Building Agents That Work

Building Agents That Work is for information workers, managers, analysts, and team leads who want to build useful agents in ChatGPT without turning the whole thing into a software project. It is written for people who need agents to do real work more than once, not just survive one impressive test run. This edition uses Custom GPTs in ChatGPT, and it teaches the subject by having the reader build as they go. It starts with the problem many people hit early: a role prompt or rough instruction block looks fine at first, then starts slipping. The answer is not endless prompt tweaking. The book shows how to give the agent a stronger structure, separate standing rules from task-specific guidance, organize knowledge files, route work more deliberately, and test results so the system holds up better over time. The material builds in sequence. Readers start with how a GPT's instructions actually work, then move through starter patterns, a worked example, and the shift from a single-file build to a modular file set. You build and test as you go, which means the ideas do not stay abstract for long. There is also a companion repository with starter and example files. That matters because the book is meant to leave the reader with something they can use, not just something they understand. By the end, readers come away with a mental model for how dependable custom GPTs are put together, along with starter materials they can adapt to their own work in communication, analysis, planning, operations, and other knowledge-work settings. This is not a coding manual, a hype book, or a bag of prompt tricks. It is a practical learning guide for people who want to build, test, and refine custom GPTs, then leave with a framework and a set of starting files they can actually use.