Can Agents Generalize to the Open World? Unveiling the Fragility of Static Training in Tool Use

2026-07-01

AI Summary

Large language model agents demonstrate strong performance on static benchmarks but exhibit significantly degraded generalization when deployed in open-world settings characterized by dynamically evolving user queries, tool sets, and interaction patterns. This work formally defines the open-world tool-using agent problem and introduces a four-layer sandbox environment—encompassing perception, interaction, reasoning, and internalization—to systematically evaluate performance degradation under distributional shifts. Addressing the fragility of existing supervised fine-tuning and reinforcement learning approaches under perturbations, the study proposes a perturbation-augmented fine-tuning strategy. Experimental results demonstrate that this approach substantially enhances the robustness and practical utility of agents operating in open, dynamic environments.

Abstract

While Large Language Model (LLM) agents demonstrate proficiency in static benchmarks, their deployment in real-world scenarios is hindered by the dynamic nature of user queries, tool sets, and interaction dynamics. To address this generalization gap, we formalize OpenAgent (Tool-Use Agent in Open-World), a problem setting characterized by distributional shifts across query, action, observation, and domain dimensions. To systematically diagnose its impact, we construct a controlled sandbox environment where we define fine-grained environmental shifts across a four-tier hierarchy, Perception, Interaction, Reasoning, and Internalization, and conduct a comprehensive series of experiments. Our analysis yields a series of key insights, demonstrating that agents trained via both Supervised Fine-Tuning(SFT) and Reinforcement Learning suffer from varying degrees of performance degradation when confronting open environmental shifts. Building on these insights, we propose Perturbation-Augmented Fine-Tuning, a disturbance-based intervention strategy for SFT that lays the foundation for enhancing agent robustness and utility in realistic environments. Our code will be released at: https://github. com/LAMDA-NeSy/OpenAgent.

Problem

• open-world generalization
• tool use
• distributional shift
• LLM agents
• environmental dynamics

Innovation

• OpenAgent
• distributional shift
• tool-use agent
• Perturbation-Augmented Fine-Tuning
• open-world generalization