AI Agents, Built by Hand: A Beginner's Guide to Making a Safe, Tested AI Agent in Plain Python
English | October 5, 2026 | ASIN: B0HLZDNBXK | 517 pages | Epub | 5.94 MB
Build a real AI agent yourself, one small step at a time, even if you have never written a line of code.AI agents are everywhere in the news, but most beginner books either skip the code entirely or start with a framework that hides how agents actually work. This book takes a third path. You build an agent by hand, in plain Python, and understand every line.Your project is Bob, a helper for a small bicycle repair shop. Bob starts as a simple question and answer. By the end it looks up repairs, adds up prices with real tools, searches the shop's documents, books appointments only with a person's approval, resists attempts to trick it, and passes its own tests.What makes this book differentMade for complete beginners. One chapter teaches exactly the Python you need, and nothing more.Nothing to install but Python. Bob uses only Python's standard library. No packages and no frameworks.No paid account needed. A built-in practice model runs the book's examples offline and prints exactly the output shown in the book. Optional sections connect Bob to a real model running on your own computer, or to a cloud model.Safety from the first tool. Approval checks, action limits, step limits, and defenses against prompt injection are part of the main path, not a final chapter.Tested, not just demonstrated. You learn to write automatic tests and to evaluate an agent's behavior, so your agent keeps its promises after every change.You will learn how to:tell a chatbot, a workflow, and an agent apart, and choose the right one;give a language model tools, and build the agent loop with a step limit;add memory, and let your agent search documents and cite its sources;defend against prompt injection, using the security community's current lists of top risks;test and evaluate an agent, and measure what it costs;connect agents to tools with the Model Context Protocol (revision 2026-07-28);read the documentation of professional agent frameworks with confidence.Every program in the book was run, and every printed output was produced by that code. Sixteen chapters, practice tasks with answers, troubleshooting help, and a companion download with all the code and data take you from your first program to a finished, guarded, tested agent.If you want to understand AI agents by building one properly, start here.
Links are Interchangeable - Single Extraction
