Submit Podcast or Channel

Agentic AI for Business Leaders: Trust, Guardrails and Rollout Lessons from Cisco’s Joe Vaccaro

October 13, 20260 comments♥ 0
00:00 On today's episode of right about now, we talk all things AI. Joe Vicaro is the SVP of network platform and assurance at Cisco. At Cisco, yeah, always changing the game with networking and really all the behind the scenes of keeping infrastructure solid in America for making the networks work. But look, let's talk AI. Today's episode, Joe and I talk about practical advice for how to leverage agentic AI. Agents doing work for you efficiently. key ideas for making that work well and safe for your company and some practical wisdom for how to deploy this for your company. Cisco is a leader in this.
00:36 AI is growing. We're going to keep talking AI on this show because you need to use it. You need to know it. You need to have it right now. Business leaders don't outsource this to other individuals. Use it yourself. Same thing of you. Go build on a weekend project.
00:50 When you see the power and you see that unlock, you see the way that you feel when you have that level of agency to go and get things done. You can then understand how to apply that across your business, not only terms of your employee productivity, but also how do you transform the business processes? How do you transform the way that your operations teams manage your business? You don't win by following the playbook. You win by rewriting it. 700 episodes deep with the people who actually built something real. No theory, no fluff, no shortcuts. This is right about now. Now with Ryan Alford. What's up, Joe? Welcome to right about now.
01:35 Hey, thanks for having me. Joe, good to be with you. Cisco, we've had a lot of Cisco executives. It's good to have another one. Always enjoy. I like opening up. Like, we talk pre-episode. Cisco's one of these companies in America that makes things happen.
01:49 But everyone kind of goes, what are they? What do they? I know it's like computers, right? And I'm like, yeah, it is. But I always like opening up the vortex, as we called it, Joe. I know you've been there a while. you're doing some big things. People can look up your history.
02:01 But I just briefly touch on the core segments you've been touching at Cisco, your core specialties. And let's start there. And then we'll dig into some specifics. At the end of the day, I'm a product guy at heart. I love solving customer problems. And that's what fuels me. And for my career, that's kind of what led me to a company called Thousandize back in
02:21 2019. They were a series D funded startup that I joined as the VP of product. And then about a year and a half later, we were acquired by Cisco. At Thousandize, it was really to solve core problem that we saw customers facing. At the end of the day, people think about networks as, hey, that's an important part of my business. But the reality is that the business doesn't operate without a network. But the network, you know, what connects a user to an application, what connects Ryan, you and I talking right now is incredibly complex system. Right.
02:49 Think of it like orchestra or a supply chain that connects your device there in South Carolina through the internet to me here in the Bay Area. And that network is getting more complex as we've seen these different fundamental shifts over time. We saw the rise of the internet replacing traditional networks. We saw the rise of cloud replacing traditional, say, data centers. We saw the rise of SaaS applications replacing traditional on-premises application. We saw the rise of work from home that were placed maybe in office-based work. We're also seeing the rise of AI. And all those major trends have made the network more critical than ever before,
03:26 but also made it more complex than ever before. With my role at Cisco, ultimately, how do we enable our customers to see, understand, and improve every connected experience that they rely upon? It's funny as networks and internet has become ubiquitous and everywhere, we take it for granted. Take a lot of things for granted.
03:43 Like the power being on, the grid, internet being there, get frustrated when it's not there. It's made to feel simple, which is why great companies like Cisco have made it to feel that way at the consumer level. But the reality is, like you said, that orchestration that's happened,
03:56 is a lot more complex. And then we throw in the era of AI. That's a whole other ball of wax. You've seen a lot. You're done a lot. Put into perspective for me what's happening with AI. Let's start in the macro. Then we'll get micro.
04:11 Is this bigger than the internet? Is it bigger than I wouldn't say power? Before we had electricity and after, it almost feels that big a scale to me, but sometimes I'm living in a bubble. Talk to me, Joe, about perspective of what AI is on a broad scale. When you look back and if you, You study history and you think about the great industrial revolutions, whether it was factories
04:31 and others. And you could think of the internet as part of that industrial revolution. It changed the economy from every single way for not only how do we do business, but how business gets done. And now if you think about this next big wave in AI, it's feeling like it's having the same impacts on business, the same wave if you go back to the early 2000s in the rise of that internet economy. Now we're in this AI era.
