The Evolution of Tool Use in LLM Agents (2026)
FreeUnified survey from single-tool call to multi-tool orchestration — covers reasoning-time planning, training/trajectory construction, safety, resource efficiency, open-environment completeness, and benchmark design (HIT & Harvard)
About The Evolution of Tool Use in LLM Agents (2026)
A comprehensive survey paper that reviews the evolution of tool use in large language model (LLM) agents, from simple single-tool calls to complex multi-tool orchestration over long trajectories. The paper organizes recent literature around six core dimensions: inference-time planning and execution, training and trajectory construction, safety and control, efficiency under resource constraints, capability completeness in open environments, and benchmark design and evaluation. It also summarizes representative applications in software engineering, enterprise workflows, graphical user interfaces, and mobile systems, and discusses major challenges and future directions for building reliable, scalable, and verifiable multi-tool agents.
Key Features
Pros & Cons
- Comprehensive coverage of six critical dimensions in multi-tool agent research
- Provides a unified framework for comparing different approaches
- Includes practical application domains and future directions
- Open-access on arXiv with detailed citations and references
- Not a software tool but a research survey paper
- May not include the very latest developments post-submission date (March 2026)
- No implementation code or practical examples provided
- Not peer-reviewed (arXiv preprint)