're not sent to AI because of ethical or statutory or other reasons. There's so much nuance between this new digital worker and your human worker that the system that you use to schedule, forecast, manage, quality analyze, train, onboard, optimize, and improve look dramatically different. So our Vem suite is built from a ground up accounting for that connected intelligence world where AI and humans are equal participants in the conversation, and built for a world where the counterparty calling in might also not be human, it might be AI. So that's I'm really excited about. Workforce might seem like in 2026 a sort of legacy topic, but this is a workforce built for a workforce of AI and human and for a customer base that is AI and human. So that's the one thing I'm generally excited about, the entire workforce suite. Um, some of them, like quality and uh management, are already generally available, some are in beta and in the pipeline. So that's one. The next, I'll come back to AI agent 360 because what we've discovered is with AI agents, either you find the right way to build them, test before deployment, monitor and run time, and observe and improve and optimize. You either do all of these things in one single place or you don't build AI agents. So for me, the reason I'm saying that is we see the immense opportunity with AI. We also see the immense threats with AI. And our job as a security and observability company, along with being a customer experience company, is to bring these capabilities together. So those are the two things I think I'm genuinely very, very excited about. I hope they create the same buzz as uh as I'm excited about, but we'll find out in a week's time. Yeah, yeah. We'll uh we'll check back in and see how accurate those predictions were. But no, yeah, it sounds like really, really exciting stuff. I guess just as the final question here, I was wondering, you know, if if this is the beginning of this AI native era. I was wondering where where it goes from here, I guess. You know, what's the perhaps the version of the contact center that you're building towards, say, two, three, four years down the line? Look, um, I think in AI, the the biggest mistake we can make is crystal ball because the world is changing so fast. You know, I was telling my teams, usually I do once a year State of the Union, which is this is where the world is, this is where it's headed. Once a year, you need to kind of come back and discuss. And I said now, if we do it once a quarter, we might be slow. So the the rate of change is unprecedented in human history. So it's hard to crystal ball. But I think a few trends are secular. Uh, the few trends that are secular are we have to use AI and automation to vastly humanize the customer experience across every touch point. You know, the lack of humanization is not about sounding nice or sounding human, it's about having the right context. Which is if you called me tomorrow morning and somebody else at the same desk picks up the phone and says, Reese, I have no idea what you're talking about, or I don't know, let me check. Vinod's not at his desk, he's gonna come back tomorrow. That's for me to just pick up and say, Hey Reese, I know you don't know me, but yeah, I know why you're calling. I know what you need. So the humanization is really using the power of AI to offer the kind of superlative, highly personalized customer experience at scale across all touch points is the time has come. We have the technology and the tools to deliver that. The second big thing is we have to account for a customer experience realm where the counterparty calling in and call is a misnomer, is not human because all of us will have agent, AI agent personas of ours, which means the foundational compact, which was how to offer this human the best possible experience using our AI or human is changing to a world where you still want the job done, but it's not you who's interacting. How do you account for a world where the counterparty changes? So you'll have to soon realize that how we build our interfaces, how we measure success, how do we measure seaside, how do we measure outcomes will change. Outcomes maybe won't change because you're calling for the same thing, but almost everything else we measure along the way will foundationally change. And that's why when I spoke about WEM, I said we need to architect for a world where AI and human are equal participants in the contact center, also outside, because they will come in. And the last big question really is, and and I say this uh mildly facetiously, um if you were to build a truly agentic enterprise today, you see startups that are three person startups that are uh going to a billion dollars. If you were to recreate an enterprise today, would any of your teams and structures and processes and software tools look the same? The short answer is no. If you and I were to start a startup tomorrow morning, we wouldn't think of teams, departments, and this and that the same way we would have even five years ago. So if everything is up for discussion and debate, what is the role of this customer engagement agent? Why should they only do support? Sales, support, marketing, a query, uh a conversational engagement, they're all part of the same continuum between brand and customer. So we'll see this dramatic coming together of all of these because agents aren't constrained by what you and I are, which is I have a boss, you know, my job is support, you have a boss, your job is marketing, or you're trained only on marketing stuff, you're sitting in the store, so you cannot process uh, let's say, uh cashbacks. Uh, maybe you can, but someone else can't. So that's my point. All these silos were the shipping of our org chart. The silos may still remain, but the agentic layer on top is able to access all of these silos, preserve the context, and give one interface to the customer. And if that is possible, will the enterprise even look like what it looks like today? That's my big question. I'm not gonna hazard, I guess, on what the answer is, but know that we are building towards that answer. Perfect. I think, yeah, I think that's a great, uh, great place to end things. Thanks, Fino. That's been a really, yeah, really, really great chat. It sounds like very exciting times at Cisco right now. Uh, like I said, I know I know the event will be live by the time this comes out, but uh, all the best with it. Um, sounds like it's gonna be a really, a really great event. No, absolutely. Our aim is to be the world's largest startup. Uh so be the scale and and reliable, like sort of dependable company that we are that comes with scale and tenure and 40 years of being in business, and yet act and deliver at the speed of a startup. So marrying the two is what generally excites me. Yeah, yeah, absolutely. I would also like to just quickly thank our audience as well for tuning in. If you enjoyed this discussion, and I'm sure you did, please remember to like and subscribe to the channel and head on over to cftoday.com for more stories like these. Until next time, thanks for watching.