1 The rise of AI agents
This chapter covers
- Defining agents and agentic thinking
- Introducing the Model Context Protocol (MCP)
- Understanding the agent foundational layers
- Advancing onto multi-agent systems
The agent is not a new concept in machine learning and artificial intelligence. In reinforcement learning, for instance, the word agent denotes active decision-making and learning intelligence. In other areas, the word agent aligns more with an automated application or software that does something on your behalf. At its essence, the term agent refers to an entity that can act on behalf of another.
An AI agent has agency, as in the ability to make decisions, undertake tasks, and act for someone or something. This is in stark contrast to what we may refer to as the classic AI assistant or chatbot. ChatGPT and Claude are becoming more agentic, allowing you to delegate work through tools or agents like Deep Research. This can all get a little muddled, so we will review the boundaries between assistant and agent shortly.
This book focuses on agents—how they work, what makes them powerful, and how we can build agents and agentic systems to automate complex tasks. While there are multiple patterns for building AI Agents, this book will focus on building agents powered by LLMS.