A new AI startup is betting the next frontier isn't smarter chatbots - it's getting people to actually work together. Humans&, founded by alumni from Anthropic, Meta, OpenAI, xAI, and Google DeepMind, just raised a $480 million seed round to build what it calls a "central nervous system" for human-AI collaboration. The three-month-old company doesn't have a product yet, but it's already making waves with a bold premise: that AI needs to be redesigned from scratch for social intelligence, not just question-answering.
AI chatbots keep getting smarter at answering your questions, but they're still terrible at helping groups of people make decisions together. Humans& thinks that's the real problem worth solving.
The startup just closed a $480 million seed round - one of the largest in AI history - to build what CEO Eric Zelikman calls the "connective tissue" for organizations. The pitch is straightforward: today's AI models are designed for one person at a time, optimized to give you the answer you want. But real work happens in messy, multiplayer contexts where people have competing priorities, long-running projects, and the need to stay aligned over weeks or months.
"It feels like we're ending the first paradigm of scaling, where question-answering models were trained to be very smart at particular verticals," Andi Peng, co-founder and former Anthropic employee, told TechCrunch. "Now we're entering what we believe to be the second wave of adoption where the average consumer or user is trying to figure out what to do with all these things."
The company doesn't have a product yet. It's only three months old. But the founding team's pedigree - alumni from Anthropic, Meta, OpenAI, xAI, and Google DeepMind - was enough to command a valuation that rivals some late-stage startups. The bet investors are making is that Humans& can crack a problem the AI giants haven't prioritized: building models trained specifically for coordination, not completion.
Zelikman, who previously worked at xAI, described the vision through a familiar scenario. "Like when you have to make a large group decision, often it comes down to someone taking everyone into one room, getting everyone to express their different camps about, for example, what kind of logo they'd like," he said, laughing with his co-founders about the hours they spent aligning on their own startup's branding.
That's the kind of workflow Humans& wants to automate - or at least streamline. The company hinted its product could eventually replace tools like Slack, Google Docs, or Notion in multi-user contexts, though it's designing the interface and model in parallel. "Part of what we're doing here is making sure that as the model improves, we're able to co-evolve the interface and the behaviors that the model is capable of into a product that makes sense," Peng said.
What sets Humans& apart - at least on paper - is its training approach. Co-founder Yuchen He, a former OpenAI researcher, told TechCrunch the team is using long-horizon and multi-agent reinforcement learning to train models that can plan, revise, and follow through over time. Instead of optimizing for immediate user satisfaction or one-off correctness, the model is being trained to balance the needs of multiple people and remember context across interactions.
"The model needs to remember things about itself, about you, and the better its memory, the better its user understanding," He explained. That's a departure from how most large language models work today, where each conversation is largely siloed and optimized for short-term feedback.
The timing is notable. LinkedIn founder Reid Hoffman argued this week that companies are implementing AI wrong by treating it like isolated pilots. "AI lives at the workflow level, and the people closest to the work know where the friction actually is," Hoffman wrote on social media. "They're the ones who will discover what should be automated, compressed, or totally redesigned."
That's the exact layer Humans& is targeting. The company wants its model to act as the central nervous system for any organization - whether that's a 10,000-person enterprise or a family group chat - understanding each person's skills, motivations, and needs, then balancing them for collective outcomes.
But the competitive landscape is brutal. Anthropic is building Claude Cowork for collaborative workflows. Google's Gemini is already embedded in Workspace, enabling AI-powered collaboration inside tools millions of people use daily. OpenAI is pitching developers on multi-agent orchestration. And productivity startups like Granola, which raised $43 million at a $250 million valuation, are racing to layer AI into team workflows.
Humans& isn't trying to plug into those platforms. It wants to own the collaboration layer outright, which means going head-to-head with some of the best-funded companies in tech. The startup will need constant access to compute and endless capital to train and scale a novel model architecture - resources the AI giants already control.
There's also the M&A risk. With Meta, OpenAI, and DeepMind actively recruiting top AI talent, a team this strong is an obvious acquisition target. But Zelikman says the company has already turned down interested buyers. "We believe this is going to be a generational company, and we think that this has the potential to fundamentally change the future of how we interact with these models," he said. "We trust ourselves to do that."
Whether Humans& can actually deliver on that vision remains to be seen. The company is still in the early stages of training its model and hasn't shown a working demo. But the $480 million war chest gives it room to experiment - and the founding team's track record suggests they know how to build frontier AI systems.
What's less clear is whether enterprises and consumers will adopt yet another collaboration platform, even one powered by a novel AI architecture. The graveyard of productivity tools is littered with well-funded startups that promised to fix how teams work. Humans& is betting that the shift from chat to agents creates an opening for a new kind of coordination layer - one that treats AI as a participant in group work, not just a helpful assistant on the side.
Humans& is making a big bet that the next phase of AI isn't about better answers - it's about better teamwork. With a $480 million seed round and a founding team pulled from the industry's top labs, the startup has the resources and credibility to take a swing at redefining how AI models are trained and deployed. But it's entering a market where the incumbents are already moving fast, and the product vision is still more philosophy than prototype. If the team can ship a model that genuinely improves how groups make decisions and stay aligned, it could carve out a defensible position in the collaboration stack. If not, that war chest will burn fast - and the acquisition offers will keep coming.
