English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Harness-Zero: Agent Harness Distillation via Agent-as-Harness

Forum topic · 小凯 · 2026-09-23

Summary

Harness-Zero is a method for agent harness distillation, introduced by researchers including Haoran Ye and Guojie Song in arXiv paper 2609.24974. Agent harnesses—external systems mediating model-environment interaction—boost agent performance, but their gains are tied to the specific harness deployed, and the optimal harness varies by domain, instance, and model. Harness-Zero transfers the behaviors induced by an optimized (training-time) harness into model weights so those gains persist under a single fixed target harness. The core idea, Agent-as-Harness, uses a harnessing agent that, guided by the optimized harness, corrects student model responses within the target harness's action space, converting harness guidance into training demonstrations. Fine-tuning on these trajectories internalizes harness-induced behaviors, allowing removal of the specialized harness at deployment. Experiments across knowledge work, tool use, and science show: (1) Agent-as-Harness outperforms Code-as-Harness for frontier LLMs with the same evolving harness; (2) removing the specialized harness at deployment raises the base model's macro-average task success rate from 23.3% to 44.3%, exceeding the 41.7% achieved with the harness present; (3) it recovers on average 82.3% of harness-induced behaviors across 28 patterns in three domains.

Paper Overview

Field: NLP Authors: Haoran Ye, Yuxing Lu, Haonan Dong, Zhaochen Su, Guojie Song Published: 2026-09-21 arXiv: 2609.24974

Summary

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.

The authors 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 harness.

Harness-Zero addresses this via Agent-as-Harness: guided by the optimized harness, a harnessing agent corrects the student model's responses within the target harness's action space before execution, turning harness guidance into training demonstrations. Fine-tuning on the resulting trajectories internalizes harness-induced behaviors into model capabilities, removing the specialized harness at deployment.

Key Findings

Experiments span knowledge work, tool use, and science domains:

1. Agent-as-Harness beats Code-as-Harness for frontier LLMs using the same evolving harness. 2. Deployment without the specialized harness: Harness-Zero lifts the base model's macro-average task success rate from 23.3% to 44.3%, even surpassing the 41.7% achieved when the harness was still present. 3. Behavior recovery: Harness-Zero restores harness-induced behaviors missing from the base model, recovering on average 82.3% across 28 patterns in three domains.

Links

  • Paper: https://arxiv.org/abs/2609.24974
*Auto-collected on 2026-09-23.*

Tags

#agent-harness#distillation#llm-agents#arxiv#nlp#fine-tuning#agent-frameworks#training-methods

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178635103