President of Cisco, Jeetu Patel, on How Trust Is the Unlock for the Agentic Enterprise

We're fully AI'd at this point.

Absolutely. [laughter] The most obvious thing. Um, let me actually take a step back and just, regardless of what company you use. Um, I know that there is a lot of negativity in America around AI, that is the meme that actually spreads, but I tend to be very long on AI. I feel like there is a tremendous amount of upside and we are about to experience the second ChatGPT moment.

And I thought it was going to happen when OpenClaw came out. I was wrong. It didn't happen when OpenClaw came out because I think it was too complicated for the average user. I think all the geeks were able to use it. They were getting their Mac minis and then Mac minis sold out and so you didn't even have a place to go put it. But um where we are now, I think the personal agents are going to be like that explosion factor that's going to be there because right now if you think about it, less than 2% of humanity uses agents in any kind of power user capacity. I think you'll see probably in the next couple years 50% or more of the people in the world will use agents in a meaningful capacity. And because it's gotten to be so simple, I don't know how many of here use Grokpot with a show of hands. Okay, few. How many have you used Muse?

Instinct. Um, I think when you use Instinct, you actually see the OG level. That's, it's completely crazy what it does. And it's um, and in my mind, what's going to happen is right now we are with 2% of humans using agents.

Um, the agent token consumption exceeded human token consumption in February. And since then, right now, they're 5x of human token consumption. So agents are consuming five times that of humans before personal agents came into the picture. So as of August, I think you'll actually see a much, much more steep acceleration of that. 60% of the entire world's compute capacity now by the end of '26 will be um, inference, not training.

Right? And so you're starting to see that. Why is that consequential? Because up until now, scaling laws were applying. People were going out and, you know, evolving their models more. But now what you're starting to see happen is people are using these models. They're getting value from it. I think that value is going to just explode on the consumer side with personal agents and then people will try to replicate that on the business side. Like we just rolled out 90,000 agents, one to every employee at Cisco. And I was skeptical, you know, whenever like if you have internal tools, you always tend to be a little bit more critical, like I'm not really sure if our people are going to build it as well. It rocks. Like we are able to go out and do things that we were never able to do before. Our operational kind of rhythm and cadence is starting to change and um, the way in which we go out and use these agents for um, you know, personal productivity and team productivity is um, is starting to change. So um, so I do feel like from that perspective we are about to enter into a um, a massive agent super cycle that we haven't seen yet and that's going to have implications on infrastructure. It's going to have implications on um, on every layer of the stack, right? And therefore it'll have implications on every company in the sense that we will all be benefactors of it if we actually do it right. So I feel like Cisco has a lot of excitement ahead but that's also because um, you know, we provide infrastructure and plumbing and security and safety and observability for um, [clears throat] for agentic workflows and for agents. And so I think like there's a secular, at this point in time there's a secular shift. I don't think when people say, you know, is this going to be a blip, I don't see how it becomes a blip in the sense that there is a secular shift. And the secular shift is going to be very consumptive on infrastructure and I don't think that's slowing down anytime soon. And so we should all uh, you know, think about this very deeply and the risk is around safety and security and we should not be delusional on that front. I think there's a real risk. Um, but the way that I think about it when someone says to me, hey, should we slow down and pace AI? I say to them, regardless of what you do, we should be thinking of speeding up security and safety. Like that's the number one priority because if we can actually do trusted delegation to agents, I think you will see massive levels of uh, productivity gains in society. Uh, if you don't do trusted delegation, it's business-ending events that'll happen.

