Tag
This paper introduces Harness-Zero, a method for distilling optimized agent harnesses into large language models via agent-as-harness, significantly boosting task performance in knowledge work, tool use, and science domains even after specialized harnesses are removed.