The archive · AI & Models · Strategic decision · 2025–2026
Humans& bets coordination is AI's next frontier: $480M seed at $4.48B
Three-month-old ex-OpenAI/Anthropic/xAI lab raises $480M seed at $4.48B to build a coordination-focused foundation model, before shipping a product.
Humans&
What the business is
Humans& is a San Francisco 'human-centric frontier AI lab' building a foundation model and product for communication and collaboration — an AI layer that coordinates people, teams and other AI agents.
Starting capital：$480M seed at $4.48B valuation (Jan 2026)
How it started
Founded around October 2025 by five AI researchers — Eric Zelikman (ex-xAI, creator of the STaR reasoning algorithm, worked on Grok 2–4), Andi Peng (ex-Anthropic, post-training on Claude 3.5–4.5), Georges Harik (Google's seventh employee), Yuchen He (ex-xAI/OpenAI) and Noah Goodman (Stanford professor) — plus about 20 staff from OpenAI, Meta, Reflection, AI2 and MIT.
What happened
In January 2026, three months in, it raised $480M in seed at a $4.48B valuation, led by SV Angel and co-founder Georges Harik, with Nvidia, Jeff Bezos, GV, Emerson Collective, Forerunner and S32. The company says it has already turned away acquisition interest and plans its first product for later in 2026; training will use long-horizon and multi-agent reinforcement learning plus memory research.
No ending yet — it is still running.
Background
Humans& is a San Francisco 'human-centric frontier AI lab' founded around October 2025 by AI researchers from Anthropic, OpenAI, xAI, Google and Meta, including ex-xAI researcher Eric Zelikman (CEO), former Anthropic researcher Andi Peng, Google's seventh employee Georges Harik, former OpenAI researcher Yuchen He and Stanford professor Noah Goodman. Its thesis is that the first scaling paradigm produced question-answering models, but the next frontier is coordination — managing teams, long-running decisions and competing priorities, which today's chatbots are not built for.
In January 2026, three months after founding, the company raised a $480M seed round at a $4.48B valuation, led by SV Angel and co-founder Georges Harik, with Nvidia, Jeff Bezos, GV, Emerson Collective, Forerunner and S32 participating. It employs about 20 people drawn from OpenAI, Meta, Reflection, AI2 and MIT.
The company describes the product as a 'central nervous system' for the human-plus-AI economy: a model plus product for communication and collaboration — effectively an AI-native replacement for messaging and collaboration tools — trained with long-horizon and multi-agent reinforcement learning, memory and user understanding. It says it has already turned away acquisition interest and plans its first product later in 2026.
The bet is unproven and expensive: training a new foundation model requires capital and compute that put it in direct competition with Anthropic's Claude Cowork, Google's Gemini-in-Workspace and OpenAI's multi-agent orchestration — none of which are rebuilding models around social intelligence, which is either Humans&'s opening or its exposure.
What has to be true
- It attacked an unclaimed position — coordination and social intelligence — instead of re-running the question-answering race, backed by a star founding team from Anthropic, xAI, OpenAI and Google.
- The mega-seed with Nvidia, Jeff Bezos and GV shows investors are pricing a thesis, not a product: three-month-old labs can command billions when the narrative is the next platform shift.
- Designing the product and the model together, co-evolving the interface with model behavior, is a credible answer to the one-user ceiling most labs ignore.
- Refusing acquisition and pre-announcing the collaboration-layer ambition tells incumbents they plan to compete for the layer Slack, Notion and Google Docs occupy.
What can be applied
When compute and talent are commodities, a contrarian thesis can still raise enormous capital — but a pre-product lab at $4.48B bets model capability, not go-to-market, decides the next platform.
Aftermath
As of September 2026 Humans& remains in building mode: no product had shipped by the time of its January coverage, with a first product planned for later in 2026. The company says it is not interested in being acquired and has turned away interested parties. Its public materials commit to 'innovations in long-horizon and multi-agent reinforcement learning, memory and user understanding,' and the risk profile is unchanged — it must fund frontier-model training while competing with Anthropic (Claude Cowork), Google (Gemini in Workspace) and OpenAI on the collaboration layer.
Sources
- Humans& thinks coordination is the next frontier for AI, and they're building a model to prove it
- Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round
- Humans& Raises $480M to Build Human-Centric AI Tools
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card