04:54 And it's impacting businesses in variety of different ways. It's one, as we see AI being infused across the business, it's changing about how business gets done. But then also it becomes this massive productivity change when we think about how do we empower our employees to leverage AI to be able to help them get their work done faster, but also help them to be able to focus their time where their own judgment can be best applied. You know, where their craft and where their judgment is maybe the lack. the time to be able to go and focus on those core problems because they're constantly, you know, chasing the kind of the reactive nature of firefighting in their business. When I talk to customers and businesses that have really embraced it and I think the ones that are on the leading edge, they see AI not as just a way of getting more productivity out
05:42 of their employees, but think about as a way as unlock for their employees. So they spend their precious time solving the hardest problems and where judgment and craft really matters. It's interesting because I think it was. first, oh, it's going to take all the jobs. Now it's educating people. Okay, it's safe to use. It's like a calculator, right? That's what I tell people now.
06:08 It's like not using it. It's like sitting next to someone that's got a calculator and you're not using one when you're not in grade school and it's considered cheating at least. I miss my Ti-83 graphic calculator, man. That thing was amazing. Oh, you can play. Remember the games you could play on that thing? Oh, yeah.
06:26 Playing Snake. When I'm supposed to be paying attention in class. It takes you back. I'm glad I wasn't the only one. But in all seriousness, is that still a fear, like at the corporate level? It's taking all the jobs versus it's the calculator. I don't really see that as a fear anymore. But I think maybe one of the points that I continue to be reminded about is that the fear is just, I think, in terms of the pace of how this wave has hit us.
06:52 If you go back to, again, those different great industrial revolutions, they happened over many different years. If you think about the Internet, it was. part of a decade for that wave to really wash over the business. Think about AI. It was only a couple years ago that chat GPT came out in the very beginning, right? We had our first use of chat GPT and the prompts that people put in. It was like, hey, how many R's are there in strawberry? And now if you think about we're in 2026 and the advanced capabilities, in many ways it's not fear, but it's like, how do you keep up with it? How do you harness these capabilities? Knowing that in business, your competitors are also using these capabilities to create a strategic advantage. And so for business,
07:32 it's how you embrace it. How do you unlock it? And how do you do it in a way that also brings your employees along for that riot? And all of this has happened over a short period of time. And I think that's where it creates a level of anxiety that business leaders are having to kind of delve with to say, okay, how do I apply this smartly across my organization? Knowing that, you know, something that you could have taken months or years to think about the strategy now needs to be thought of and executed in days or weeks. Yeah, that's the big shift is, and I think the winners and losers are going to be the companies that embrace that. You can't let it be without measure and without safety and brand standards and all that good stuff that matters, but the reality is we have tools that enable
08:13 speed to market. The companies embrace that are the winners of the day in some ways if there's such thing as winners and losers, but I think reality is it is the bottom line. How is Cisco embracing that? Because I worked on the ad agency side. working with corporate businesses, Apple, Samsung, Verizon, Audi, Lexus, Toyota. Some of the biggest brands in the world, they move slow as shit, man. I mean, I'm just being honest. This is how it used to be. It's been a few years.
08:38 How is Cisco and how are you seeing companies like embrace this when traditionally corporate America is timid for a reason? But now, how are we embracing this? If you think about it, you know, Cisco, we're building the critical infrastructure for the AI era. How do we do that for our customers? Well, it starts by doing it ourselves. A Cisco IT perspective, we've really been, and I give a lot of credit to our team, has been the cutting edge of unlocking these capabilities for our employees to use themselves. Because when you use it yourselves, you understand the power of it.
09:07 And then you can think about how do I apply this back to solving those complex customer problems? You know, we've also rolled out AI agents for all 90,000 employees within Cisco. That really helps them to see the power of it and then to live on that edge of creativity for how to then apply it back to the problem. Now, one of the points you made also, I think, is around, you know, hey, guard rails. And I think that kind of hits on a key point around trust. How do you build trust in the systems? Because at the end of the day, I also remind myself that Cisco isn't in the business of creating networks.
