LLM agents improve task handling by refining their own execution setups
Harness^2: Recursive Agent Harnessing for an Open World
Artificial IntelligenceComputation and LanguageMachine Learning
Summary
Large language model (LLM) agents rely on how they are set up to perform tasks, not just on the model itself. The paper introduces Harness², a framework that lets these agents improve their task-handling methods by repeatedly editing and learning from their own setup without needing outside feedback. This recursive approach helps them adapt better to new tasks they haven't seen before. The authors tested this system on several benchmarks and found consistent improvements in performance.
What this means in practice
- •For enterprise ai teams: Automatically improve AI task workflows in changing business environments without manual redesign of system settings.
- •For software engineers: Enhance AI agent robustness by enabling systems to self-adapt execution strategies for unknown tasks encountered during deployment.
- •For customer support platforms: Improve AI response systems by recursively refining their operational logic to handle novel user queries more effectively.$Commercial implications: Makes it possible to sell continuously self-improving chatbot platforms that adapt to varied customer questions without manual updates.
Authors
Ruiyao Xu, Yanfei Chen, Zhongying CuiZhu, Bhavana Dalvi Mishra, Yifei Ming, Han Yu, Rujun Han, Chen-Yu Lee, Tomas Pfister
Abstract
LLM agents' capabilities depend not only on their underlying models but also on their execution harnesses. Recursive self-improvement (RSI) offers a promising approach to automatically evolve these harnesses. However, deploying RSI in open-world settings still faces two critical challenges. First, open-world tasks are typically unseen during the harness design phase and novel at test time, leaving no ground-truth verifiers to reliably guide online harness edits. Second, the RSI process remains highly task-specific and fails to generalize to novel scenarios. We introduce Harness^2, a plug-and-play, recursive agent harnessing framework that addresses these challenges through two connected levels of recursion. Task-level recursion refines the current task's harness using a contrastive Propose-Probe-Compose step that reliably guides edits without ground-truth verifiers, and can be scaled further via parallel or sequential recursion. Harnessing-level recursion then extracts experience from these edits to update a continual improvement harness, enabling the agent to better adapt to future tasks. Across three professional benchmarks and four model-harness configurations, a single step consistently improves GLM-5 and Gemini 3.5 Flash over their base harnesses, yielding gains of up to 11.5 points on WorkBuddy-Bench, 4.5 points on JobBench, and 11.8 percentage points on LAB, with further gains as the task-level recursion budget increases.