Self evolving agents fix mistakes but struggle to apply lessons broadly
Self-Evolving Agents Can Repair Failures but Struggle to Generalize from Them
Computation and Language
Summary
Some AI agents try to get better by learning from their mistakes without any outside help or clues about what works. This paper finds these agents can fix problems they already saw fail, but they have a hard time using those fixes to solve new, different problems. The researchers show that even when the agents know what corrections are right, it’s tough for them to generalize that knowledge. Adding feedback or more tries helps only a little, while seeing successful examples helps much more.
What this means in practice
- •For interactive ai developers: Improve AI agents that learn from errors without external guidance by focusing on repairing specific failures rather than expecting broad skill growth.
- •For software testing teams: Design AI tools that iteratively fix bugs seen during testing but remain cautious about relying on these tools to prevent future unseen issues.
Authors
Weixiang Zhao, Yichen Zhang, Yanyan Zhao, Jinming Bai, Bin Liang, Ruifeng Xu, Bing Qin, Ting Liu
Abstract
Experience-driven self-evolution offers a promising path for large language model (LLM) agents to continually improve through interaction. Yet existing approaches often rely on successful trajectories or external evaluation signals, leaving unclear whether agents can evolve when such positive anchors are unavailable. We study this question through Cold-Start Failure-Only Evolution (CFOE), where agents start without accumulated experience and learn only from failed interactions without privileged guidance. Across three self-evolving paradigms, multiple LLM backbones, and four interactive environments, we uncover a consistent repair-generalization gap: failure-derived experience can help repair previously failed tasks, but rarely yields sustained improvements on unseen tasks. Further analysis shows that agents often misinterpret failures, while even verified-correct experience remains difficult to transfer. Relaxing CFOE reveals that verifier feedback and additional interaction provide limited gains, whereas successful trajectories substantially improve transfer. These findings suggest that current self-evolving agents are better at repairing observed failures than building generalizable capabilities from failure.