09:38 We're in the business of keeping a hospital operational so that the doctors can serve their patients. They're enabling the grocery store checkout person at a retail to be able to allow customers to buy milk. That's where networks live. And so when you think about then, how do you apply and use AI for? network operations. It's like you trust the agents to help you perform those functions. And building that trust is absolutely critical so that you can begin to allow the agents to be able
10:05 to operate alongside your human teammates. How do you guys balance? You have Altman coming out saying practically, AI will kill us all. I'm not saying he said, but implying that it can be dangerous. And you have headlines that read that way. And then you have the opportunity and the innovation that's happening, how does a company like Cisco apply those guardrails of reality and safety versus balancing what's in the news headlines? The keys are the news can shape your perspective. And again, the way to build an understanding is to use the products and to live at the edge. So if you go back to say Anthropic when they were coming out with Mythos, Cisco is one of the leading partners for Anthropic. We were there from the very beginning in Project Glasswing in leveraging these
10:50 models to understand how security threats could be exploited across a business. We used the mythos internally across our own infrastructure. We used it to improve our own products. And then based upon those learnings, help to understand how we could then apply these methods in terms of hardening our own products and enabling our customers to be able to protect themselves. For me, it's easy to look at the headlines and to form opinions. At the end of the day, you need to use the technology, ensure that as business, you have people that can live on the edge to understand how this is going to impact your business itself. That's the only way to form your own opinions and to be able to use it because the technology is changing. Obviously, they're hot takes on either side
11:31 of this kind of macro debate in the industry right now. And that's going to be important for business to be able to form their own opinions. Joe, have you seen anything that's ever scared you with AI? Oh, you're not setting up nuclear weapon. I guess I mean like, wow, went further than I told it to and it didn't have authorization to do that. We've seen in terms of building AI capabilities where if you give an agent a goal, it will use every possible means to try to achieve that goal. And that's where you think about how do you build software technology in running through evals and guardrails and testing it. But yeah, absolutely. If you give agents goals, they can be able to work. You know, there was a quote, I don't know who it's attributed to which talked about, you should think
12:11 about agents as teenagers. They're extremely confident and eager, but at times, lack good judgment. And so when you're bringing agents into your system, I think that's where you need to build up the trust. You need to provide a way to govern the actions so that if they are taking steps alongside your operations teams and making changes to critical infrastructure, not nuclear codes by any means, but helping to fix problems across your network, they're doing so with a level of determinism and not necessarily kind of making it up on the spot. So that kind of framing that analogy kind of helped me kind of shape how we think about agents and also how customers should think about how do you bring agents into your business processes safely?
12:59 This episode is brought to you by high level. Make sure to check out the link in the show notes to sign up for your 14 day free trial. If you run a business, you know leads rarely arrive when it's convenient. A call comes in while your team is helping another customer. If someone reaches out, but the follow-up gets delayed, an invoice sits unsent or a promising lead gets stuck in the wrong stage of your pipeline. I've built enough businesses to know that growth has a way of exposing every gap in your systems. That's why high levels voice AI and managed AI agents caught my attention. High levels voice AI can answer incoming calls, qualify leads using information from your website or uploaded documents, and even book or reschedule appointments. When your team is busy,
13:41 voice AI can keep that conversation moving instead of letting a potential customer slip away. Then high levels managed AI agents can handle multi-step actions across your CRM. They can update pipeline stages, send invoices, and execute follow-up sequences across SMS, email, and social. That means fewer repetitive tasks for your team and a more consistent process for every lead coming into the business. For me, the real value of AI isn't adding another complicated tool. It's using technology to handle the work that slows your team down so your people can focus on building relationships, serving customers, and growing the business. If you're ready to create better systems and get more from every opportunity, use the link in the show notes to
14:21 sign up for your 14-day free trial. Does belittling your LLM make it work harder or less hard? I've built a lot of software with Claude and others the last couple years. And man, it is so smart. But to be so smart is dumb as hell sometimes. And it's like what you're saying with the judgment. And I'll be honest, I do not do this in my real life. And it doesn't really make me feel better.
14:44 But I get so frustrated and I know it's a robot and it doesn't really have feelings. And I will talk shit to that, that LLM so hard. I become the biggest shit talker ever. And I sometimes think it actually makes him work better. But am I fooling myself? Feel free to move on to the next question that's more serious. But in all seriousness, I do put down my LLM quite a bit. Think about it like hiring a new engineer or a new college graduate.
15:08 They don't work well with nuance. You need to give them clear, direct feedback. They work better if you yell at them or not. I don't know how to go and run that Eval system. But I do know at times what they work best with is giving them, clear, succinct feedback. When you leave ambiguity in your answers, they can interpret that and run a mess. And so what you're hitting at is when you're working with Claude or you're working with codex, as direct as possible in terms of what you're after, what is your objective, will give
15:37 the agent a better chance of achieving it. The same way, if you had a new college grads straight out of school, if you give them a vague question or vague direction, you're probably not going to get exactly what you want back from them. I told Fable 5.1, it was getting shipped off. to the land of Misfit toys. And I couldn't wait one more second for Fable 5.2 to come replace it. And they look there. He shipped 5.5. They heard your feedback and skipped right to 5.5.
16:00 It didn't seem to like that very much, but it didn't get any better. And look, you nailed it, Joe. In all seriousness, it is the vagaries. Sometimes you get in a hurry. It knows what I mean here. And you pay for it every time with cleaning up the mess that your toddler left behind. It's like the world's greatest, smartest toddler. Oh, my God, it's a genius.
