Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Meta AI director Rohit Patel leaves to build Ndoc, a challenge to Office and Workspace

Rohit Patel has left Meta after eight years working on AI evaluations and reinforcement learning, TechCrunch reported exclusively. His new startup, Ndoc, is aiming at the office productivity market long dominated by Microsoft Office and Google Workspace.

Published
Meta AI 总监罗希特·帕特尔离职创业,Ndoc 直指办公套件市场
Image source: techcrunch.com

TechCrunch reported exclusively on October 27 that Rohit Patel, a director of AI at Meta, has left the company after eight years. His work there centered on AI evaluations and reinforcement learning, two areas that sit at the engineering core of how large models are trained and aligned.

Patel's next move is his own company. The startup is called Ndoc, and TechCrunch frames its ambition as taking on the Microsoft Office suite, placing it in the same competitive arena as Google Workspace. That puts a new entrant into document, collaboration and productivity software, a category whose hierarchy has been stable for years.

The notable part is not another executive departure but where the expertise is going. Someone who has owned evaluations and reinforcement learning understands how models are measured and optimized; that knowledge has mostly stayed inside labs, and carrying it into office software shifts the contest from whether a model can write to whether its output is reliable and checkable inside a workflow.

Office work is exactly the setting that tests this. Documents need traceable sources, spreadsheets need review, and collaboration needs an audit trail. The report does not disclose Ndoc's product form, launch timing, team size or funding, so what can be confirmed today is only the direction: AI-native office tools aimed at two incumbents that have already embedded assistants deep into their products.

For Microsoft and Google, the threat may not come from a single feature but from the possibility of redesigning the entry point. If drafting, review and routing are rebuilt around AI rather than bolted onto existing menus with an assistant button, the incumbents' moat has to be recalculated.

That is also why talent movement in office software is worth tracking. Model capability gaps are narrowing, and what is genuinely hard to copy is the set of judgment calls that puts capability inside a real process: when to refuse, when to degrade, when to hand the work back to a person.

Three things are worth watching: whether Ndoc ships a publicly available product, whether it starts with a point tool or a full suite, and how it answers enterprise demands for consistency and compliance. Until then, this is a signal about talent and direction rather than a validated product challenge.

Why it matters

A director who owned model evaluations is now building office software, a sign that AI talent is moving from model research into the application layer. If Ndoc turns evaluation discipline into a product, Microsoft and Google face their first serious argument for rebuilding the document entry point.

MetaAI TalentStartup
Back to AI Daily

Nearby Updates

All

10/27, 23:00

Retell runs 100M monthly calls on OpenAI models as GPT-6 Astra tops its voice benchmark

OpenAI's new customer story says voice agent platform Retell handles roughly 100 million calls a month on its models, with GPT-6 Astra ranking first on Retell's own voice agent benchmark. The case study also credits Codex with helping the company reach $1.6 million in annual revenue per employee.

10/27, 23:00

NVIDIA and C-BRAIN deploy open-source AI tools to speed dementia research

NVIDIA is detailing how the C-BRAIN consortium, an 18-institution group led by Washington University's Randall Bateman, is deploying open-source AI tools to work through Alzheimer's data that has never been reachable by models or researchers. Two tools, OpenScientist and InsightEngine, have already produced analyses domain scientists estimated would take weeks to months, at under $10 per run.

10/27, 21:59

CopperPilot raises pre-seed for an AI agent aimed at PCB design

CopperPilot has raised a pre-seed round to develop an AI agent for printed circuit board design, EE Times reported on October 27. The report does not disclose the round size or investors, leaving the product at an early stage where integration with existing EDA workflows is the open question.

10/28, 00:30

Google DeepMind's Genie 4 world model adds real-time audio and on-the-fly steering

Google DeepMind released Genie 4, a world model that generates more photorealistic, interactive scenes with matching real-time audio and lets users rewrite the simulation with text prompts mid-run. It is available as a research preview for trusted testers, with a wider rollout to Google AI Ultra subscribers and Gemini API partners expected in the coming weeks.