I think the personal agents, the way I think about this is what it does most visibly is it gives everyone the, the beauty about ChatGPT when it came out was every [clears throat] person, your grandmother started using ChatGPT and they would tell you about it, you know, like everyone was starting to use it. And so there was this relatableness that happened because everyone had stories about how they were using ChatGPT and that was the pull that got created for AI and that helped the entire industry. I think that same thing is going to happen with personal agents because I don't think people still have, you know, we all talk about agents. We've been talking about agents for a while. I don't think people fully understand what happens in single-shot task execution all the way through and you tell the agent, go ahead and cancel my subscription with Xfinity, right? And it'll just go ahead and do that. Go ahead and sign me up for this other service. It just goes ahead and does that. You don't have to do anything. It just does it. And that to me is, I don't think everyone, like we are all living in a bubble and we all kind of live in the tech community and so it's different for us. But when you go outside, you go in the middle of Iowa, you go in some parts of Tennessee and you ask people what they're doing. Like I'll give you an example. I was in um, in Baltimore. I was doing my polygraph test because I was getting cleared. I was getting, you know, you kind of tested for security clearance. And so I was sitting there because they take away your phones. You're just sitting there. You're not going to do anything. So you are forced to talk to people. So you know, like so the guy next to me was sitting down. He was a young kid. He's like 25, 26 years old. And I'm like, so what do you do? He's like, well, I was in the military and now I'm actually going to work in the government. So I'm getting my polygraph test done because I have to work in the government. I'm like, oh, cool. He's like, what do you do? I'm like, oh, I work at Cisco. He goes, Cisco? Never heard of it. [laughter] Like, oh, okay. Uh, good. I'll let my marketing team know. But um, but like, uh, have you heard of Microsoft? He's like, of course I've heard of Microsoft. Like, who are you? You think I'm an idiot? Uh, I'm like, oh yeah. I'm like, have you heard of OpenAI? Nope. Never heard of OpenAI. I'm like, have you heard of ChatGPT? Absolutely. Love ChatGPT. I'm like, oh, okay. Have you heard of Anthropic? Nope. No idea who Anthropic is. I'm like, have you heard of Claude? Yes. Love Claude. So my point is that I just don't think that people are as entrenched into the details on this stuff as we all are. And for us it's all common knowledge on all these different, you know, kind of companies that we have. But I think when you go out in the market, I think there's still a level of delta between what is common knowledge in the inner circle and what's common knowledge within the rest of the country. And when the rest of the country and the rest of the world gets consumerized with something where it becomes an aha moment like ChatGPT, like there's a super cycle that forms right behind it. And I just, I think like that is about to happen again with personal agents. You know, and your question, what was your question? I didn't answer your question.

From an enterprise perspective, I think what was happening with AI for a while is it was really easy to get good demos, um, and then this wasn't working in real life. And now what you're starting to see is um, the technology has gotten advanced enough, the iteration cycles and the iterative loops are getting better. I think you're starting to see these agents self-improving quite a bit, that um, I think the production levels are getting to be pretty exciting. And so we had built some technologies which were like, you know, they were demoing really well but they weren't really quite hitting the mark on the users. And one of the things we do is we use Cisco as customer zero and I tell my IT team, I'm like, be brutally honest and do not just placate and give me like false niceties. If you don't like it, your verbiage to me is this sucks, it does not create value, right? And then every week they give me a detailed report of what's working and then I forward it onto my team. And the only goal that they have is when they've built something, I'm like, don't add another feature. Get the customers to start loving the product that you've built. Like that's the most important thing. Um, and we actually put some forward-deployed engineers whose only job was to make Cisco IT successful because you want to learn in that iterative kind of iterative loop. And we're finally now hitting that state of flow where these IT people will get on a call with us and for 35 minutes they'll just talk about how everything is so great and they were never able to do something like this before. That's an exciting thing when you start to see product-market fit and it's actually starting to hit. And I feel like um, you know, Instinct is that way. You're starting to see Muse being that way. And so I think this will have a secular shift 2.0 that'll start happening in AI because of it.

We should all stay pretty excited about this. We should also be very paranoid about security. It's scary what can happen. And so we should um, we should make sure that, and especially with agents, like you only get as much out of an agent as the amount of access you give the agent, but you give the agent too much access and then you're screwed. And so like you're kind of like you have to make sure that you're kind of modulating how you do it. Um, so I, for example, I gave um, Instinct tools access with my email, for my personal email, for three days and then I'm like, you know, this is just a bad idea. I'm a Section 16 officer of a Fortune 100 company. Like, can I just take it back? And so I took back the access and then I'm like, well, are they going to use my cache? And so I asked the agent and they're like, no, we explicitly will delete your cache if you tell us to. And so I'm like, delete the cache. Deleted the cache. Um, I had some inside sources who told me like, yes, yes, yes, that's what they do. But I do feel like that's an area that I just don't think we are in the state yet where people are trusting the system. So how long it takes for us to start trusting these agents is going to be the definitive um, you know, kind of window of time that you need for um, it not getting delayed too much.