AI Insight
HarnessEvolve is a new framework for autonomous AI agents that enables them to reliably improve themselves by learning from reference trajectories—execution paths generated using correct answers. The system addresses three key problems in agent self-evolution: difficulty identifying which steps caused failures, agents memorizing task-specific shortcuts instead of learning generalizable skills, and new updates degrading previously learned capabilities. By separating execution from evolution and implementing quality and performance gates to validate updates, HarnessEvolve demonstrates consistent improvements over existing methods across multiple benchmarks and agent frameworks.
Why it matters
This research advances autonomous AI systems that can improve themselves without constant human intervention, which is crucial for deploying AI agents in complex, real-world environments. The framework's ability to prevent performance degradation while enabling continuous learning could accelerate the development of more reliable and capable autonomous agents for enterprise and open-domain applications.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Self-evolving agents advance toward autonomy by optimizing their harness—prompts, skills, tools, and execution logic—based on environmental feedback. This paradigm, however, is hampered by three challenges: textit{credit assignment failure}, where terminal success/failure feedback makes it ambiguous which step caused the error; textit{shortcut learning}, where agents memorize task-specific patterns rather than acquire generalizable capabilities; and textit{catastrophic forgetting}, where unguarded updates degrade previously acquired competence. In this paper, we introduce HarnessEvolve, a self-evolving framework that learns from reference trajectories to achieve reliable agent self-evolution. HarnessEvolve decouples the execution agent from the evolutionary pipeline, assigning execution, evaluation, optimization, and gating to independent agent modules, enabling generalizable and stable harness improvements. Specifically, HarnessEvolve overcomes credit assignment failure by generating reference trajectories (execution paths produced when given the ground-truth answers) and aligning failed executions against them to extract error signals, which are clustered to reveal systematic failure patterns. To prevent shortcut learning and catastrophic forgetting, candidate harness updates must pass two gates: a quality gate that filters data leakage and prompt bloat, and a performance gate that accepts each update if it improves on the current batch without degrading recent batches, with epoch-end validation on a held-out set selecting the best-performing accepted agent snapshot. We conduct extensive experiments on several benchmarks spanning open-domain and enterprise scenarios, using different models and agent frameworks. Results demonstrate that HarnessEvolve consistently outperforms state-of-the-art baselines across all benchmarks and settings, confirming reliability across task domains.
Source: HarnessEvolve: Learning from Reference Trajectories for Reliable Agent Self-Evolution