The file system has never had a good publicist. It works in folders, respects names such as “final_v7_really-final,” and asks only to be noticed during an outage. Artificial intelligence arrived with better lighting. Yet between the clever agent and the useful answer sits a stubborn old question: which file may it read, which version is current, and who is allowed to know what is inside?
Karthik Ramamurthy has spent more than two decades living beneath that question. His career runs through Symantec, NetApp, EMC's Isilon business and Veritas, a tour of enterprise storage's changing vocabulary. In January 2026, he joined Panzura as chief product officer. By April, after leading product, engineering and go-to-market strategy, he had been appointed chief executive.
The quick promotion looks less abrupt when viewed from below the org chart. Ramamurthy has moved repeatedly between the machine room and the operating model: engineering and product management, protocols and alliances, cloud architecture and subscription transformation. His present thesis is the accumulated version of those jobs. Enterprise AI will only be as useful as the file data it can reach - and only as trustworthy as the permissions and governance it preserves.
A career built around failure
Not personal failure. System failure - the wonderfully concrete sort that engineers can diagram, reproduce and then try to prevent. Ramamurthy is named on two granted U.S. patents developed during his Symantec years. One describes how to provide high availability to service groups inside a data center. The other uses the physical location of cluster nodes to keep a shared point of failure from taking everything down at once.
The patents are revealing because they concern topology as much as technology. Reliability depends on knowing what is connected, what is shared and what can disappear together. Nearly two decades later, that same instinct appears in his AI argument. A model's answer is not reliable merely because it sounds complete. The underlying data needs a map: where it lives, who controls it, when it changed and whether the person asking has permission to see it.
Co-files two applications on high availability and avoiding single points of failure.
Moves into EMC/Isilon product leadership as both inventions become granted patents.
Expands at Veritas from storage-product leadership into subscription, cloud and operating transformation.
Joins Panzura as CPO, becomes CEO, and leads a file-data agenda built around AI access and governance.
Former colleagues have described him as detail-oriented, results-driven and unusually good at breaking complicated problems into a sequence an engineering team can act on. One Symantec colleague recalled how he connected customer needs with the roadmap and kept a sales-enablement focus while shaping emerging storage products. These are not cinematic traits. They are useful ones when the product's triumph is that thousands of people open a file without thinking about why it worked.
“Unstructured data is now the backbone of AI and enterprise transformation.”Karthik Ramamurthy, on becoming Panzura CEO
The Isilon years: all the protocols, please
In 2016, a Dell EMC community introduction gave Ramamurthy a wonderfully maximalist brief. As an Isilon director of product management, he ran “all things” data services and protocols, plus big data and Hadoop, cloud and software-defined storage. The list sounds like somebody lost an argument about prioritization. It also captures the reality of enterprise infrastructure: customers bring heterogeneous systems, long memories and inconvenient requirements. A clean whiteboard rarely survives first contact with their data center.
During an online expert session on Isilon all-flash systems, the conversation descended quickly into the exact questions storage buyers ask: network interfaces, cache behavior, mixed drive capacities, data movement and disruption. This was Ramamurthy's habitat before the executive biography polished the nouns. Product management here meant translating physical constraints and customer workloads into choices that could survive production.
At Veritas, the aperture widened. Ramamurthy led appliance and software-defined storage product work, helped define product development and delivery strategy, and handled alliances with AWS, Microsoft, VMware, Google, Nutanix and Pure Storage. He later held senior vice president roles in subscription and cloud transformation, then transformation and operations. The engineer's problem had become an organizational one: how does a mature enterprise-software company change what it sells, how it builds and how it operates without losing the customers already depending on it?
Files become an AI problem
Panzura's world is full of project files, design documents and office records scattered across teams and locations. That material is called “unstructured,” a description that is technically serviceable and socially rude. The structure exists. It lives in folder conventions, filenames, metadata, access lists and the institutional knowledge of people who remember why a drawing was revised three years ago.
Ramamurthy's argument is that companies should not have to extract all of that into a second data estate, rebuild its context through middleware and hire a small caravan of consultants before an AI system can be useful. He wants the file platform itself to supply an AI-ready data layer. The difficult part is not merely letting Copilot find a document. It is keeping the answer synchronized with the live file and faithful to the access controls already governing it.
+ metadata
+ governance
AI answer
That is the purpose of Panzura Nexus, launched in April 2026. It connects CloudFS data with Microsoft 365 Copilot through an event-driven pipeline. When a file or its access-control list changes, the system captures the change rather than waiting for the next scheduled crawl. A person should not discover a confidential document through an AI conversation simply because an index has not noticed that their permission vanished yesterday.
CloudFS supplies the global file layer; Nexus makes selected content and metadata conversational; Symphony discovers and orchestrates unstructured data across heterogeneous environments. The three products form an increasingly clear strategy. The file is not a forgotten attachment at the edge of the AI stack. It is the record of work, wrapped in context and rules.
“Enterprises have spent decades accumulating file data and now realize it is the fuel for AI.”Karthik Ramamurthy, August 2026
A product-led CEO, in sequence
Ramamurthy arrived at Panzura when “AI-ready” had become an obligatory adjective. His version is unusually operational. On becoming CEO, he named three priorities: strengthen leadership in AI-ready data, scale through partners and channels, and help customers turn data into a strategic advantage. Each priority implies work outside a model window. The data must be governed. The partner needs an economic reason to carry the platform. The customer needs an outcome worth changing infrastructure for.
The release sequence tells the story better than a mission statement. CloudFS 8.7 lowered infrastructure and management costs while preparing the platform for agentic workflows. Nexus connected governed files to Copilot. Support expanded for service providers building managed file, security and AI-data businesses. Then Panzura Express packaged file services, ransomware resilience and AI access for smaller single-site organizations.
Express reveals the operator in Ramamurthy. Instead of beginning with a grand theory of democratization, he begins with a cramped IT budget and a hardware queue. The product runs as software on existing virtualization or in a major cloud, against compatible object storage. His pitch is blunt: a two-person IT team can approve an operating expense and stand the system up in an afternoon, rather than waiting for a capital project and a distant delivery date.
It is also a channel-first move. Express is sold through partners, a decision consistent with his view that service providers deserve a platform on which they can build recurring businesses. Infrastructure companies do not scale on architecture diagrams alone. Somebody has to sell, deploy and operate the diagram in a world of procurement calendars and understaffed teams.
The old craft inside the new wave
In August, Ramamurthy joined Forbes Technology Council, where he plans to write about making unstructured data usable by AI, governing data that agents can act on, and preparing file infrastructure for agentic work. These themes are new in label and old in substance. They return to the questions running through his patents and product roles: What is the dependency? Where is the failure point? Is the system's picture of reality current?
This continuity is the most interesting part of his career. He did not leave storage behind for AI. AI wandered into storage and found him already there, fluent in the awkward details. The fashionable layer now depends on the unfashionable one, as fashionable layers have a habit of doing.
For builders, Ramamurthy's stealable idea is simple: start beneath the demo. Identify the condition everyone assumes will magically remain true. In enterprise AI, that condition may be a trustworthy, permission-aware source of data. In another market, it will be something equally uncelebrated. The opportunity often sits where applause is scarce and failure is expensive.
Panzura's wager still has to play out across customers, partners and competing architectures. But it has the virtue of beginning with reality. Companies already possess years of knowledge in files. Employees already use permissions to decide who sees them. AI systems already want access. The useful question is not whether those facts are elegant. It is whether somebody can make them work together.
Ramamurthy has built a career making complicated systems behave. Now his brief is to make their accumulated memory answer back - without forgetting the locks on the doors.