Harness Learning Enables Test-Time Adaptation Without Weight Updates

Researchers from CMU and other institutions propose Harness Learning, which adapts agents to unseen tasks at test time by revising the agent harness via execution feedback instead of updating model weights, nearly doubling performance.

2026-10-01 ~ 2026-10-01 · 2 related posts

• Harness Learning nearly doubles unseen-task performance without weight updates — DanielKhashabi · 2026-10-01
• Harness Learning: Adapting Agent Scaffolds at Test Time Without Touching Model Weights — burny_tech · 2026-10-01