Rohit Patel's departure from Meta to build Ndoc is the clearest signal yet that the office suite is no longer a fortress, it's a target. After eight years refining AI evaluations and reinforcement learning, Patel isn't leaving to make a better word processor. He's leaving to rebuild the underlying logic of how documents, spreadsheets, and presentations interact with human intent. That is a meaningful distinction, and it should change how you think about your daily tools.
The timing matters. We've been watching AI move from text-based reasoning into more complex domains, whether that's AI That Reasons in Motion, Depth, and Words Expands Beyond Text or the push toward simulated environments for smarter models. Those developments share a common thread: AI is getting better at understanding structure, sequence, and consequence. An office suite is exactly that, a structured environment where actions produce outcomes. Patel's expertise in reinforcement learning is not incidental. It's the core competency required to build software that doesn't just store your data but actively reasons about it.
Microsoft Office has dominated for decades because it became the default grammar of professional work. But default is not the same as optimal. The suite is powerful, yet it remains fundamentally passive. It waits for you to know what you want, then helps you execute it. Ndoc's implied ambition is to flip that relationship. Instead of you interrogating a spreadsheet, the spreadsheet interrogates itself, spots inconsistencies, suggests next steps, and drafts the memo that summarizes its own findings. That is not a feature update. That is a different category of tool.
We've seen the raw potential of this approach in research contexts. An MIT expert recently unpacked how AI models proved something significant about a major math theory, work that required models to hold and manipulate abstract relationships over long chains of reasoning (An MIT expert unpacks what AI models just proved about a major math theory). If that same capacity for sustained logical inference gets applied to your quarterly budget or your project timeline, the productivity gains are not incremental. They are structural.
Our take is straightforward: the office suite is the next real battlefield for AI, and Patel is betting that the winning move is to start from scratch rather than bolt intelligence onto legacy architecture. That bet has a specific consequence you should watch closely. If Ndoc succeeds, the metric will not be how many features it matches from Microsoft Office. The metric will be how much of your thinking it can handle before you even open the file. The open question is whether users will accept a tool that sometimes does the work differently, or even better, than they would have done it themselves. That is the adoption hurdle, and it is far higher than any technical one. Watch for how Ndoc handles that friction, because that, not the feature list, will determine whether Patel's reimagining becomes the new standard or a fascinating footnote.