Research
Research & blog
- May 5, 2026
meta-agent: continual learning for agents
An open-source library that automatically improves agent harnesses from production traces. On tau-bench v3 airline, it raised holdout accuracy from 67% to 87%.
- April 15, 2026
From knowledge-work agents to self-improving systems
How deploying Canvas with enterprise customers led us to build the continual-improvement layer for production AI agents.