AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Meta AI director Rohit Patel leaves to take on Microsoft Office and Google Workspace

Collected Oct 27, 2026

Rohit Patel, most recently a director at Meta Superintelligence Labs, is leaving Meta after eight years, TechCrunch reported. He has launched Ndoc, a startup he pitches as an AI-native replacement for Microsoft Office and Google Workspace. The company has a team of three and no funding, and it began an enterprise pilot on Tuesday while opening an alpha version to a limited number of customers who join its wait list. Signing up is free; customers pay for usage and tokens. Patel hopes to ship a beta in the coming weeks and a full launch next year.

Patel's departure is the latest in a series of high-profile exits from Meta. The article names Chief Global Affairs Officer Joel Kaplan, AI researcher Andrew Tulloch and Head of Product for AI-for-Work Emily Dalton Smith among those who left before him. At Meta, Patel worked on reinforcement learning, where agents learn decisions through trial and error to maximize a reward signal, and on evaluating whether AI output can be trusted.

That background shows up in how Ndoc is organized. Rather than bolting an assistant onto Excel or Word, the product puts three views in one layout:

- The Stage tab holds presentations of key findings — slide decks, dashboards, simulations or small apps. - The Narrative tab holds the analysis: numbered citations, generated figures and the reasoning behind the numbers. - The Model tab holds the underlying datasets and the Python scripts that produce the output.

A user can upload a dataset, ask Ndoc to run an analysis and write a narrative about what it found, and then inspect the script behind any figure. Patel's point is auditability: even without knowing Python, a user can ask the built-in chat to explain what a script does and which line implements a given formula. He also says the tool keeps memory across a person, a single document, a team and the whole company, and that users can specify context such as audience and tone per document.

On the model side, Patel says Ndoc is built and fine-tuned on a mixture of open-source models rather than frontier APIs, and that this makes it considerably cheaper to run than alternatives built on frontier models. He told TechCrunch that early trial users were indifferent to that cost advantage; what excited them was having the whole workflow in one place. This is a familiar pattern in developer tools — infrastructure savings are real but rarely the reason someone switches, and a cheaper stack can become a constraint if the open-source models lag on hard reasoning tasks. That trade-off is my reading, not something the article establishes.

The bigger ambition is the file format. Ndoc documents can be exported 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 being locked into Ndoc the way he says companies have been locked into Microsoft or Google. He partly cites that goal as a reason for leaving: inside a large company, agreeing on an open format would involve hundreds of people and opinions, while he wants to move at the pace AI allows. An open HTML-based container is a plausible route to interoperability, since browsers already render it; whether it becomes a real standard depends on outside adoption, which a three-person team cannot compel. That is inference on my part.

Patel is entering a crowded field. OpenAI, Anthropic, Meta, SpaceXAI and Google have all released enterprise AI tools that connect to Google Workspace, Microsoft 365 and other workplace apps so their models operate inside tools people already use. Last month OpenAI launched Space, a shared workspace inside ChatGPT that competes more directly with established productivity suites. Patel acknowledges Space could be a threat but argues the labs have the wrong approach, comparing filling in spreadsheet cells with AI to putting lights on the oxen that pull a plow after inventing electricity — his view being that the better move is to build the tractor. The counterargument, which the article notes, is that habits and whole workflows are hard to replace, which is why the labs go where the work already happens. Patel says he understands this and has built connectors so customers can import data from existing tools.

His stated standard is personal: he will not release something he does not love using, and he keeps a "graveyard" of tools that failed that test. He also wants proof that people love the product before raising money. The competitive reality is that switching costs run both ways — incumbents benefit from entrenched habits, while a newcomer has to be dramatically better, not merely different, to pull teams off suites they already pay for.

Why it matters: Product teams weighing an AI layer on top of Office or Workspace now have a third design point to compare against — one document holding the dataset, the script, the narrative and the presentation — and the planned open-source ".ndoc.html" format is the piece developers should watch, since it would let other tools read Ndoc output without a proprietary lock-in. The alpha is waitlisted and paid by usage, and with three people and no funding, the near-term risk for anyone adopting early is durability as much as capability, which is my inference rather than a stated concern. Teams should treat the current release as a pilot to evaluate, not a suite replacement.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Rohit Patel is departing Meta after eight years working on AI evaluations and reinforcement learning, TechCrunch has exclusively learned. His startup, Ndoc, aims to take on Microsoft Office Suite.