preface
This book began as a goal. Richard was mapping out the year ahead, and his main aim was to write a book.
The groundwork was already there. For years, he had been building AI systems: medical AI at Intelligent Ultrasound and Tendertec, trademark search at the Intellectual Property Office, and machine translation at Guildhawk. Alongside this, he had built a library of roughly 600 books on AI, most read over the past eight years. He was working in generative AI while it was still cutting-edge research, well before ChatGPT and Claude Code carried it into the daily work of software engineers.
His mentor, Felix Hovsepian, knew this. Two weeks later, while Richard was away on holiday, Felix introduced him to Andy, an acquisitions editor at Manning. That introduction became an unexpected invitation: would he write a book on prompt engineering and modern AI?
He said yes.
The invitation made sense for the same reason you are holding this book. In their own work, the prompt was usually the least engineered thing in the codebase: a string literal, tuned until a demo passed, copied between services, and owned by no one. They wanted to talk about a prompt the way we talk about a function or a class: to point at what is failing and fix it deliberately, rather than by trial and error.