Cisco’s Jeetu Patel: Businesses Need To Be Able To ‘Delegate Fearlessly’ To AI Agents

‘The thing I would say in 12 months is: We should make material progress on a customer’s ability to trust AI, and then I’d say we have to make it economical for customers,’ Jeetu Patel, Cisco’s president and chief product officer, tells CRN in an exclusive interview.


Cisco Systems sees AI as a massive opportunity for the channel, but realizing its full potential will depend on solving some of the technology’s biggest challenges, which include trust, security, infrastructure and cost.

Jeetu Patel, Cisco’s president and chief product officer, is no stranger to solving technology challenges. He believes that not addressing these big four challenges could prevent AI from reaching its full potential, which is not an acceptable outcome, the executive told CRN in a recent interview.

Patel has been busy with his teams bringing together Cisco’s networking, security, observability, data and AI infrastructure into a full-stack strategy designed to help partners tackle those challenges. With brand new tools and platforms such as AI Defense and Cloud Control, the company’s massive Splunk purchase and resilient infrastructure services, partners have new opportunities to help customers securely move AI agents into production, control the cost of intelligence and reach Cisco’s ultimate goal: allowing businesses to “delegate fearlessly” to AI.

Trust may ultimately dictate how far AI can go, and Cisco believes partners will be key to overcoming that hurdle.

Patel told CRN exclusively about how Cisco sees a major opportunity for partners as customers look to deploy AI agents securely, economically and at scale, and how the company’s full-stack AI approach is giving partners new ways to help customers move AI projects into production.

Here are excerpts from the conversation.

What’s the biggest, if you could pick one biggest thing out of Partner Summit, that you want partners to know that you think is creating the biggest opportunity for them?

Firstly, the world is excited about the possibilities of AI, and the thing that could, oddly enough, prevent AI from reaching its full potential is safety and security. And so we have to make sure that we come together as a tech community and make sure that we solve that problem because it’s not an acceptable outcome to not effectively use AI to its fullest potential because people are afraid. And right now, the trust quotient is pretty low. People don’t feel like they can delegate work to an agent in a trusted way, and so we have to make sure that we enable that condition. We’re spending billions on our side in making sure that happens.

The second big area is, we are in a very long-term supply-constrained world where, firstly, there’s no shortage of demand, and that demand is growing exponentially. The exponential curve is almost like a vertical line. But the thing about the supply that’s really interesting in this AI movement that’s happened is the supply is instantly consumed. It’s not like during the 5G days there was a lot of dark fiber and the fiber didn’t really come online, and you had to wait for the demand to catch up with the supply that was there. There’s none of that this time around. The supply gets instantly consumed. That, in my mind, is a very interesting dynamic because what we have done is we have made sure that we’ve provided critical infrastructure for AI for every tier of the market: hyperscalers, neoclouds, sovereign clouds, service providers, and then for the enterprise.

The big shift [for Cisco] is we are now a completely vertically integrated platform that’s a co-designed full stack. We make our own silicon, we make our own photonics, we have our own systems hardware. We have our own systems software. We have our own platform for data. We have our own platform for security and observability. We’ve got our own models. We’ve got our own applications. We’ve got our own agents. And they all work in harmony with one another. What we are doing for partners that’s going to be really important is … Model Foundry, and that’s a capability within Cloud Control that allows the partners to post-train models on behalf of their customers with their data in a way that they can actually lower the cost of intelligence quite materially for things like cybersecurity, for things like infrastructure management, and we’ve provided all the tooling and apparatus for them to be able to do that. That’s a huge, huge step forward for the industry, because I’ll say this: this is really important in a world where intelligence is abundant. The true differentiator is who is going to be able to more effectively and efficiently use tokens. That becomes your strategic lever and your moat. If you can, as a cyber defender, use intelligence where the density of intelligence per token is higher than the attacker, you’re going to have an advantage.

Cisco is increasingly talking about being an open platform and supporting third-party infrastructure. How far are you willing to take that if a partner isn’t going to sell an entire Cisco stack?

We’ve taken it pretty far already. A lot of our competitors end up being our partners, and I think this notion of the distinction between: This company is a competitor versus this company is a partner, is actually disappearing. The way that we’re thinking about this is: Start from the customer and work backwards. If the customer is invested in two technologies and we happen to be one of them, we owe it to the customer to make sure that they get the fullest return on investment from the other technology by us partnering with them. For example, all this started with Microsoft Teams running natively with our Webex devices and Cisco devices. We also now have that same capability [with] Google, as well as Zoom. That is an example of direct competitors in a certain area and how we’ve said: Let’s make sure that we don’t let that come in the way. Now, we’ve seen direct indication of what the level of impact that strategy has had on our business because our devices business is amazing because of that strategic pivot that we took.

With Cisco bringing together its different business units, like networking, observability, and security, what can a partner sell or deliver today that they couldn’t even a year ago?

I think the level of comfort that a customer can get. Firstly, you have to make sure that the way that an organization decides to use agents is going to be material to their long-term success and differentiation. One of the big paradoxes with agents is the tools access paradox. You have to give agents access to your tools, so that the agents can actually reach their full potential for you. But in order to give it access to your tools, you have to feel like you can trust the agent. If you don’t trust the agent, you basically don’t get the most out of AI. That’s one of the things that we are helping solve as a problem, which is, we want everyone to use agents to the fullest capacity. As a result, we’ve provided a full apparatus with AI Defense, not just giving customers full visibility in what agents happen to be in your environment, but what can you do with those agents, and how do you make sure that those agents are going to be protected from the world of attacks, and how do you also make sure that you’re protecting the world from those agents by making sure that it stays within the guardrails and is aligned with human values and expectations. That’s an area that there’s a whole bunch of technologies underneath that on identity for agents and discovery of what an agent does within your organization and making sure that you can monitor agents’ behavior and dynamically intercept at runtime if you start to see drift from an agent’s behavior … All of that did not exist a year ago.

