Agentic Tool Use in Large Language Models

TL;DR

This paper reviews and organizes the different paradigms of agentic tool use in large language models, analyzing their methods, strengths, and challenges to provide a unified understanding.

Contribution

It categorizes existing approaches into three paradigms, analyzes their differences, and highlights key challenges to advance the field.

Findings

Identifies three main paradigms: prompting, supervised learning, and reward-driven policies.

Provides a structured review of evaluation methods and challenges.

Highlights the evolution and differences among tool-use approaches.

Abstract

Large language models are increasingly being deployed as autonomous agents yet their real world effectiveness depends on reliable tools for information retrieval, computation and external action. Existing studies remain fragmented across tasks, tool types, and training settings, lacking a unified view of how tool-use methods differ and evolve. This paper organizes the literature into three paradigms: prompting as plug-and-play, supervised tool learning and reward-driven tool policy learning, analyzes their methods, strengths and failure modes, reviews the evaluation landscape and highlights key challenges, aiming to address this fragmentation and provide a more structured evolutionary view of agentic tool use.