Tickets Don't Compile

AI coding agents have transformed how software gets built, but most teams discover the same painful truth: the agent is only as reliable as the request that drives it. Vague tickets produce vague code. Half-formed acceptance criteria produce half-finished features. The faster the agent works, the faster the gap between intent and implementation widens. This book is the practitioner's guide to closing that gap. It teaches spec-driven development, a methodology that treats the specification as the durable artifact and code as the regenerable output, and it teaches it to the three audiences who have to live with the result: the product manager who defines intent, the engineer who shapes the contracts, and the AI coding agent that executes against both.

Most resources on the topic pick one tool and one reader. This book picks neither. Its templates, patterns, and workflows work across Claude Code, GitHub Copilot, Cursor, Kiro, Spec Kit, Google Antigravity, and any agent that supports the Model Context Protocol, and the accompanying Creative Commons appendix ships drop-in configuration files that move the methodology from page to repository in under an hour. By the end, readers will know how to write a specification that an agent can execute against, how to configure agents so they follow team standards from the first prompt, and how to detect specification drift before it becomes the cost of doing business. Tickets, as it turns out, don't compile. Specifications do.