Rohit Patel, a director at Meta Superintelligence Labs, is leaving the company after eight years, TechCrunch reported on Tuesday. His new startup, Ndoc, is not another assistant layered on top of Word and Excel. It is an attempt to replace the stack.
Patel's argument is that enterprises do not have an AI problem in their documents; they have a fragmentation problem across them. Analysis lives in a spreadsheet, the explanation in a slide, the actual computation in a script nobody opens, and the final answer often in a chat window that gets copy-pasted into all three. Ndoc collapses that into one file with three views: a Stage tab for decks, dashboards, simulations and small apps; a Narrative tab for the analysis, numbered citations and generated figures; and a Model tab holding the datasets and Python scripts that produced the numbers.
The Model tab is the thesis. Patel spent his Meta years on reinforcement learning and on evaluating whether model output can be trusted, and he is betting that the differentiator for AI-generated work is auditability. Users can open the script, or ask the assistant which line computes a given formula. "There's nothing hidden behind the curtain here," he told TechCrunch.
Ndoc is built on fine-tuned open-source models rather than a frontier API, which Patel says makes it considerably cheaper to run than competitors. Early users, he concedes, did not care: what they liked was having the whole job in one place. The company also plans to open-source its file format, a single .ndoc.html file that opens in any browser — deliberately aimed at the lock-in that Microsoft and Google have enjoyed for decades. Wanting to move fast on that decision, he says, is part of why he left: at Meta it would have required agreement from hundreds of people.
He is entering a crowded room. OpenAI, Anthropic, Google and Meta all now sell enterprise tools that plug into Microsoft 365 and Google Workspace, and OpenAI's shared workspace Space is a direct competitor. Patel's view is that they are solving the wrong problem: "Companies are basically attaching AI to spreadsheets and it fills your cells for you," he said. "It's like if you just figured out electricity and decided the best use case is to put lights on my bulls that till my soil. No. Get a tractor."
Whether a three-person, unfunded startup can become the place where enterprise work happens is the open question, and the odds are not kind. Ndoc is running an enterprise pilot and opening a limited alpha from Tuesday; signup is free, with usage and token charges to follow. A beta is expected in the coming weeks and a full launch next year.
Patel is also part of a pattern at Meta. He follows the departures of chief global affairs officer Joel Kaplan, researcher Andrew Tulloch and head of product for AI-for-work Emily Dalton Smith — a steady drain of senior people from the unit Mark Zuckerberg has made the company's biggest bet.
