Dynamic adapter selection boosts AI agent capability and accuracy
WeAct: Dynamic Weight Switching as Agent Actions
Machine LearningArtificial IntelligenceComputation and Language
Summary
AI agents often struggle to keep old skills when learning new ones, because they update shared parts of their model and lose earlier knowledge. The authors created WeAct, which lets the agent switch between different skill modules during its work instead of sticking to one fixed setup. This approach helps the agent pick the right skill on the fly and improves overall performance. It also allows adding new skills later without messing up existing ones.
What this means in practice
- •For ai system builders: Enable AI agents to dynamically switch skill modules to maintain and combine multiple capabilities during long tasks.
- •For software development teams: Support incremental updates of AI capabilities by adding or improving skill modules without retraining entire systems.
Authors
Rui Shi, Xunjian Yin, Jiacheng Sang, Zichuan Liu, Shuyan Zhou
Abstract
Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories. Updating shared parameters to learn new capabilities can degrade previously acquired ones, resulting in catastrophic forgetting. As a trajectory unfolds, different stages may require different capabilities, yet conventional joint training and static adapter merging maintain a fixed parameter configuration throughout execution. We introduce WeAct (Weight-level Internal Actions), which trains independent capability-specific LoRA adapters over a frozen backbone and adds adapter selection to the agent's action space. These internal actions use the model's native tool-calling interface for on-demand capability composition in a shared context. Across eight deep-research benchmarks, Qwen3-32B with WeAct improves average accuracy by 6.2% relative to full fine-tuning counterparts despite using less compute, and by 27.5% relative to the strongest static LoRA-merging method. WeAct also supports modular continual adaptation, by adding a new code computation adapter to expand the agent's capabilities and updating the existing writing adapter to improve report quality. Both updates leave the backbone and other adapters unchanged, supporting modular continual adaptation in agents.