Cisco is taking steps to fundamentally change enterprise network management with agentic AI. AgenticOps should move the needle beyond traditional AIOps to autonomous systems powered by reasoning models and 40 years of networking expertise.
At Cisco Live EMEA, Anurag Dhingra, SVP and GM for Enterprise Connectivity and Collaboration at Cisco, visited our studio on-site to discuss the company’s ambitious vision for AI-driven network operations. The strategy centers on what Cisco calls AgenticOps. You could see this as an evolution from the AIOps approaches that dominated before the advent of large language models and reasoning capabilities.
From AIOps to AgenticOps
“AIOps really started before the ChatGPT moment, before LLMs were a thing, before reasoning models were a thing,” Dhingra says. “Think of AgenticOps as sort of the natural evolution of that with really a model that can reason, that can plan, that can think, and then you apply that as the core of agents that help you do operations.”
The foundation of Cisco’s approach rests on cross-domain telemetry aggregated from its extensive product portfolio. Dhingra is responsible for a big chunk of that: campus and branch networking, industrial networking, Thousand Eyes observability, and the WebEx collaboration suite. This data is then combined with what Dhingra describes as “CCIE level information”. In other words, decades of accumulated networking expertise codified into systems that autonomous agents can leverage.
AI Canvas and dynamic dashboards
One of the flagship capabilities we talk about with Dhingra is AI Canvas. This is a platform that generates dashboards on the fly using generative UX. Unlike traditional static dashboards, AI Canvas creates visualizations dynamically based on what users are investigating in the moment. “It generates dashboards on the fly, and then those dashboards are completely created in service of what you’re asking in the moment,” he says.
The system allows network administrators to save useful views, share them with colleagues, and even vote on dashboard configurations. All of this creates a collaborative, evolving approach to network visualization. Perhaps more importantly, it should also create dashboards that people actually use.
The orchestration platform built into AI Canvas can pull information from third-party systems like ServiceNow and coordinate across specialized agents that understand different parts of the infrastructure.
Autonomous troubleshooting and proactive operations
Cisco’s agentic approach wants to shift network operations from reactive to proactive. Autonomous agents continuously monitor networks against predefined experience metrics, such as Wi-Fi connection times. When thresholds are exceeded, agents can investigate and troubleshoot autonomously without human intervention.
“These agents can be completely autonomous. They’re constantly monitoring your network. They’re looking at all the telemetry that’s available to them coming from network devices, coming from clients, coming from security appliances,” Dhingra says. The system has already demonstrated results with wireless infrastructure, where AI agents constantly tune radio resource management for optimal performance.
Technical architecture and MCP servers
The technical implementation leverages MCP (Model Context Protocol) servers deployed at the controller and management layers rather than on individual network devices. “Pretty much everything that we used to do in the past with APIs is now available wrapped into MCP servers. Those are the tools that agents can use,” Dhingra states.
Cisco employs a mixed approach to AI models. It combines frontier models with custom-trained models that contain network-specific information. This ensemble strategy helps manage inferencing costs while maintaining capability. After all, not every task requires the most powerful and expensive models.
Managing costs and building trust
Dhingra acknowledges that AI inferencing costs remain a concern but expresses optimism about continuing improvements in model efficiency and cost. Currently, Cisco focuses on delivering value to customers rather than monetizing AI capabilities directly. “Right now we really focus on getting the technology in the hands of our customers, making sure they’re getting value out of that. We’re not trying to monetize or make money out of that right now,” he says.
Building customer trust represents another critical challenge. Cisco addresses this through transparency about models and data sources. It implements AI Defense for model testing and guardrails, and carefully designed human-in-the-loop workflows. Agents can present plans for approval before execution, giving administrators visibility and control over autonomous actions.
The evolving role of network administrators
Will all of this agentic AI integration make network admins redundant? Certainly not, Dhingra says. It reimagines those roles. “If you are a seasoned network administrator or network ops person now, you have very capable, almost digital teammates available to you and so you can delegate a bunch of routine tasks to them,” he explains. This allows professionals to focus on designing networks and operating them at scale rather than handling small, repetitive tasks.
Future vision: collaborative agents
Looking ahead, Dhingra envisions an ecosystem of specialized agents coordinated by orchestrator agents that act as supervisors. These specialized agents will focus on specific domains like enterprise networking, data center operations, or security, while orchestrators handle cross-domain tasks.
The interaction model will also diversify beyond dedicated applications. Agents may proactively send messages through Slack, Webex, or Microsoft Teams. In doing so, they integrate seamlessly into existing workflows and don’t require users to navigate to specific management interfaces.
Customer adoption and market readiness
Dhingra acknowledges that technology is advancing faster than many customers can absorb it, with adoption currently lagging. Cisco is addressing this through close collaboration with design partners to understand their workflows intimately and ensure technology fits naturally into existing processes.
Current availability and early results
Cisco’s agentic capabilities are already available in production through the AI Assistant in Cloud Dashboard, which customers have been using for several quarters. AI Canvas is currently in early access with select customers, who are reporting reduced troubleshooting times and improved network performance, particularly in wireless environments.
Also read: Cisco’s AI Canvas goes beyond where SaaS and AIOps fall short