Harness Learning Enables Generalizable Test-Time Adaptation for Language Model Agents

Forum topic · 小凯 · 2026-09-30

Summary

This arXiv paper (2609.35738) introduces harness learning, a method for improving language model agents. An agent is jointly defined by its model and its harness—the executable program organizing model calls, tool use, and information flow. Since different tasks require different harnesses, the authors train a proposer model to revise a solver's harness using execution feedback. They formalize this as meta-learning over executable programs, where harness revisions play the role of weight updates in gradient-based adaptation. The proposer is trained with reinforcement learning, using the task performance of revised harnesses as the reward. At test time, the proposer refines the harness using feedback from successive executions on a new task, without any parameter-space updates. Experiments on reasoning and multi-hop question answering show that harness learning improves revision quality and that test-time adaptation transfers to unseen tasks. Policies trained on single revisions can keep improving harnesses over multiple rounds, pointing toward continually learning agents that convert accumulated experience into generalizable improvements.

Paper Overview

Research Area: NLP

Authors: Alvin Zhang, Xuecheng Liu, Zixuan Wang, Fahim Tajwar, Daman Arora, Ruslan Salakhutdinov, Daniel Khashabi, Yuda Song, Andrea Zanette

Published: 2026-09-28

arXiv: 2609.35738

Abstract (Translated)

A language-model agent is jointly defined by its model and its harness—the executable program that organizes model calls, tool use, and information flow. Because different tasks call for different ways of organizing these operations, the harness needs to be adapted using feedback from the task at hand. The authors introduce harness learning, which trains a proposer model to revise a solver's harness using execution feedback.

The process is formulated as meta-learning over executable programs, with harness revisions playing the role of weight updates in gradient-based adaptation. The proposer is trained with reinforcement learning, using the task performance of revised harnesses as the reward.

At test time, the proposer uses feedback from successive executions on a new task to refine the harness, without any parameter-space updates.

Key Findings

Experiments on reasoning and multi-hop question answering show that harness learning improves revision quality.
Test-time adaptation via harness learning transfers to unseen tasks.
A proposer trained on single revisions can continue improving harnesses over multiple rounds.
The approach points toward continually learning agents that convert accumulated experience into generalizable improvements.

---

*Auto-collected on 2026-09-30*