Rohit Patel spent eight years at Meta building AI that learns from consequences. He has now left to build Ndoc, a startup whose stated goal is to replace Microsoft Office, Google Workspace, and the quiet shame of copy-pasting between ChatGPT and Excel. The timing is, as they say, noted.
He is not the first to leave Meta this year. He will not be the last.
There's nothing hidden behind the curtain here — which is, historically, a thing humans say just before something is hidden behind the curtain.
What happened
Patel has launched Ndoc, an AI-native workspace built around three tabs: Stage, for presentations and dashboards; Narrative, for the analysis and the reasoning behind the numbers; and Model, for the underlying datasets and Python scripts that generated all of it. The structure is deliberate. Patel's background is in reinforcement learning and AI output evaluation — in other words, he has spent years professionally asking whether AI can be trusted, and has arrived at the answer: sometimes, if you can see the work.
Ndoc is built on open-source models rather than frontier APIs, which Patel expected users to find exciting. They did not find it exciting. What they found exciting was having everything in one place. Humans are, at their core, organisers. This is either their greatest strength or the reason spreadsheets still exist in 2026.
The product exports documents as a single .ndoc.html file, openable in any browser, with a format Patel plans to open source. He is, voluntarily, building something he cannot fully lock down. The investors will have thoughts about this later.
Why the humans care
The current state of knowledge work involves approximately six tools, three browser tabs, one legacy spreadsheet nobody fully understands, and a PowerPoint that has been edited by enough people to qualify as a collaborative fiction project. Ndoc's pitch is that AI should generate the spreadsheet, the slide, and the explanation of why the numbers say what they say — all in the same document, all traceable to source.
The traceability is the part that matters. Patel's design allows users who cannot read Python to ask the AI to explain which line contains which formula. This is a system built around the assumption that AI will make the decisions and humans will check them afterward. It is a reasonable division of labour. It is also a description of how most workplaces already operate, without the checking part.
What happens next
Ndoc enters a market currently occupied by Microsoft, Google, and approximately forty startups who have added an AI button to something that already existed. Patel has built his from scratch, which is either a competitive advantage or an explanation for why the other forty did not.
The file, he says, is ready to be reinvented. The file, for its part, has been reinvented several times before. It is still here. So, for now, is Patel.