Meta Superintelligence Labs director exits to launch Ndoc, an AI-native Office rival

Rohit Patel, a director at Meta Superintelligence Labs, has left the company after eight years to launch Ndoc, a startup positioned as an AI-native replacement for Microsoft Office and Google Workspace, TechCrunch reports.

His exit extends a string of high-profile departures from Meta. According to TechCrunch, Chief Global Affairs Officer Joel Kaplan, AI researcher Andrew Tulloch, and Emily Dalton Smith, who led product for AI-for-Work, have also left.

A workspace built around AI from the start

Ndoc is not another AI assistant bolted onto existing software. Patel's goal is to collapse work that currently jumps between Excel, Word, PowerPoint, Tableau, coding tools such as Replit, and copy-pasting into a chatbot — into one shared workspace that colleagues can access even if they are not Ndoc users.

The interface has three tabs. The Stage holds the presentation layer: slide decks, dashboards, simulations, and mini apps. The Narrative carries the written analysis, with numbered citations and generated figures explaining the reasoning behind the numbers. The Model tab contains the underlying datasets and the Python scripts that produced everything.

That last tab reflects Patel's background. At Meta he worked on reinforcement learning and on evaluating whether AI output is trustworthy, and his position is that if AI writes the spreadsheets and decks companies use for decisions, someone has to be able to check the work. Even users with no Python knowledge can ask the built-in chat to explain a script or identify the line containing a formula — in his words, there is nothing hidden behind the curtain.

The product also takes context seriously: users can set audience and tone per document, and Ndoc maintains memory across the individual, the document, the team, and the company.

Open models and an open file format

Ndoc is built and fine-tuned from a mix of open source models, which Patel says makes it considerably cheaper to run than products relying on frontier models. He told TechCrunch that early trial users were largely indifferent to the cost story — what they responded to was seeing their work end to end in one place.

The bigger ambition is the file format itself. Documents export as a single .ndoc.html file that opens in any browser, and Patel plans to open-source the format so other developers can build compatible apps without the lock-in that has kept companies tied to Microsoft and Google for years. He said this was a key reason for leaving: inside Meta, settling on an open format would have meant reconciling hundreds of opinions, while a founder can move at the pace AI allows.

Three people, no funding, crowded field

Ndoc launches with a team of three and no outside money. Patel says he deliberately avoided fundraising until he has evidence that customers love the product.

The competition includes the largest AI labs. OpenAI, Anthropic, Meta, Google, and SpaceXAI all ship enterprise AI tools that plug into Workspace and Microsoft 365, and OpenAI recently launched Space, a shared workspace inside ChatGPT that targets the same territory more directly. Patel concedes Space could be a threat but argues the labs have it backwards: bolting AI onto spreadsheets to fill in cells is, in his telling, like discovering electricity and using it to strap lights onto the oxen ploughing a field instead of building a tractor. Because changing habits is genuinely hard, Ndoc includes connectors that import data from the tools customers already use.

As of Tuesday, Ndoc is running an enterprise pilot and opening an alpha to a limited number of waitlist sign-ups. Joining is free, with customers paying for usage and tokens; a beta is expected in the coming weeks and a full launch is planned for next year. Patel's operating philosophy, as he put it to TechCrunch, is that execution has become so cheap that shipping a suboptimal plan beats perfecting one — by the time the best plan exists, someone else will have executed.

Why it matters

The central question in enterprise AI right now is whether it arrives through existing tools or replaces them. The major labs have mostly bet on distribution, embedding their models inside Office and Workspace. Patel is making the opposite bet: that documents created natively by AI — with the data and code attached for verification — will displace file formats designed long before AI existed. An open .ndoc.html format would also attack the lock-in that has protected Microsoft and Google for decades. And his departure is one more signal that Meta, mid-race to build superintelligence, is losing senior AI staff to tiny startups that can ship faster than a large company can deliberate.