humans& is a frontier AI lab working on core research and systems for human-centered AI. The team focuses on long-horizon and multi-agent reinforcement learning, memory systems, and models of user behavior - treating these as foundational pieces for AI that operates effectively within human contexts rather than abstractions of them.
The lab integrates science and product development directly, which creates specific trade-offs: velocity in evaluation against architectural purity, practical constraints shaping research direction. This coupling matters because the problems they're targeting - how models maintain coherent behavior across extended horizons, coordinate across agents, maintain and query memory, and build accurate models of what people actually need - have different character in deployment than in isolation.
Their stated aim centers on AI as connective infrastructure for organizations and communities, which frames the bottleneck differently than pure capability maximization. The binding constraint is not raw model performance but trustworthiness, interpretability of behavior, and whether people can meaningfully understand and collaborate with these systems. This orientation shapes what they build and how they measure progress.