Cisco Live: Red carpet rolls
out for AgenticOps 

Another week, another AI evolution, but this time it was a leap forward rather than another tweak of the bot.

Cisco Live 2025 in San Diego unveiled a model of enterprise operations built for the AI era, and it doesn’t come with training wheels. The concept of AgenticOps, announced in the opening keynote, signals Cisco’s full-scale shift into an operational future where humans and intelligent agents collaborate at every layer of the network.

“This is perhaps the most consequential time for us to come together,” said Cisco CEO Chuck Robbins in attempting to convey the significance of the moment. “The payload of innovation that you’re going to hear about today is probably the most significant we’ve ever delivered here at Cisco Live in my career.”

That payload includes the agentic model Cisco has spent the last two years constructing: layering telemetry, automation, and observability across Meraki, Catalyst, ThousandEyes and Splunk. AgenticOps isn’t a rip-and-replace proposal. It’s a design philosophy that treats AI agents as full participants in enterprise workflows: sensing, deciding, and acting, while keeping humans in the loop.

Jeetu Patel, Cisco president and chief product officer, formally announced AgenticOps during the keynote session.

“We’re moving from an era of chat to an era where agents autonomously conduct tasks and jobs on behalf of humans,” he said. “This will fundamentally change the workforce. It will feel like 8-billion people have the throughput capacity of 80-billion.”

Robbins outlined the backdrop for this transformation in his keynote: “Last year, we were talking about chatbots and inquiries and efficiencies. Now we’re talking about: how do we manage the workforce -humans and agents?”

_{Chuck Robbins, Cisco CEO and chairman,during a media briefing at Cisco Live in San Diego. Photo by ARTHUR GOLDSTUCK}

The answer, Cisco says, lies in a convergence of the network and AI that goes far beyond integrations.

“Everything about what we’re doing today starts with this technology that we’ve all been around for a very long time: the network,” Robbins told the audience. “You’ve told us that the network is more important as we make this transition to agentic AI than it ever has been.”

The model builds on three pillars: infrastructure modernisation, embedded security, and intelligent automation. Said Robbins: “97% of you said modern networks are critical to make this happen.”

That sentiment underpins Cisco’s investments in programmable silicon, integrated security, and full-stack analytics.

“We believe that the era of agentic AI is only going to be secured by applying security services in the network to those workflows.”

AgenticOps introduces a logic engine of sorts to every enterprise function, powered by Cisco’s newly revealed Deep Network Model (DNM). Developed by Cisco’s AI team, DNM provides an AI-native representation of enterprise network behaviour.

“The DNM encodes decades of networking expertise into a continuously evolving model,” said DJ Sampath, Cisco’s VP of AI. “It captures not just the network topology but also application behaviours, device characteristics, and end-to-end flows.”

According to Sampath, the model draws on more than 550 billion data points daily and simulates network behaviour under different conditions.

“That kind of predictive reasoning enables agents to anticipate outages before they happen or recommend policy adjustments that improve performance. It’s a shift from reactive to proactive operations, grounded in AI that understands the language of the network.”

Where other AI initiatives focus on LLMs that answer questions, Cisco’s emphasis is on agents that drive decisions.

Anand Raghavan, senior director of product management, wrote in a Cisco Blog: “AgenticOps isn’t about building another chatbot. It’s about putting agents in the operational loop, with a clear mapping between outcomes, personas, and accountability.

“These agents can observe metrics, run diagnostics, initiate workflows, and even propose remediations. But humans still steer. It’s a design pattern – not a delegation of control.”

This pattern is already visible in Cisco’s products. AI Assist in Meraki helps IT teams troubleshoot campus networks using natural language. ThousandEyes agents recommend bandwidth shifts based on traffic predictions. In Splunk, agents assist with anomaly detection and root cause analysis. Each operates within domain-specific bounds, with built-in oversight features.

The architecture is as much about security as it is about scale.

Robbins said: “We need to view security as fused into the underlying network. To be able to apply network services and security services at wire speed while running one of these applications – that’s the only way to deliver secure AI.”

This idea of fusing networking and security, once separate disciplines, is central to Cisco’s claim of competitive differentiation.

“None of our networking friends have security. None of our security friends have networking,” Robbins said. “It’s an interesting place to be.”

AgenticOps also signals a broader shift in enterprise infrastructure philosophy.

“Private data centres are back,” Robbins declared. “You’re going to use the cloud, but you’re also going to have smaller models running at the edge. You’re going to need to work hard on building out your own private infrastructure to go with your public infrastructure.”

That dual-track approach ties directly to Cisco’s new partnerships with Nvidia and others for AI-ready data centres, including the recently launched Secure AI Factory.

“We want to help you build AI-ready data centres, future-proof your workplace, and underpin it all with digital resilience,” said Robbins.

That future-proofing goes hand-in-hand with agent augmentation, said Patel: “The workplace is going to have a fundamental augmentation of agents and robots and IoT devices … what we wanted to do was focus on three very specific things: operational simplicity powered by AI, scalable devices that are ready for AI, and fusing security into the fabric of the network.”

Patel unveiled a unified cloud management dashboard for Meraki, Catalyst and next-gen devices, available immediately, as well as AI-powered “multi-player infrastructure management” through AI Canvas: “Rather than a single-player model where everyone troubleshoots alone, we’ve created a collaborative environment powered by the Deep Network Model.”

Patel highlighted another cornerstone of Cisco’s approach: trust.

“Security is a prerequisite for AI adoption. If people don’t trust the system, they’re not going to use the system,” he said. That includes extending zero trust to not just users, but also to things and agents. “We want to reimagine zero trust for today’s language-driven world.”

Cisco’s innovation engine also included Live Protect – an automated vulnerability shielding system – and hybrid mesh firewalls designed for AI-native applications. Patel called the new firewall suite “the most comprehensive security platform for GenAI the industry has seen”.

Part of that resilience comes from unifying previously siloed data systems.

“Everybody was trying to build some big data lake. Now, with AI, agentic models, and new protocols to access legacy data sources, we can do this faster than ever to help you keep your organisation up and running.”

The role of trust was another major theme.

“You have to build it with integrity,” said Robbins. “And you have to build it with a consideration for humanity. That’s underpinned by trust.”

That trust must extend to the agents themselves. Raghavan underscored the importance of transparency: “AgenticOps needs to be observable. Users must be able to see what the agents are doing, why they’re doing it, and intervene at any time. We’re not hiding the logic; we’re exposing it.”

This ethos of responsible AI is baked into the system’s design, said Sampath: “It’s not enough for agents to be effective. They must also be auditable, explainable, and governable. That’s the only way we’ll scale trust alongside automation.”

AgenticOps enters a market gripped by urgency.

“98% of you feel extreme urgency to deliver on AI,” said Robbins. “85% of you think you have to do it within 18 months.”

The opportunity is vast, but so is the pressure.

“There’s a FOMO element,” he said. “Your CEOs want to move fast. They’re worried about safety and security, but they’re just as worried about falling behind a competitor who moves faster.”

Cisco’s message is clear: AI is already shaping enterprise decision-making, and AgenticOps offers a framework to navigate that complexity with intention.

Robbins closed the keynote with a cool summary of what’s at stake: “The bad news is: you’re all at the heart of making this happen. The good news is: you’re all at the heart of making this happen.”

* Arthur Goldstuck is CEO of World Wide Worx, editor-in-chief of Gadget.co.za, and author of The Hitchhiker’s Guide to AI.