Learning to Configure Agentic AI Systems
This addresses the inefficiency and brittleness in agentic AI systems for users dealing with large combinatorial design spaces, offering a learned alternative to hand-tuned methods.
The paper tackled the problem of configuring LLM-based agent systems, which typically use fixed templates leading to inefficiency, by introducing ARC, a reinforcement learning-based policy that dynamically tailors configurations per query, achieving up to 25% higher task accuracy and reduced token and runtime costs across benchmarks.
Configuring LLM-based agent systems involves choosing workflows, tools, token budgets, and prompts from a large combinatorial design space, and is typically handled today by fixed large templates or hand-tuned heuristics. This leads to brittle behavior and unnecessary compute, since the same cumbersome configuration is often applied to both easy and hard input queries. We formulate agent configuration as a query-wise decision problem and introduce ARC (Agentic Resource & Configuration learner), which learns a light-weight hierarchical policy using reinforcement learning to dynamically tailor these configurations. Across multiple benchmarks spanning reasoning and tool-augmented question answering, the learned policy consistently outperforms strong hand-designed and other baselines, achieving up to 25% higher task accuracy while also reducing token and runtime costs. These results demonstrate that learning per-query agent configurations is a powerful alternative to "one size fits all" designs.