What Does an Agent Need? Bundle-Aware Agent Resource Composition
Summary
- Researchers formulated Task-Conditioned Agent Resource Composition (AgentComposition) to jointly select backbone models and tool-skill bundles for maximizing end-to-end execution utility.
- The proposed Retrieval-Grounded Bundle-Aware Agent Generation method retrieves candidate resources and past configurations, utilizing constrained beam search and a bundle-aware critic.
- The method introduces AgentSelect-Exec with 5,060 executed configurations, recovering complete reference configurations within ten candidates for 67.73% of queries.
- Results show that proxy-trained critics and execution feedback significantly improve top-ranked agent execution scores beyond existing baselines.
Why it matters
Optimizing how AI agents select and compose backbone models with tool-skill bundles addresses a major bottleneck in end-to-end execution efficiency and task performance.