This chapter explores the rapidly evolving field of large language model agents, focusing on their architectures, functionalities, and the challenges they face. As Large Language Models (LLMs) like ChatGPT continue to advance, there has been significant interest in enhancing their capabilities to perform complex tasks autonomously through integrated agent architectures. The chapter begins by examining Single-Agent Architectures, where individual LLMs utilize memory, tool-use, planning, and reflection modules to execute tasks with minimal human intervention. It then discusses Multi-Agent Systems, which enable multiple LLM agents to collaborate and coordinate their actions, achieving goals that are beyond the capacity of a single agent. The chapter also highlights Task-Specific Agents, which are designed and optimized for specialized domains such as software development, gaming, and psychological assessment. Subsequently, the chapter delves into Agent Tuning strategies, detailing methods to fine-tune open-source LLMs to enhance their performance as agents, thereby narrowing the gap between open-source models and advanced commercial systems like GPT-4. Finally, we discuss how to use agentic workflows to enhance the reasoning intelligence of LLMs. Throughout, the chapter underscores the importance of these agent architectures in extending the interactive and autonomous capabilities of LLMs in various application domains.