Backed by Y CombinatorBuilding AI systems that improve from experience.
Canvas Coworker
An AI coworker for long-running GTM and knowledge-work tasks,built to learn from feedback and improve with use.
Research & Blog
AutoHarnessBench
Can frontier models automate agent harness R&D? A benchmark for models that improve task-specific harnesses under a fixed budget.
August 2026PreviewPrivileged On-Policy Self-Distillation
Post-training method that builds privileged repair hints from failed rollouts and distills back into the policy.
July 2026ReadMeta-Reward: Reward Modeling as Harness Optimization
Optimizing LLM-judge evaluator harnesses to build better agent reward models.
May 2026ReadMeta-Agent: Continual Learning for Agents
An open-source library that automatically improves agent harnesses from production traces.
May 2026Read