This path is for the product person who is going to ship an AI feature — or is being asked to and wants to stop guessing.
It does not assume you can read a Python notebook. It does assume you are prepared to think carefully about trade-offs, hold engineers accountable to the right metrics, and make calls in conditions of genuine uncertainty.
The spine of the path is the twelve-chapter AI Manual. Each chapter ends with numbered, citable Rules (ai-1, ai-2, …) you can reference in PRDs, post-mortems, and strategy docs. The stages here group those chapters around the seven decisions a PM actually makes when shipping AI: whether to use it at all, which model, how to spec the prompt and gate the launch, how to design for uncertainty, which architecture to commit to, how to run the production economics and safety controls, and where the whole thing sits in the company's strategy.
Work through the stages in order the first time. After that, treat it as a reference — when costs spike, return to Stage 6; when the model says something your users will not forgive, return to Stage 4; when the next quarter's roadmap lands on your desk, start at Stage 1 again and ask whether AI is actually the right answer this time.
Four of the case slugs referenced (klarna-ai-deflection, air-canada-chatbot-lawsuit, github-copilot-adoption-curve, linear-ai-summary) are being authored in parallel and are flagged optional so the path renders and is completable today. The two Capstone courses (course-1-model-selection-and-evals, course-2-prompt-as-spec) are placeholders pointing at the Manual chapters that will become their backbone.