Harness model collaboration improves long term learning and reuse
Harness-Model Collaboration for Long-Horizon Continual Learning
Artificial Intelligence
Summary
Many intelligent agents need to learn and remember useful step-by-step procedures over a long time, but current methods struggle to balance quick recall and lasting understanding. The authors introduce HaMo, which combines fast experience storage with slower learning to improve how agents remember strategies and adapt to new tasks. HaMo turns detailed action sequences into easy-to-retrieve summaries and trains the model to apply them flexibly. Their tests show HaMo works better than methods relying only on memory or only on learning models, especially when tasks change.
What this means in practice
- •For web automation developers: Enhance web agents to better remember and reuse complex multi-step workflows across different scenarios, improving automation effectiveness.
- •For enterprise ai teams: Improve AI systems that must adapt and retain procedural knowledge over time for business processes, reducing retraining needs.
Authors
Zhen Bi, Zhizhi Peng, Yan Wang, Xueshu Chen, Haofei Zhang, Yaobo Zhang, Zeyu Yang, Qing Shen, Jungang Lou
Abstract
Long-horizon agents continuously encounter reusable procedural knowledge, yet existing systems typically rely either on model adaptation or external memory, trading persistent learning for rapid reuse. This limitation is especially important for Web agents that repeatedly execute related multi-step workflows. We propose HaMo, a Harness-Model collaborative framework that combines fast non-parametric experience accumulation with slow parametric adaptation. HaMo distills trajectories into strategy-level experience cards for immediate retrieval, while an experience-conditioned continual learning objective trains the model to interpret, ground, and execute retrieved strategies across changing contexts. A retrieval-coverage-based selection mechanism further identifies compact and complementary trajectories for parameter updates. Experiments show that HaMo outperforms harness-only and model-only adaptation under within-domain and cross-domain shifts, improving both retention and adaptation. These results highlight the importance of jointly preserving procedural experience and internalizing the competence to transfer and execute it across tasks.