Meta AI Director Exits to Challenge Microsoft and Google in Enterprise Productivity

Rohit Patel spent eight years building AI evaluation and reinforcement learning systems inside Meta. Now he's walking away from that stability to bet on a three-person startup with no outside funding. The thesis: enterprise software's biggest incumbents, Microsoft Office and Google Workspace, are vulnerable to a ground-up AI rebuild rather than an AI feature bolted on top.

Patel, formerly a director at Meta Superintelligence Labs, has launched Ndoc, which TechCrunch first reported exclusively. He's not alone in leaving. Chief Global Affairs Officer Joel Kaplan, AI researcher Andrew Tulloch, and Head of Product for AI-for-Work Emily Dalton Smith have all exited Meta in recent months. That pattern matters for anyone tracking talent retention at the big labs, though Patel's departure is distinct in one respect: he's not joining a rival lab, he's founding a direct competitor to the tools those labs are racing to augment.

The Product Bet

Ndoc's pitch is structural rather than cosmetic. Instead of layering a chatbot onto Excel or Word, Patel designed three integrated tabs: Stage, for finished outputs like slide decks and dashboards; Narrative, for the analytical "why" behind the numbers, complete with citations; and Model, where the underlying datasets and Python scripts live in full view.

That transparency is deliberate. Patel's Meta work centered on verifying whether AI output could be trusted, and he's carried that discipline into the product. Users who don't know Python can still ask the embedded AI chat to explain any line of code or formula. "There's nothing hidden behind the curtain here," he said. Given how often AI-generated financial models get deployed without scrutiny inside companies, that auditability claim is worth testing rather than taking on faith.

The platform runs on fine-tuned open source models rather than frontier systems from the major labs, which Patel says lowers operating costs considerably. Interestingly, that cost advantage hasn't been the draw for early users. "I thought that's what people would be excited about, but people don't care about that," Patel said. What resonates instead is consolidation: pulling Excel, Word, PowerPoint, Tableau, Replit, and ad hoc ChatGPT sessions into one workspace that colleagues can view even without an Ndoc account.

Patel's longer-term ambition is a file format, not just an app. Documents export as a single ".ndoc.html" file, viewable in any browser, which he intends to open-source so other developers can build on it without the lock-in that has defined Microsoft and Google's grip on enterprise workflows for decades. He says that open-source ambition was a direct factor in leaving Meta. Coordinating a format decision at a company that size would require "hundreds of people and opinions," he said, and he wanted to move at the pace of the technology rather than the pace of internal consensus.

The competitive field here is not empty, and that's the central tension in this story. OpenAI, Anthropic, Meta, xAI, and Google have all shipped enterprise AI tools that connect directly into existing Workspace and Microsoft 365 environments. OpenAI went further last month with the launch of Space, a shared workspace inside ChatGPT that TechCrunch described as functioning like ChatGPT's own Office suite, putting it on a more direct collision course with Ndoc's ambitions than any of the connector-based tools.

Patel views Space as a threat but argues the entire category is solving the wrong problem. "Companies are basically attaching AI to spreadsheets and it fills your cells for you," he said. "It just sort of kills my soul. This is like if you just figured out electricity, and then you decided the best use case is to put lights on my bulls that till my soil. No. Get a tractor." It's a colorful line, but it also reveals the bet he's making: that incremental AI features inside legacy tools will eventually lose to a workspace built natively around AI reasoning and verification.

He's not ignoring the switching-cost problem, though. Ndoc includes connectors so customers can import data from the tools they already rely on, an acknowledgment that habits are sticky and that a pure rip-and-replace pitch rarely wins enterprise deals on its own.

Patel's go-to-market discipline is notably conservative for a founder in a hype-driven category. He declined to raise capital before securing proof that customers actually preferred Ndoc to their existing workflows. "I didn't want to take money until I had some proof that people love the product," he said. As of this week, Ndoc is running an enterprise pilot and has opened an alpha to a waitlist, free to join, with usage and token-based pricing to follow. A beta is expected in the coming weeks, with full launch targeted for next year.

The Bottom Line

Ndoc is an unfunded, three-person operation entering a category where OpenAI, Anthropic, Google, Meta, and xAI have already committed significant capital and distribution advantages. Patel's technical credibility on AI verification is real, and his cost structure using open source models is a genuine differentiator worth tracking. But the lack of funding, the absence of a committed enterprise customer base, and the sheer scale of incumbent resources mean this remains a high-variance bet rather than an emerging market leader. Investors and enterprise buyers should watch whether the alpha pilot converts into paying logos before treating Ndoc as anything more than an interesting experiment. The file-format open-sourcing plan, if it gains developer adoption independent of Ndoc's own success, could prove the more durable asset regardless of how the core product fares against entrenched competitors.