Abstract: From symbolic and reactive agents to reinforcement learning and transfer learning-based agents, and then to the current large-model-based agents, AI agents have evolved through three generations, gradually becoming the core of AI research and applications. Among them, agents based on large models are intelligent computational entities capable of more accurately perceiving the environment, reacting and making judgments, forming, and executing decisions. The wide range of application prospect has been demonstrated in various fields including image generation, video generation, data analysis, graphic image editing assistance, negotiation guidance, educational instruction, academic research, life assistant services, website development, etc. Foreign companies such as OpenAI, Google, Apple, and NVIDIA, as well as domestic companies like Tencent, Baidu, SenseTime, Lenovo, and iFlytek, have actively explored and practiced in the field of large-model-based agents, launching their respective large-model-based agent products that cover various fields such as games, daily life, online assistants, marketing, and education. This paper reviews the definition and development of AI agents while focusing on the concept and development frontiers of large-model-based agents as well as their representative achievements in industrial practice. It aims to provide reference for readers who aspire to engage in research and development of large-model-based agents.