The Evolution of Tool Use in LLM Agents: From Single-Tool Call to Multi-Tool Orchestration

Haoyuan Xu; Chang Li; Xinyan Ma; Xianhao Ou; Zihan Zhang; Tao He; Xiangyu Liu; Zixiang Wang

cs.SEMar 24, 2026

The Evolution of Tool Use in LLM Agents: From Single-Tool Call to Multi-Tool Orchestration

Authors: Haoyuan Xu, Chang Li, Xinyan Ma, Xianhao Ou, Zihan Zhang, Tao He, Xiangyu Liu, Zixiang Wang,

Organizations: Harbin Institute of Technology, China · Harvard University, USA · Huawei Technologies Co., Ltd., China

Abstract

Tool use enables large language models (LLMs) to access external information, invoke software systems, and act in digital environments beyond what can be solved from model parameters alone. Early research mainly studied whether a model could select and execute a correct single tool call. As agent systems evolve, however, the central problem has shifted from isolated invocation to multi-tool orchestration over long trajectories with intermediate state, execution feedback, changing environments, and practical constraints such as safety, cost, and verifiability. We comprehensively review recent progress in multi-tool LLM agents and analyzes the state of the art in this rapidly developing area. First, we unify task formulations and distinguish single-call tool use from long-horizon orchestration. Then, we organize the 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. We further summarize representative applications in software engineering, enterprise workflows, graphical user interfaces, and mobile systems. Finally, we discuss major challenges and outline future directions for building reliable, scalable, and verifiable multi-tool agents.

Figures & tables

Figure 1. Paradigm Shift in LLM Tool Use: From Single-Tool Call to Multi-Tool Orchestration

Figure 3. Taxonomy of Methodologies for Multi-tool Agent Tuning Paradigms

Risk	Failure Mode	Defense
Concurrent Conflict	Race conditions, write interference, and corrupted shared states.	Dependency-aware scheduling, policy gating, and pre-dispatch verification ( 92 ; 38 ; 165 ) .
Partial Commit	Irreversible side effects leave globally inconsistent states.	Transaction-style isolation, delayed commit, compensation, and rollback ( 17 ; 121 ; 88 ) .
Post-Execution Anomaly	Safety violations emerge only after tool execution.	Critic/verifier-based auditing and execution-time checking ( 45 ; 51 ; 221 ) .
Memory/Plan Poisoning	Malicious context persists across steps and reactivates later.	Trajectory-level verification, meta-verification, and evidence-based rejection ( 34 ; 136 ; 116 ) .
Long-Horizon Drift	Error accumulation, state pollution, and confused-deputy behavior.	Lookahead search, rollback, and trace diagnosis ( 189 ; 230 ; 100 ) .

Table 1. Representative safety risks and defense mechanisms for multi-tool agents.

Figure 4. Overview of the completeness evolution in multi-tool agents. The framework characterizes functional completeness as a continual process with three sequential yet interrelated stages: capability boundary perception, which identifies missing tools, ambiguous intents, and parameter anomalies through clarification and safe termination; autonomous tool expansion, which compensates for capability gaps via tool creation, tool synthesis, API integration, and composite skill construction; and open environment adaptation, which consolidates successful tools and interaction traces into reusable experience memory to support dynamic updating and long-term evolution in changing environments.

Benchmark	# Tools	# Instances	Year	Link	Env Type
NESTFUL ( 11 )	900+	1,800+	2024		S
ToolHop ( 205 )	3,912	995	2025		S
TRAJECT-Bench ( 57 )	1,228	5,670	2025		I
ELT-Bench ( 72 )	200	100	2025		I
Toolcomp ( 123 )	11	493	2025		S
ComplexFuncBench ( 229 )	43	1,000	2025		S

Table 2. Taxonomy of Long-Horizon Multi-Tool Benchmarks (2023–2026). Benchmarks are categorized by four core dimensions defining long-horizon intelligence. Key metrics include tool space size (Number of Tools), evaluation scale (Number of Instances). Env Type : S =Static Dataset, I =Interactive Simulator, H =Human/Dual-Control, R =Real-World OS/Web.

Figure 5. A unified abstraction of application-side multi-tool orchestration. Real-world agent systems operate over heterogeneous environments, instantiate different orchestration topologies, and rely on iterative plan-act-observe-revise loops under verification, safety, and cost constraints.

LLM Tool Use

Large language model tool use studies how models select, call, and coordinate external tools to extend their reasoning and take actions.

See all 289 papers

Utility-Guided Agent Orchestration for Efficient LLM Tool Use
Mar 20, 2026Boyan Liu, Gongming Zhao, Hongli XuLLM Agent Orchestration LLM Tool Use
University of Science and Technology of China, Hefei, Anhui, China · Suzhou Institute for Advanced Research, University of Science and Technology of China
Beyond the Leaderboard: A Synthesis of Tool-Use, Planning, and Reasoning Failures in Large Language Model Agents
Jul 7, 2026Wael Albayaydh, Rui Zhao, Ivan FlechaisAI Agent Reliability LLM Agent Evaluation
University of Oxford
Empirical Comparison of Agent Communication Protocols for Task Orchestration
Mar 24, 2026Ivan DobrovolskyiAgent Communication Protocols LLM Agent Orchestration
Independent Researcher, Sunnyvale, CA, USA

Mar 20, 2026J/K move · Enter open · S save

Utility-Guided Agent Orchestration for Efficient LLM Tool Use

Boyan Liu, Gongming Zhao, Hongli Xu

University of Science and Technology of China, Hefei, Anhui, China · Suzhou Institute for Advanced Research, University of Science and Technology of China

LLM Agent Orchestration LLM Tool Use LLM Agent Workflow Optimization LLM Decision-Making

Open paper AskPDF arXiv

Tool-using large language model (LLM) agents often face a fundamental tension between answer quality and execution cost. Fixed workflows are stable but inflexible, while free-form multi-step reasoning methods such as ReAct may improve task performance at the expense of excessive tool calls, longer trajectories, higher token consumption, and increased latency. In this paper, we study agent orchestration as an explicit decision problem rather than leaving it entirely to prompt-level behavior. We propose a utility-guided orchestration policy that selects among actions such as respond, retrieve, tool call, verify, and stop by balancing estimated gain, step cost, uncertainty, and redundancy. Our goal is not to claim universally best task performance, but to provide a controllable and analyzable policy framework for studying quality-cost trade-offs in tool-using LLM agents. Experiments across direct answering, threshold control, fixed workflows, ReAct, and several policy variants show that explicit orchestration signals substantially affect agent behavior. Additional analyses on cost definitions, workflow fairness, and redundancy control further demonstrate that lightweight utility design can provide a defensible and practical mechanism for agent control.

Abstract

LLM Tool Use

Explore similar work

Utility-Guided Agent Orchestration for Efficient LLM Tool Use

Beyond the Leaderboard: A Synthesis of Tool-Use, Planning, and Reasoning Failures in Large Language Model Agents

Empirical Comparison of Agent Communication Protocols for Task Orchestration