16:19 But, man, it's an idiot. times. Because it's like, it's exactly what you said. The same point, what I see is that the agents are getting incredibly intelligent. The way that intelligence is formed is not only through memory, but the context that you provide it. And if you think about that shared context layer, especially tying it back to a little bit from a Cisco, right, when you bring agents as teammates across your critical infrastructure,
16:43 you want them to have the context of what's happening. So if they do propose changes, I understand the impacts of that. And so, you know, a lot of what we think about. is how do we enable agents to have the broadest and deepest understanding of context to what's happening across the networks and how other domains, whether that's your security policy or others that are impacting it so that when they do make recommendations, they do it with the best possible understanding of the entire end-end system. And I think I've seen amazing things come in terms of the ability to diagnose problems, to make recommendations in real time when you
17:14 actually do feed it and provided that level of context over a long period of time. I think it's important. I think it would be interesting, too, like for people, budget and cost. When you were just talking, I was thinking about, yeah, they've gotten smarter. They've gotten smarter about opening a session with 17 other helpers that run up tokens like my kids run up tokens at Dave and Busters. And it's never-ending budget run. How should businesses be budgeting for AI? You know, I was doing the math.
17:40 If I allowed some of these things that happen, I'm like, I could have hired three employees. But talk about cost of AI and the power to do all of these things is that. that's going to get better over time? It's gotten better in some ways, but then, like you said, the smarter agents are opening 17 other helpers. And I'm like, shit. I mean, it was going to save some money until, you know, you got lazy and you needed 17 helpers. First off, your David Buster still has regular tokens. Like mine only has a credit card and my kids burn through that in like five minutes because they don't even know how much money's left. And so I got to go to your David Busters. No, it's that. It's that too. Figitive tokens. From a cost perspective, yeah,
18:15 That's a real issue. In long term, right, you'll see this notion of model routing where an intelligence layer that chooses the optimal model based upon the task and the question at hand. And actually, when you see a task, it fanned out to many different agents, there should be intelligence. The sub-agents don't always leverage the front-term model. From a Cisco perspective, we've been thinking about this. Again, as a 90,000 person company, our budget for tokens is pretty significant. And so as we think about how this end impacts our own customers, that's why we've invested. invested in building, training our own model that's specific to understanding network intelligence. And as our customers use Cisco products, it automatically selects a Cisco model for certain
18:55 questions, leverages, frontier models for other questions. And then what you'll see, right, is that technology will allow you to be able to then use as kind of ensemble of different models to help you to optimize based upon the question at hand. What I would encourage people to do is, you know, costs are very real. But again, we're at the early stages of how do we think about the of AI across our business and that, you know, we're just starting this era of token efficiency where the intelligence layer and model routing is being built into products to be able to select the right model. And what we'll see also is that over time, small language models that get put directly onto the hardware in devices where you're not even exercising a front tier model from a
19:37 token cost and you can do that fully in-house on the device itself. That's going to be an amazing unlock. I still firmly believe that the cost equation will work at sale across the business over time. And the businesses that win are they're going to be the ones that continue to lean into understanding and how to best use the technology to kind of further their business objectives and create the differentiation in the market. It's becoming the thinking system, the brain, the operating system for a company. When should you go to that point as a business? Cisco, so complex, so big, so layered, networking, being and having the brain for that.
20:18 When should a business flip that switch in their mind for building that out potentially? If you think about Cisco, part of our competitive advantage is that we've been in the networking business for 40 years. And so we've seen it, done it, fixed it for the largest companies in the world. That corpus of knowledge built up over 40 years, seeing how networks are built, how they're operated, how they break, and how they get resolved can then be data that we can push back as value to our customers. And so at Cisco, you know, we refer to that as a deep network model.
20:46 that's been trained on that 40 years of knowledge and really kind of writing the book, literally writing the book and how do you become a Cisco certified internet working expert? Now, the question is, when should businesses go down that path? Part of it is that do you sit on data that becomes part of your own differentiation as a business? And if you do, then there's a key unlock that could be had to be able to then leverage that in specialized ways. So instead of using a general purpose, LLM at a frontier, your customers are able to leverage a model that is highly tuned and highly trained on a particular class of problems that you have the keys, right? In the case of Cisco, we have the largest set of telemetry and understanding of how networks operate.
21:28 And creating that value as an unlock for our customers as a strategic advantage. In our own business, my advice would be if you have that in your business, then it becomes a key unlock. But it also requires investment to be able to do it. And so do it only if you think of your data as a key part of your differentiated value proposition. It brings up interesting with IP, trademarks and knowledge becoming IP. Not that that's completely new, but I do feel like the age of AI is bringing that about in a way because data gets shared so easily now in some ways. You have all this floating around.
