Harness-Zero: Harness Distillation via Agent-as-Harness
Harness-Zero internalizes specialized agent-harness behavior, lifting task success from 23.3% to 44.3%.
Harness-Zero studies distilling an optimized agent harness into model weights so gains persist under one fixed deployment harness. Because the harnesses differ in action space and available information, a harnessing agent corrects student responses before execution and those trajectories become fine-tuning data. Across knowledge work, tool use, and science tasks, success rises from 23.3% to 44.3% after the specialized harness is removed, above the 41.7% reached with it still attached, and 82.3% of 28 induced behavior patterns are recovered.
- Harness-Zero transfers optimized-harness behavior into model weights.
- A harnessing agent rewrites student actions into the target action space.
- Macro-average success rises from 23.3% to 44.3% without the specialized harness.
- That exceeds 41.7% achieved while the harness remains attached.
- It recovers 82.3% of 28 harness-induced behavior patterns.
Full article248 words · extracted from arxiv.org · click to collapse
Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instances, and models, a general-purpose agent must either settle for a suboptimal shared harness or route among an ever-growing set of specialized ones. We therefore study agent harness distillation: using a domain- or instance-optimized harness as training-time guidance and transferring the behaviors it induces into model weights, so that its gains survive under a single fixed target harness. The challenge is that the two harnesses differ in action space and available information, so guidance from the optimized harness cannot serve directly as supervision for the target one. We introduce Harness-Zero, which enables harness distillation through agent-as-harness. Guided by the optimized harness, a harnessing agent corrects student responses before execution in the target harness's action space, turning harness guidance into training demonstrations. Fine-tuning on the resulting trajectories internalizes harness-induced behavior into the model, so the specialized harness can be removed at deployment. Our experiments spanning knowledge work, tool use, and science domains show that: (1) For frontier LLMs using the same evolved harness, agent-as-harness outperforms code-as-harness. (2) With the specialized harness removed at deployment, Harness-Zero improves the base model's macro-average task success from 23.3% to 44.3%, even exceeding the 41.7% it reaches with that harness still attached. (3) Harness-Zero recovers harness-induced behaviors absent from the base model, with 82.3% average recovery across 28 patterns in the three domains.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.24974