Meta AI Director Rohit Patel Launches Ndoc as Alternative to Microsoft Office

Rohit Patel is leaving Meta after spending eight years developing AI at the company. Patel, a former director at Meta Superintelligence Labs, has launched Ndoc and pitches the startup as an AI-native alternative to Microsoft Office or Google Workspace. Patel's departure follows those of Chief Global Affairs Officer Joel Kaplan, AI researcher Andrew Tulloch and Head of Product for AI-for-Work Emily Dalton Smith.

Patel aims to combine work across Excel, Word, PowerPoint, Tableau, Replit and ChatGPT into one workspace accessible to colleagues without Ndoc accounts. Ndoc organizes work into three tabs within a single layout. The Stage tab displays findings through slide decks, dashboards, simulations or mini apps. The Narrative tab contains analysis, numbered citations and generated figures that explain the results. The Model tab stores the datasets and Python scripts behind the work.

Patel's work at Meta concentrated on reinforcement learning and assessing the reliability of AI output. Reinforcement learning trains AI agents to make decisions through trial and error while maximizing rewards. Patel believes people should be able to verify AI-generated spreadsheets and presentations used for business decisions. Users can inspect the Python scripts behind documents, and Patel says users can ask the AI chat to explain scripts and identify formulas for verification. Users can upload datasets and request analysis accompanied by a written explanation of the findings. Users can also supply context about a document's audience, tone and memory. Patel says Ndoc retains memory at the individual, document, team and company levels.

Patel says Ndoc uses a combination of open-source models that the startup has fine-tuned. Patel says developing the platform from scratch makes operating costs considerably lower than those of alternatives using frontier AI models. Patel says early trial users were more interested in having the entire workflow together than in the lower operating costs. Patel says users can export documents into a single file with the extension ".ndoc.html" that opens in any browser. Patel plans to make the format open source so developers can create compatible applications without dependence on Ndoc. Patel's wish to open source the format partly motivated the decision to leave Meta and work independently. Patel believed agreement on an open-source format at Meta would require input from hundreds of people. Patel wanted to work at the speed AI enables rather than at the pace of human coordination.

Ndoc has a team of three and no funding, making competition for enterprise workflows a major challenge. OpenAI, Anthropic, Meta, SpaceXAI and Google have introduced enterprise AI tools that connect models to Google Workspace, Microsoft 365 and other workplace applications. OpenAI launched Space last month as a shared workspace within ChatGPT that competes more directly with established workplace tools. Patel acknowledges that Space could threaten Ndoc but argues that the AI labs largely misunderstand the opportunity. Patel criticizes companies for adding AI to existing spreadsheets primarily to fill cells rather than fundamentally changing how work happens. Changing established habits and replacing complete workflows is difficult. Patel says Ndoc includes connectors that let customers import data from existing tools because Patel recognizes that difficulty.

Patel says a product must be useful enough to replace Patel's own optimized AI workflows before Patel will release the product. Patel says numerous earlier tools failed that test. Patel says customer enthusiasm must provide evidence of demand before Patel seeks funding. As of Tuesday, Ndoc is conducting an enterprise pilot and opening an alpha version to a limited group of customers joining the wait list. Customers can register without charge but will pay for usage and tokens. Patel hopes to introduce a beta version in the coming weeks. Ndoc plans a full launch next year.

Patel prioritizes rapid execution over delivering a perfect initial product. Patel argues that execution is now extremely inexpensive and that acting on an imperfect plan is preferable to waiting while competitors finish the work.