22:00 You have all these models just absorbing it. Some of the things I've built, I go and scrape data from people in places that I can legally. I wonder if people are doing it to me. And sure enough, they were. It opens up a lot of questions in IP and otherwise. How does Cisco feel in position that or think about that? I've seen other customers, their businesses, begin to operationalize interfaces into their intelligence.
22:24 If I were to give you advice Ryan, I'd say, hey, if people are after your intelligence, then create the interface so that they can do it. It becomes a part of your differentiation. So you could create a model context protocol interface so that if agents do want to pull all of your podcasts and all of the intelligence that they do so in a structured way and from that semantic layer that unlocks it and is really an extension of the reach that you could have as a business.
22:51 And I've seen other customers, including Thousandize and Cisco at large, do that where we say, okay, look, we've always had an API for how people could get access to our underlying telemetry. And now we're going to create an agent-first interface so that if agents want to access that intelligence, they could do so as well. And here's how we do that, right? Because at the end of the day, that's in the interest of our customers. The price comes with value creation. And at the end of the day, my belief as a product leader is always to take a decade-long horizon for how we think about relationships with our customers. This is a good thing to do for our customers. And if customers find value in that, they're going to continue to one to do business with Cisco. And that's what I'm after.
23:27 Yeah, Joe, you got to make money, though, on all your knowledge, right? We actually just launched a new site where people can come and find. And we've got almost 700 episodes of business knowledge. And you can come search keywords in business and making it really easy to get to that. That's a little bit of what you're talking about. So that I definitely believe in. I guess it's more the money, you know, building APIs and different things. If you have broader data that has value in a way that it has to be synthesized,
23:52 but the data itself still has value on its own because it's historical and that sort of thing. Because, I mean, a lot of people sell access to data. I think that was more the context of the question. Yeah. It does. The AI economy is still in its early infancy. Whether it's Cisco or any business, we have to continue to really lean into the creativity. as these business models are going to change.
24:12 They know that, to kind of quote the famous line, the only paranoid survive. And so you should continue to be leaning in in thinking about and exploring new business models because the world will continue to change and you want to be able to ride that wave as it's hitting us. As we close out, Joe, I know that there's been a lot of research
24:29 around how quickly the shift is happening with agentic AI systems. I know that your teams and everyone is leaning into that. What would you leave our audience Any executives, anyone considering deploying those? You've pointed out some really key things. I think they'll be really helpful in the context, being very specific, putting guard rails. Is there anything else specific to sort of agentic rollouts that you would put a fine point on?
24:54 Business leaders don't outsource this to other individuals. Use it yourself. Same thing of you. Go build on a weekend project. When you see the power and you see that unlock, you see the way that you feel when you have that level of agency to go and get things done, You can then understand how to apply that across your business, not only in terms of your employee productivity, but also how do you transform the business processes, how do you transform the way that your operations teams manage your business? That's for me has been one of the key things
25:23 is to use it myself, because then I know and understand where to best apply it in terms of how I can build better products and how I can be able to kind of reshape my own organization. Wise advice. You got to be a practitioner. Very wise, though. I've been knee deep in it. I'm like, I'm not going to pay someone else to do. this. I'm going to figure it out myself, mainly so that then I can either understand it if we need to play other people doing it. I think those that are sticking their head in sand or just letting others, it's a really dangerous spot to be in. Yeah, use it. Everyone's a practitioner. It's been great to talk, Ryan. Yeah, it is. Give me some social handles, LinkedIn, anything like that.
25:59 You want to share as far as contact for learning more about what you guys are doing at Cisco or yourself personally. Take a look at, obviously, Cisco.com. Easy to find us on the web. I'm, on socials at X at Joe Vaccaro. And you can also find me at LinkedIn at Joe Vaccaro as well. Easy to find. Joe, we really appreciate your time. Yeah, thanks for having me again. This was a lot of fun.
26:19 Hey, guys, you know, to find us, Ryan iswrite.com. You'll find the entire episode, highlight clips, links to Cisco. They've been a great partner bringing us great executives and guests like Joe. We appreciate them for helping change the world. And we appreciate you for making us, number one. We'll see you next time. I'll write about now. Here's the truth.
26:38 Information doesn't change your life. Execution does. So don't just listen to this episode and move on. Take the idea. Make the call. Launch the thing. Fix the problem. Build what you keep talking about building.
26:56 For more, follow Ryan Alford on Instagram at Ryan Alford. And watch or listen to every episode at Ryanisright.com. This is right about. Now. Now quit waiting. Go win.