A Meta Director Resigns to Launch a Platform Rivaling Office and Workspace

After eight years working on artificial intelligence systems inside Meta, Rohit Patel, a former director at Meta Superintelligence Labs, decided to walk away from the tech giant to start a small venture with an enormous ambition: building a complete alternative to the Microsoft Office and Google Workspace suites — but this time designed around artificial intelligence from the ground up rather than having it bolted on afterward. The startup is called Ndoc, its team numbers no more than three people, and it has yet to raise any venture funding.

An Ongoing Wave of Departures from Meta

Patel's exit is not an isolated incident; it comes as part of a string of high-profile departures the company has seen recently, including global affairs chief Joel Kaplan, AI researcher Andrew Tulloch, and Emily Dalton Smith, head of AI products for work environments. This movement reflects the intensity of the current competition for AI talent, and the inclination of many researchers to launch their own ventures rather than wait out the long decision cycles inside corporate giants.

What Does Ndoc Actually Offer?

The core idea is that office work today is fragmented across multiple tools: spreadsheets, a word processor, presentation software, charting tools, a coding environment, and then repeated copying and pasting from AI chat interfaces. Ndoc attempts to merge these layers into a single space that colleagues can access even if they aren't subscribed to the platform.

The platform organizes work across three interconnected views:

• Stage: where the final outputs live — presentation slides, dashboards, simulations, and mini-applications.
• Narrative: the written analysis, with numbered citations and generated charts that explain the reasoning behind the numbers.
• Model: the original datasets and the Python scripts everything was built on.

Verifiability Instead of a Black Box

This architecture stems from Patel's own experience; at Meta he worked on reinforcement learning and on evaluating how trustworthy AI outputs are. That led him to the conviction that any system producing spreadsheets and presentations on which corporate decisions are based must be auditable. A user can open the script that produced the document, ask the AI assistant what each line does, and pinpoint and verify the line containing a particular formula — even without knowing Python.

In practice, you can upload a dataset, request an analysis, and have a narrative report written about the results, while adding document-specific context such as the target audience, the tone of writing, and saved information. Patel says the platform maintains memory at multiple levels: the individual user, the specific document, the team, and the company.

Open Models and a New File Format

Ndoc was built using a mix of open-source models with fine-tuning, which — according to its founder — makes its operating costs far lower than those of products built on expensive frontier models. Ironically, early users weren't excited by that technical point, but rather by the idea of seeing the entire workflow in one place.

The bigger ambition involves redefining the "file" itself: documents are exported as a single file with the extension .ndoc.html that can be opened in any browser, and Patel plans to open-source the format so developers can build their own applications on it without being locked into his platform — unlike the long-standing dependence on Microsoft and Google formats. This desire for openness was one of the reasons he left, since agreeing on an open format inside a company the size of Meta requires the decision to pass through hundreds of opinions, whereas he wanted to move at the speed of AI rather than the speed of administrative hierarchies.

Fierce Competition over the Workspace

The battle a three-person venture is entering is crowded with heavyweights: companies such as OpenAI, Anthropic, Google, and Meta have launched enterprise tools that connect to Google Workspace and Microsoft 365 so their models can operate directly within those environments, and OpenAI recently introduced a shared workspace inside ChatGPT that competes with the traditional suites more openly.

Patel believes this approach tackles the problem from too narrow an angle: attaching AI to spreadsheets to fill in cells instead of redesigning the way work itself is done. At the same time, he acknowledges how hard it is to change user habits, which is why he has provided connectors to import data from existing tools.

Conclusion

Ndoc is currently undergoing enterprise testing, with a limited alpha for those who join the waitlist; signing up is free with pay-as-you-go pricing based on usage and tokens, with a beta due soon and a full launch next year. The founder stresses that he won't pursue funding until he has proven that users love the product, and he sums up his philosophy by saying that execution has become cheap: it's better to execute an imperfect plan quickly, because the perfect plan may only be finished after someone else has already executed it.

✦ بقلم فريق دروب أيديا