Act While Thinking: Accelerating LLM Agents via Pattern-Aware Speculative Tool Execution

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sui, Yifan, Zhao, Han, Ma, Rui, He, Zhiyuan, Wang, Hao, Li, Jianxun, Yang, Yuqing
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912974294220800
author Sui, Yifan
Zhao, Han
Ma, Rui
He, Zhiyuan
Wang, Hao
Li, Jianxun
Yang, Yuqing
author_facet Sui, Yifan
Zhao, Han
Ma, Rui
He, Zhiyuan
Wang, Hao
Li, Jianxun
Yang, Yuqing
contents LLM-powered agents are emerging as a dominant paradigm for autonomous task solving. Unlike standard inference workloads, agents operate in a strictly serial "LLM-tool" loop, where the LLM must wait for external tool execution at every step. This execution model introduces severe latency bottlenecks. To address this problem, we propose PASTE, a Pattern-Aware Speculative Tool Execution method designed to hide tool latency through speculation. PASTE is based on the insight that although agent requests are semantically diverse, they exhibit stable application level control flows (recurring tool-call sequences) and predictable data dependencies (parameter passing between tools). By exploiting these properties, PASTE improves agent serving performance through speculative tool execution. Experimental results against state of the art baselines show that PASTE reduces average task completion time by 48.5% and improves tool execution throughput by 1.8x.
format Preprint
id arxiv_https___arxiv_org_abs_2603_18897
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Act While Thinking: Accelerating LLM Agents via Pattern-Aware Speculative Tool Execution
Sui, Yifan
Zhao, Han
Ma, Rui
He, Zhiyuan
Wang, Hao
Li, Jianxun
Yang, Yuqing
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
LLM-powered agents are emerging as a dominant paradigm for autonomous task solving. Unlike standard inference workloads, agents operate in a strictly serial "LLM-tool" loop, where the LLM must wait for external tool execution at every step. This execution model introduces severe latency bottlenecks. To address this problem, we propose PASTE, a Pattern-Aware Speculative Tool Execution method designed to hide tool latency through speculation. PASTE is based on the insight that although agent requests are semantically diverse, they exhibit stable application level control flows (recurring tool-call sequences) and predictable data dependencies (parameter passing between tools). By exploiting these properties, PASTE improves agent serving performance through speculative tool execution. Experimental results against state of the art baselines show that PASTE reduces average task completion time by 48.5% and improves tool execution throughput by 1.8x.
title Act While Thinking: Accelerating LLM Agents via Pattern-Aware Speculative Tool Execution
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
url https://arxiv.org/abs/2603.18897