Recursion Lab builds agents that learn from real work.

General models are useful, but they produce generalist agents. Recursion Lab helps companies and agent builders turn real workflows, expert feedback, and production traces into agents that improve over time.

• capture how agents and experts work across tools, tasks, and edge cases
• turn feedback, outcomes, and judgment into training data and evaluations
• train agents to get better at company-specific work over time

Learn more:

• Email us if you'd like your agents to improve from real work.

Built in Bangalore and San Francisco