Agentic Software Engineering: Directing Coding Agents to Build Good Software – Hardcover

Coding agents now write much of the world’s software, yet it is the people who direct them who answer for its quality.

Good software has nine characteristics, as the international standard ISO/IEC 25010 defines them: functional suitability, performance efficiency, compatibility, interaction capability, reliability, security, maintainability, flexibility and safety. This concise textbook teaches how an engineer who directs agents achieves them. The method rests on three duties: the engineer respects the agent as a capable collaborator, equips it before it acts, and verifies what it returns. Both the structure of the book and the parts of each chapter were tested in the author’s own teaching. Every chapter opens with learning objectives and closes with a summary of what it established, and its exercises ask the reader to recall and apply rather than reread.

Topics and features:

• Follows the work in the order in which it happens, from framing a problem and writing its specification to verification, multi-agent orchestration, review and security
• Builds a project: the Tribunal, a multi-agent system that brings a historical decision before a panel of model advocates and judges
• Includes 99 exercises of seven kinds, from short classification tasks to work on the project and at the reader’s own workplace
• Prints the prompts behind the worked examples, many of them with the answer that the agent returned, and shows the method in 127 figures
• Explains how a language model behaves through five mental models, and shows where four common failure modes come from
• Dates every observation about a particular tool and keeps it apart from the lasting method

This unique book is written for advanced undergraduate and graduate students of computer science and software engineering, as well as for the instructors who teach them. Software engineers who already work with coding agents will find in it a method that they can use on their next project.

Mikael Alemu Gorsky is an AI researcher and lecturer in the School of Computer Science at the Holon Institute of Technology, Israel, and the author of The AI Pravda newsletter.