Learning to Configure Agentic AI Systems

2026-02-12

AI Summary

Current large language model agent systems often rely on fixed templates or manual tuning, which limits their adaptability to queries of varying difficulty and leads to performance rigidity and resource inefficiency. This work proposes ARC—the first reinforcement learning–based dynamic configuration framework—that formulates agent configuration as a per-query decision problem. ARC employs a lightweight hierarchical policy network to dynamically optimize workflows, tool selection, prompting strategies, and token budgets in an end-to-end manner. By enabling adaptive adjustments for both task reasoning and tool-augmented question answering, ARC transcends the conventional “one-size-fits-all” paradigm. Experimental results demonstrate that ARC achieves up to a 25% improvement in task accuracy over strong baselines across multiple benchmarks while significantly reducing token consumption and execution time.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Adversarial Agents
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work

Abstract

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.

Problem

• agentic AI
• configuration
• query-wise decision
• combinatorial design space
• resource efficiency

Innovation

• agentic AI
• dynamic configuration
• reinforcement learning
• query-wise policy
• resource efficiency