Hajra Naz
Rohit Patel is departing Meta after spending eight years building AI solutions for the tech giant. The former Meta Superintelligence Labs director launched Ndoc, a new startup. He pitches the platform as an AI-native replacement for Microsoft Office and Google Workspace.
Patel represents the latest high-profile departure from Meta. His exit follows Chief Global Affairs Officer Joel Kaplan, AI researcher Andrew Tulloch, and Head of Product for AI-for-Work Emily Dalton Smith.
Patel designed Ndoc to go beyond basic AI features layered onto old workflows. His goal is to replace disconnected tools like Excel, Word, PowerPoint, Tableau, Replit, and manual ChatGPT pasting. Ndoc replaces them with one unified workspace. All colleagues can access shared projects even without active Ndoc accounts.
Ndoc combines three core components inside a single user interface. The Stage tab hosts final presentation displays like slide decks, data dashboards, simulations, or mini applications. The Narrative tab provides written analysis, numbered references, generated charts, and the context behind raw data. The Model tab contains underlying source datasets alongside active Python code.
Read More: Microsoft offers buyouts to around 60,000 workers
This structural design reflects Patel’s technical background. During his time at Meta, Patel focused heavily on reinforcement learning systems. That field of machine learning helps AI agents make decisions through trial and error to maximize results. He also focused on verifying if AI outputs are accurate and trustworthy. Patel believes humans must verify AI work if models generate corporate spreadsheets and executive slides. Users can inspect the exact Python scripts used to generate every file.
“Even without Python knowledge, you can ask the AI chat to explain the code,” Patel noted. “You can ask which line holds your formula and verify it directly. Nothing stays hidden behind the curtain.”
In practice, users upload a dataset and request an instant analysis with written takeaways. They can also input document context, including intended audience, tone, and system memory. “It preserves memory across your profile, that specific document, your team, and your broader organization,” Patel explained.
Patel explained that Ndoc runs on fine-tuned open-source models. Building the system from scratch keeps operating costs far lower than tools using third-party frontier AI.
“I expected users to get excited about low costs, but they did not care about that,” Patel said, reflecting on early test feedback. “Users care about seeing their end-to-end workflow in a single workspace.”
Read More: Google Docs and Drive Let You Edit Markdown Files Without Converting
Patel aims to reinvent the file format itself. Documents export as custom “.ndoc.html” files that open inside standard web browsers. He plans to open-source this file type. That move allows outside developers to build compatible apps without vendor lock-in to Microsoft or Google.
His desire to open-source the project drove his decision to leave Meta. At a large corporation, agreeing on an open-source standard requires input from hundreds of employees and conflicting opinions.
Patel wanted to execute rapidly. “Humans move at a human pace, but I can move at the speed of modern AI development,” he remarked.
Replacing established workplace platforms is a massive gamble for a founder leading a three-person team without venture backing. Other major AI organizations are competing for the same market share.
OpenAI, Anthropic, Meta, SpaceXAI, and Google have released enterprise AI tools. Those integrations connect models directly to Google Workspace, Microsoft 365, and existing productivity software. OpenAI recently launched Space, a shared workspace inside ChatGPT that directly targets traditional software tools.
Patel acknowledges that Space poses competition, but he believes major AI labs misunderstand user needs.
“Companies attach AI to traditional spreadsheets to auto-fill cells,” he said. “It kills my soul. That resembles discovering electricity and using it to put lights on working bulls. You should buy a tractor instead.”
AI labs target existing tools because changing user habits remains difficult. Patel understands this challenge, so he created connectors to import data from existing software.
“My standard was not releasing anything I do not personally love using,” he stated. “I maintain highly optimized AI workflows. If I keep returning to old routines while testing Ndoc, I will fail to build a good enough tool. I maintain a graveyard of failed tools that missed that bar.”
Read More: Google Pauses Open Source Bug Bounty Program After AI Reports Surge
Patel sets a similar standard for customer adoption before raising venture capital. “I refused outside capital until proving people genuinely love the product,” he explained.
Ndoc is running an enterprise pilot and opening a limited alpha waitlist for early customers. Registration is free, though users pay for resource usage and token consumption. He plans to launch a public beta in the coming weeks before a full release next year.
For Patel, execution speed matters far more than launching a perfect product initially.
“Execution is extremely cheap, so execute a suboptimal plan and worry later,” he said. “By the time you draft a perfect plan, someone else will finish executing it.”
