Vimarsh Puneet has spent more than a decade at Cisco working on two of enterprise networking’s most consequential shifts: the move to software-defined campus networks in the mid-2010s, and the current push to run those networks with AI agents. He holds patents from the first era and now leads product management for the platform defining the second.
The Patents Behind the Platform
Puneet is a named inventor on U.S. Patent 10,887,175, which lets network policy servers enforce identity-based security rules even when Network Address Translation obscures a device’s real IP address, and on U.S. Patent 10,454,882, which enables DHCP address assignment to scale across large, software-defined network fabrics using overlay architecture and anycast addressing. Both patents were filed during the development of Cisco Software-Defined Access, the campus fabric platform Cisco introduced on June 20, 2017.
Cisco has described that launch in strong terms since. Cisco SVP Scott Harrell called it “a paradigm shift for our industry and a completely new era of networking” at the time. The hardware platform built around it, the Cisco Catalyst 9000 switch line, went on to become what Cisco CEO Chuck Robbins has repeatedly called the fastest ramping product in the company’s history, reaching 5,800 customers within nine months of launch, according to Cisco’s Q3 fiscal 2018 earnings call.
From Provisioning to Reasoning
Puneet’s patents solved a specific problem: how to make a network manage itself at scale, assigning addresses and enforcing security policy across thousands of devices without manual configuration. That work sits at the foundation of what the industry now calls intent-based networking, where administrators state a desired outcome and the network configures itself to match it.
Puneet now leads product management for Cisco AI Canvas, a generative interface for network operations that he presented at Cisco Live Amsterdam in 2026 under the title “Six Stages. One Story: Cisco AI Canvas and the Future of AgenticOps.” Where Software-Defined Access automated how a network provisions itself, AI Canvas is built to automate how a network operates day to day, using AI agents to handle tasks that have historically required a human engineer’s judgment.
“AI’s value comes less from replacing engineers than from turning the network into a continuously reasoning system,” Puneet said. He described the shift as moving the engineer’s job from operating individual devices to defining policy, constraints, and acceptable outcomes for a system that reasons across a network’s current state on its own.
What Comes After Intent-Based Networking
The jump from Puneet’s patent work to his current role is not a change in subject so much as a change in altitude. Software-Defined Access addressed provisioning and policy enforcement at the device level. AI Canvas is aimed at the operational layer above it, the day-to-day decisions that used to require a network engineer watching dashboards and responding to alerts.
Puneet has said the harder problem in that shift is not the sophistication of the AI model doing the reasoning, but the quality and consistency of the data feeding it. Enterprise networks, in his description, generate telemetry that is inconsistent across vendors and infrequent by design, which limits what any model, however advanced, can reliably conclude from it. Fixing that data layer, in his framing, has to come before an AI system can be trusted to act rather than simply recommend.
That distinction, between a network that observes and a network that acts, is also where Puneet has said the industry is furthest behind. Confidence in a model’s output is not the same as organizational willingness to let a system make changes on its own, and closing that gap involves questions of accountability and reversibility as much as questions of model accuracy.
Ten years after his name first appeared on a Cisco patent filing, Puneet’s work now sits at a different layer of the same problem: not how a network configures itself, but how it decides what to do next. The patents made the network provision itself. The current project is making it reason.
The two eras share a throughline beyond Puneet’s own career. Software-Defined Access proved that a large, complex enterprise network could be treated as a single system rather than thousands of individually managed devices. AI Canvas extends that same premise to operations rather than configuration, betting that the same unified view of the network can support automated reasoning as well as automated provisioning. Whether that bet succeeds at the scale Software-Defined Access did is still an open question, one that will play out over the next several years of deployment rather than a single product launch.
