Translation: This article, jointly published by the Fudan NLP Laboratory and Mihoyo, provides a comprehensive and systematic review and outlook on the development trajectory, key elements, application scenarios, and social impacts of intelligent agents based on large language models (LLM-based Agents). The article divides the composition of an Agent into three parts: the brain, perception, and action, analyzing their internal working mechanisms. Additionally, the authors discuss the practical aspects of single-task applications, innovative applications, and group collaboration of agents, as well as the social impact issues they face, offering significant guidance for the exploration and application of LLM-based Agents.