Polymath
ActiveSimulation environments to train & evaluate long-horizon AI agents
About
We’re heading towards a future where AI agents will be able to perform useful work over long horizons, with little or no human supervision. To increase the reliability, performance, and safety of autonomous agents, they must be trained in simulation environments that reflect the real world. Polymath builds simulated worlds for agents to practice and learn through experience. We're a team of researchers and engineers from UC Berkeley, Hume AI, Plaid, and Amazon. We have years of experience post-training frontier models in industry, and building large scale data systems. Polymath is backed by Y Combinator.
Founders · 2
AmazonCo-Founder / CEO @ Polymath. Previously @ Hume AI, AWS, UC Berkeley
AmazonCo-Founder / CTO @ Polymath. Previously @ Plaid, Amazon, UC Berkeley
Launch
Simulation environments to train & evaluate long-horizon AI agents
Polymath builds simulated worlds where AI agents learn to operate autonomously over long horizons using running applications, real tools, and multi-step tasks. They launched Horizon-SWE, a benchmark placing frontier models in a simulated software company to measure end-to-end software engineering tasks across the full lifecycle.
Formerly “Palette AI”
Related startups
Personalization infra that improves retention, LTV, and CAC
Agent simulation and RL for researchers
