Harness2: Recursive Optimization for Open-World Agent Systems
October 19, 2026
Harness2 is a framework for recursive agent optimization that uses task-level and harnessing-level recursion without changing model weights. It achieved up to 11.8 percentage point gains on LAB's partial pass rate by refining execution harnesses through behavioral contrasts.
HOW THIS AFFECTS YOU
●
builderYou can optimize your agent's tool-use and orchestration layers through automated, recursive refinement.
●
researcherThis demonstrates a way to improve model performance in unseen environments without weight updates.