humans&

Market position

humans& is an early-stage, pre-product frontier lab that raised one of the largest seed rounds on record. TechCrunch notes the round is smaller than Thinking Machines Lab's $2 billion seed at a $12 billion valuation (July, the largest to date) and comparable to Unconventional AI's $475 million at a $4.5 billion valuation in December. The company's launch is characterized by press as part of a trend of large financings for startups founded by breakaways from major AI labs, occurring amid warnings from analysts and industry insiders about an AI bubble. The New York Times situates the company within a broader 'human-centered AI' movement that includes Stanford's Institute for Human-Centered Artificial Intelligence, founded in 2019.

The company positions itself against labs optimizing for autonomous task completion and worker replacement. Co-founder Andi Peng contrasted the approach with Anthropic's emphasis on models working autonomously for many hours at a time, saying she views machines and humans as complementary. Zelikman argues that the prevailing question-and-answer paradigm is insufficient because chatbots are not good at asking questions or understanding what people want. Forbes notes that few startups are attempting new model-training methods given the compute and capital required, placing humans& among a small set of labs — alongside Thinking Machines Lab and Safe Superintelligence — that raised large sums before shipping a product.

Technology

The company's research agenda centers on long-horizon and multi-agent reinforcement learning, memory and user understanding, with the goal of models that proactively request information from users, retain it, and reason about individual and group preferences, goals and constraints. Zelikman previously authored what his website describes as the first paper training language models to reason in natural language, and worked on pretraining data collection, reasoning and agent infrastructure at xAI. Grokipedia characterizes the approach as combining emotional intelligence with cognitive capability, using reinforcement learning variants adapted for cooperative rather than competitive outcomes, plus iterative human feedback loops. The company has published a blog post on stability and performance trade-offs in NVFP4 reinforcement learning. Forbes reported the company told investors the new training paradigm will require more compute than current strategies.

Go-to-market

No go-to-market motion is described in the sources. The company's site directs prospective employees to a careers page, states an intention to contribute back to open source and academic research, and invites research collaboration via its @humansand account on X. Public visibility to date has come from a coordinated launch announcement and press coverage in January 2026.

Sources indicate an intended focus on people working together in organizations, teams and communities — software to facilitate collaboration among colleagues, friends or family, with AI acting as connective tissue across an organization. No specific customer segments, industries or named customers are identified.

Geography

The New York Times and thesaasnews describe humans& as a San Francisco-based startup. No other offices or international operations are mentioned in the sources.

History

Grokipedia states Eric Zelikman founded the company in September 2025 as co-founder and CEO, launching the humansand.ai website and the @humansand account on X that month, shortly after leaving xAI. The New York Times reports Zelikman left xAI in September 2025 and then assembled the founding team. On October 31, 2025, Forbes reported that the company was in talks to raise $1 billion at a $5 billion valuation, citing six sources; humans& declined to comment. On January 20, 2026, the company published its public launch post and press reported a completed $480 million seed round at a $4.48 billion valuation, with the company described as roughly three months old and employing about 20 people. Note: internal records date the company's founding to 2019 and place its headquarters in El Segundo, while the sources here describe a 2025 founding and a San Francisco base.

Risks & controversies

Forbes reported that one investor who passed on the fundraise called the round size 'too big of a number' for an early-stage company. Coverage notes the company is valued at $4.48 billion with about 20 employees, no released product and a three-month operating history, and that its proposed training paradigm requires more compute than current approaches. TechCrunch cites the example of Thinking Machines Lab, where half the founding team departed after a record seed round, as evidence that capital and pedigree do not guarantee success, and notes broader warnings of an AI bubble. Sources conflict with internal records on founding year and headquarters location.