-- Most enterprise software looks correct in a demo, in a pilot, and in the first year of a controlled deployment.Vimarsh Puneet, Director of Product Management for Enterprise AI at Cisco, has spent the last several years working on network infrastructure at a scale where those early impressions rarely survive contact with real operational conditions.
What Different Customers Reveal About the Same Platform
Puneet's product work at Cisco covers software that sits inside customer environments as different as a hospital network at Barts Health NHS Trust in London, a bank in Buenos Aires at BBVA Argentina, a manufacturing operation at Audi, and a university network at the University of Western Australia. All of these are public Cisco customer case studies, and they run very different operational patterns on the same underlying platform.
Photo Courtesy of Vimarsh Puneet
A hospital network measures success in patient safety and uninterrupted access to clinical systems. A factory floor measures it in production throughput. A university measures it in supporting simultaneous research computing and student-facing services on the same fabric. A bank measures it in transaction integrity under regulatory scrutiny. The software has to hold up under all of those conditions, and the failure modes look different in each.
That range is exactly why Puneet has said enterprise software cannot be built the way pilot software is. A design that works cleanly for one customer's operational pattern often breaks for another's, and the gaps only surface once the same platform is running under many different kinds of pressure at once.
"The network isn't just infrastructure. It's the foundation everything else depends on. You can't just build software and hope it holds; you have to build it enterprise-grade, because when the network fails, nothing else works," Puneet said.
That framing shapes how he approaches product decisions. An enterprise-grade platform, in his description, is designed from the start around the assumption that customer environments will be more varied, more constrained, and more consequential than any test environment can simulate.
Why Independent Validation Matters More Than Adoption Numbers
Deployment numbers describe how many organizations installed a piece of software. They do not describe whether it delivered the outcomes the organizations installed it for. On that question, external validation matters more than internal claims.
An IDC white paper studying Cisco Software-Defined Access and Assurance deployments reported a 462% return on investment over five years for the customers IDC interviewed, based on operational data collected from organizations running the platform in production. The finding is IDC's, not Cisco's, and it reflects a specific set of customers rather than every deployment, but it moves the conversation from adoption to outcomes, from "how many bought it" to "what happened once they were running it."
Puneet has said that distinction is one he pays close attention to as he moves further into enterprise AI product work. Adoption metrics are easy to produce and easy to overweight. Outcome measurement, especially independent outcome measurement, is harder to get and more informative when it exists.
That preference for outcome data over adoption data is not incidental to how Puneet approaches his current AI work. Enterprise AI, in his framing, is going to attract the same pattern of overstated deployment claims and understated real-world results that earlier waves of enterprise software attracted, and the counter to that pattern is the same: measured, external, operationally-grounded evidence of what a system actually does inside customer environments, not what its vendor says it does.
What Scale Teaches About What to Build Next
Working across a customer base as varied as Puneet's product responsibility covers is a form of continuous product research that no isolated design environment can substitute for. The aggregate operational data across many industries reveals patterns about what enterprises repeatedly struggle with, and what earlier generations of enterprise software repeatedly failed to solve, that no single customer engagement could show.
Puneet has said that base of operational experience directly shapes how he approaches enterprise AI. AI for enterprise infrastructure is not, in his framing, primarily a modeling problem. It is a problem of understanding what operators across many different kinds of organizations actually spend their time on, and what kinds of automated action a wide range of enterprises will trust an AI system to take. The answers come from years of operational contact with different customer environments, not from cleaner but narrower research settings.
That vantage point is what Puneet has said shapes the questions he asks about enterprise AI product design now, and it is what he has pointed to as the reason enterprise software has to be built the way he described in his own words: as infrastructure the rest of the business depends on, not as something layered on top of it.
Contact Info:
Name: Vimarsh Puneet
Email: Send Email
Organization: Vimarsh Puneet
Website: https://linkedin.com/in/vimarsh-puneet
Release ID: 89206202
Google
RSS