Every year, Cisco becomes an unrecognizable company for the better because the pace at which the innovation is happening is so rapid. You can see it in every single product that we’ve released. By the way, Splunk is a huge part of the entire equation. We just hired a senior exec that used to work at Google, Sunil Potti [Cisco’s senior vice president and general manager of security, observability, and data platform] who is now running both Splunk and security. We’ve combined those two businesses together under one leader. That changes things quite materially, so I feel like there’s a lot of things to be excited about.

With Cisco encouraging partners to lead with resilient infrastructure services, what does that look like in practice and where do you see the biggest opportunity for partners there?

I think the biggest opportunity for partners in general is in making sure that they can continue to keep adding value for customers.

Your ability as a customer to have infrastructure that is scalable, reliable, dependable and secure is directly proportionate to the strength of your business. It’ll be very hard to imagine a company that doesn’t understand how to efficiently and securely generate and use tokens. If they can’t do that, it’s very hard to imagine how they’ll be a very differentiated and strong, durable company in the future.

What we want to do is we want to provide all the apparatus for that happening. If you do have an outage, can we predict that outage before it happens? Can we make sure that the time to investigate and detect an outage gets compressed, and the response and remediation happens as quickly as possible? If it is a breach, can we ensure that that breach gets proactively intercepted, so that there’s no lateral movement? It is going to get harder and harder for organizations to distinguish between a security breach or an agent just misbehaving because they thought it was the right thing to do because they took your instructions very literally and tried to follow them. For us to be able to take the telemetry between the security and the observability data sets and correlate them together, so that you’re not looking at those in isolation, but you’re looking at this together and saying, “Hey, this agent is actually doing something that might look suspicious, even though it’s not a security breach, and so we should be able to intercept it. This agent is consuming way too many tokens, and they could actually blow our token budget because they’re just keeping on going in a loop. We need to make sure that we pull the kill switch on the agent." All of those things need to be provided in one cohesive platform, and that’s what we’ve done, from the way that we can actually monitor GPU usage, to the way that you actually monitor the use of tokens.

Many enterprise projects don’t make it from pilot to production. From a product perspective, what’s preventing customers right now from moving more of those into production?

I think it’s gotten a whole lot better today than it was six months ago, and I think you’ll see an exponential improvement in that over time. I think part of it is, we got into an experimentation phase for a while initially, and the thing with AI was, it was very easy to build seductive prototypes. It’s very hard to build durable value from these agents and applications over a long period of time. So I do feel like, from that perspective, you had this phase where people just were going out doing demos because they were so easy, but they hadn’t really thought about: What does it take to make that demo truly a production application?

I feel like in the next 6-12 months, the enterprise take rate will be much higher because this is one of those areas that will be led by consumer. Everyone’s now starting to use these agents, the personal agents like Muse and Instinct and Dots, [and] that’s giving them a lot of ideas on the kind of things you can do. Then, the other thing that you have to keep in mind is just slapping AI on top of a broken process doesn’t really bring you the returns. You have to actually re-engineer the process itself.

When we started using AI, one of the things we looked at was, yes, we have to make sure that everyone’s coding using AI, but how are we coding? What are we doing in the way in which we think about products to build? That core process had to get rethought as we were using AI. We weren’t just thinking about slapping AI on top of a broken old process, and I feel like that’s something that every company is going to need to learn to do a really good job at.

A year from now, what will tell you that Cisco’s AI strategy is resonating through the channel?

The first thing we have to do is make sure that we make infrastructure available at scale to our customers, so that they can feel like they have the appropriate amount of tokens to go out and do the things that they need to do. The first thing we have to do is give them an underlying apparatus for that to happen. Which is why we have partnerships with Supermicro. Which is why we’re building out multiple different families of silicon chips. Which is why we’ve got such a huge investment we’ve made in photonics. Which is why data center buildouts that are happening are so sophisticated.

We want to make sure that regardless of where the customer wants to consume and generate tokens, whether it’s hyperscaler, whether it’s the neocloud, whether it’s a sovereign cloud, whether it’s through their service provider or whether it’s within the enterprise or on the edge, we’ve given them full flexibility to make sure that they can actually use our technology. We don’t care what model and architecture they pick. We will actually support them with that. That’s a pretty important dynamic that we’ve actually set, and that hopefully will get customers to feel like: "I’m going to get a huge amount of benefit from this." A year from now, I would say that that will just get easier and easier.

The second thing is our platform, especially with Cloud Control, we feel it will be the primary way that customers are actually managing the infrastructure. I think in a year you’ll see huge progress. The third area is safety and security, and us making sure that we have enabled our customers to be able to, in a trusted way, delegate fearlessly to agents. I think that is going to be a pretty important dynamic. We’re spending a lot of effort in making sure that people aren’t afraid of having agents work on their behalf, but that they actually feel comfortable that they’ll be able to make sure that agents work. That means that you’ll have to lower the cost of tokens. That means that you’ll have to make sure that you have the right level of visibility. That means that you’ll have to make sure that you have the right level of runtime intercepts. All of those things will have to be in place.

Our AI Defense product is the core engine for that, and that takes into account everything that we have from all of the underlying components, whether it be eBPF and the Hypershield technologies that we have, or whether it be Splunk or it be Zero Trust. We want to make sure that all of those are just easily serviceable with AI Defense.

So, the thing I would say in 12 months is: We should make material progress on a customer’s ability to trust AI, and then I’d say we have to make it economical for customers.