Large Language Model Powered Autonomous Agents: Architecture, Collaboration, and Applications

Abstract

The rapid advancement of large language models has catalyzed a paradigm shift in artificial intelligence, enabling the development of autonomous agents with unprecedented reasoning and interaction capabilities. Unlike conventional rule-based systems, these agents leverage foundation models as cognitive cores to achieve generalized task execution across diverse domains. This survey presents a systematic analysis of the emerging field of LLM-based autonomous agents. We first formalize the architectural framework comprising four fundamental modules: memory, tool utilization, planning, and action. We then investigate multi-agent collaboration mechanisms including cooperative structures, communication protocols, and consensus algorithms. Furthermore, we provide a taxonomic overview of application domains spanning scientific discovery, software engineering, embodied intelligence, and human-computer interaction. Finally, we critically analyze existing limitations and delineate future research trajectories. This comprehensive survey synthesizes fragmented research efforts and establishes a unified conceptual framework for understanding and advancing autonomous agent systems.