Most AI chatbots are built around a simple exchange: one person asks, the system answers. Humans& is betting that the next major AI challenge is not just better answers, but better coordination among people, teams, and AI systems.
The three-month-old startup, founded by alumni of Anthropic, Meta, OpenAI, xAI, and Google DeepMind, has raised a $480 million seed round. Its goal is to build what it describes as a central nervous system for the human-plus-AI economy, with a foundation model designed for social intelligence rather than only information retrieval or code generation.
Why Coordination Is the Target
AI tools have become stronger at answering questions, summarizing documents, and solving mathematical equations. But the source article makes clear that many still operate like assistants for a single user, not systems built for the complex work of collaboration.
That distinction matters because real work often involves competing priorities, long-running decisions, and groups that need to stay aligned over time. Humans& sees that gap as the next frontier for foundation models.
The company’s public framing emphasizes AI for empowering humans. Behind that message is a more specific technical ambition: build a model architecture that can help people communicate, coordinate, and make decisions together.
That puts Humans& in a different category from a chatbot that simply plugs into an existing workplace tool. The startup wants to own the collaboration layer itself.
A Product That Is Still Taking Shape
Humans& does not yet have a product, and the company has not clearly defined what that product will be. The team has suggested it could replace multiplayer or multi-user contexts such as communication platforms like Slack or collaboration platforms like Google Docs and Notion.
The possible audience is also broad. The company has hinted at both enterprise and consumer uses, which means its target could range from workplace teams to families.
What is clear is the direction of the work. Co-founder and CEO Eric Zelikman, a former xAI researcher, told TechCrunch that Humans& is building a product and model centered on communication and collaboration. The aim is to help people work together more effectively with each other and with AI tools.
The startup’s own experience shows the kind of problem it wants to solve. Zelikman described the difficulty of getting a large group to agree on something as ordinary as a logo. That example is small, but it points to a familiar coordination burden: people may have different preferences, and someone has to collect, compare, and resolve them.
Training for Social Intelligence
Humans& says its model will need to behave differently from today’s question-answering systems. Zelikman argued that current chatbots often ask questions without understanding the value of the question itself.
The company wants a model that asks questions more like a friend or colleague who is trying to understand the user. In that framing, a useful AI system would not only respond accurately. It would learn what matters to each person, remember context, and help balance individual needs with the needs of a group.
Yuchen He, a co-founder and former OpenAI researcher, told TechCrunch that Humans& plans to train the model through more interaction and collaboration between humans and AIs. The model will also use long-horizon and multi-agent reinforcement learning.
In plain terms, the training focus is about time and multiple participants:
- Long-horizon reinforcement learning is intended to help a model plan, act, revise, and follow through over time.
- Multi-agent reinforcement learning focuses on settings where several AIs and/or humans are involved.
Those ideas fit the company’s broader thesis. If AI is going to support coordination, it needs to do more than generate a strong one-off response. It needs to handle context that stretches across people, tools, and decisions.
He also emphasized memory and user understanding. The better the model remembers information about itself and about the user, the more effectively it can understand what is needed.
The Competitive Pressure Is Obvious
The opportunity is large, but the risks are also clear. Training and scaling a new model is expensive, and Humans& will need significant resources to compete for compute and talent.
The company is not only moving near workplace software such as Notion and Slack. It is also entering territory where major AI companies are already active. Anthropic has Claude Cowork, Gemini is embedded into Workspace, and OpenAI has been pitching developers on multi-agent orchestration and workflows.
AI collaboration and productivity tools are also attracting capital elsewhere. The startup AI note-taking app Granola raised a $43 million round at a $250 million valuation as it launched more collaborative features.
Several prominent voices are also presenting the next phase of AI as coordination rather than simple automation. LinkedIn founder Reid Hoffman argued that companies are implementing AI incorrectly when they treat it as isolated pilots, and that the real leverage is in how teams share knowledge and run meetings.
For Humans&, that makes timing both favorable and difficult. The market appears interested in AI workflow and collaboration, but the strongest AI companies already have distribution, products, and research teams aimed at related problems.
What Would Make Humans& Different
The startup’s strongest differentiator is its claim that social intelligence should be built into the model itself. The source article notes that major players do not appear poised to rewrite a model around that idea.
That could give Humans& room to build something distinct. It could also make the company an acquisition target, especially as Meta, OpenAI, and DeepMind seek top AI talent. Humans& told TechCrunch it has already turned away interested parties and is not interested in being acquired.
The central question is whether a coordination-first AI model can become a real product before larger platforms absorb the same need through existing tools. Humans& is making a large bet that collaboration is not just a feature to add on top of AI, but a foundation-level problem that requires a new model and a new interface together.
If that bet is right, the next wave of AI may be judged less by how well a system answers one person’s question and more by how well it helps many people decide what to do next.