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

The Great Unbundling of the Office Suite

I have seen this movie before, but the cast is changing. Rohit Patel, a man who has spent the last eight years at Meta sharpening the tools that make AI behave, is walking away. He is not leaving for another safe director role at a rival big tech firm. He is going after the king of all moats: the productivity suite.

His new venture, Ndoc, is aiming for the throat of Microsoft Office and Google Workspace. For a founder, this is either the balliest move in the current cycle or a sign that the incumbents are finally vulnerable. Patel has been at the center of reinforcement learning and AI evaluation at Meta. He knows exactly how these models fail and, more importantly, how they can be steered to be useful.

Why Productivity Software is Stale

We have been using essentially the same document and spreadsheet mental models since the 1990s. Sure, Google brought us into the cloud and collaborative editing, and Microsoft eventually caught up with Office 365, but the underlying structure is archaic. It is a digital version of paper.

Most builders today realize that AI isn't just a feature you sprinkle on top of a Word document. Adding a chatbot to the side of a spreadsheet is a band-aid, not a revolution. Patel likely saw that from the inside of Meta. When you are at a company that large, you see the limitations of trying to retro-fit legacy systems with modern intelligence.

The Founder's Gamble

Taking on Microsoft is usually a death sentence. They own the distribution. They own the IT departments. They own the enterprise seat. But there is a crack in the armor. As teams shift from 'writing documents' to 'generating workflows,' the old file-based system starts to feel heavy.

If Ndoc is building from a builder-first perspective, they aren't just making a better text editor. They are likely building a data-first environment where the AI isn't an assistant, but the core engine. For those of us in the crypto and AI space, we know that the biggest friction is context switching. If Patel can solve the context problem—making the software understand the project rather than just the paragraph—he has a fighting chance.

The Challenge of AI Evaluation

Patel’s background at Meta wasn't just in making AI 'smart,' it was in evaluation. This is a boring topic to most, but it’s the most important one for builders. How do you know the AI is actually helping and not just hallucinating polished-sounding nonsense?

Current productivity tools suffer from a trust gap. You ask an AI to summarize a meeting or analyze a budget, and you still have to spend twenty minutes double-checking the work because you don't trust the output. If Ndoc can leverage Patel’s expertise in reinforcement learning to create a 'verifiable' productivity environment, they might actually steal some market share from the giants.

What This Means for the Ecosystem

We are entering a phase where the best talent is leaving 'Big AI' to build 'Applied AI.' The research phase is maturing, and now the builders are getting restless. They don't want to just train Llama 4; they want to build the tools that people use to run their businesses.

• Incumbents are slow: Even with billions in R&D, Microsoft and Google are burdened by their own success. They cannot break their user interfaces without risking millions of enterprise users.
• The 'LLM-Native' Advantage: New startups like Ndoc don't have to support legacy .doc files or backwards compatibility. They can build for how work happens in 2026.
• Evaluation is the new UX: The winner won't have the prettiest buttons; they will have the most reliable intelligence.

A Skeptical Lens

I have to be honest: the odds are stacked against him. We have seen Notion, Coda, and many others try to chip away at the Microsoft/Google duopoly. While they’ve found success in the startup niche, the vast majority of the world still runs on Excel.

Patel's departure is a signal that the 'safety' of a director role at Meta is no longer as appealing as the potential of the AI-native productivity market. For founders watching this, the takeaway is clear: the infrastructure is mostly built. Now is the time to build the applications that actually change the way we work.

The biggest moat Microsoft has isn't their software; it's the fact that nobody ever got fired for buying Office. Patel has to build something so much better that the risk of switching becomes a necessity.

The Takeaway for Builders

Don't just build 'wrappers.' If you're going to compete with the giants, you have to find the one thing they can't do because of their size. In this case, it's rebuilding the document from the ground up with AI at the center, not as a sidebar. Patel is betting eight years of Meta equity on that premise. It’s a bold play, and it’s exactly the kind of move this stagnant productivity market needs.

Read the original at TechCrunch Startups →