Efficient training method improves large language model agent harnesses
Ecdysis: Efficient and Effective Training of Runtime Harnesses for LLM Agents
Software EngineeringArtificial Intelligence
Summary
Training the control systems, or harnesses, that manage large language model agents can be slow and sometimes leads to systems that work well only on tested tasks but not new ones. The authors identify that this happens because these training methods often don't clearly understand the causes of failures. They propose a new way called Ecdysis that looks at failures across many tasks together to better pinpoint common problems and avoid unnecessary fixes tied to specific model quirks. This method makes training faster and results in harnesses that reason more accurately across different tasks.
What this means in practice
- •For machine learning engineers: Train runtime control systems for language model agents faster and with better generalization to new tasks.
- •For software developers building ai assistants: Improve reliability and reasoning of AI assistants by evolving their runtime harnesses using efficient failure diagnosis.
Authors
Ruiqing Yue, Yu Cui, Zhuoyu Sun, Sicheng Pan, Xianhong Xue, Tingyu Li, Ting Li, Wenzhuo Zhu, Yi Chen, Yifei Liu, Baohan Huang, Zhe Cui, Haibin Zhang, Cong Zuo
Abstract
Self-evolving runtime harnesses can substantially improve the capabilities of large language model (LLM) agents and provide a promising paradigm for optimizing agent execution. Existing harness evolution methods typically rely on iterative search, repeatedly evaluating and revising candidate harnesses based on execution feedback from task instances. While this paradigm enables continuous harness optimization, it incurs substantial time overhead due to repeated agent executions and code modifications, and may overfit to observed tasks and specific failure patterns, resulting in degraded generalization to unseen tasks. We identify the lack of principled failure diagnosis as a key bottleneck in harness evolution: an observed failure can reflect either model-specific deficiencies or systematic harness deficiencies, and directly optimizing against individual failures can lead to unnecessary model-specific accommodation. We therefore propose Ecdysis, an efficient and effective framework that distinguishes model-specific accommodation from harness-level repair and biases adaptation toward systematic harness deficiencies by identifying recurring cross-task failure patterns. Ecdysis adopts a batch-level cross-instance failure aggregation paradigm to jointly analyze failure evidence from multiple task instances and further introduces Failure-Driven Collaborative Refinement to diagnose failure causes and iteratively refine harness modification specifications. By combining cross-instance failure analysis with multi-role diagnosis, Ecdysis enables more effective harness evolution with lower training time. Experiments show that Ecdysis achieves up to a 1.84x speedup in harness training compared with existing harness evolution methods, while improving the reasoning accuracy of the resulting harnesses by 18.56%.