Specifications have always mattered. Agentic AI makes them matter more.
The revolution started in software engineering. The term spec-driven development is recent: build the specification first, then code. The coding half is increasingly automated by long-running AI agents, and they get better every week. This shift takes the load off the coder and lands it on the specification. An autonomous agent produces a meaningful result only if the spec is near-perfect, and writing specs by hand is no longer enough. This book focuses on building specifications at the level of detail that autonomous development demands — and leans heavily on AI in the process.
The same shift is reaching other knowledge-work fields. Before a lawyer drafts the final contract, they work out the parties, the obligations, the consideration, the edge cases, and the termination conditions. Drafting a final contract is the "coding" step; the working-out is the "specification" step. They may not call it that, but it is the same shape of work. A product manager runs discovery and writes a PRD before engineering work begins. A researcher writes a protocol before the study runs. A policy writer drafts intent and scope before finalizing the policy itself. An author drafts a synopsis and an outline before writing the book. A consultant writes a scope of work before the engagement begins. In every case, a highly detailed document is produced before the final stage of the work starts; that document is the specification. The methodology in this book applies to most such cases, not just to engineering.
An AI agent can now write the code, draft the contract, and run the research. What it can't do is design a specification for itself; it needs human judgment. So the spec is the new bottleneck — and writing specs by hand is no longer enough. The book teaches how to leverage frontier AI for this type of work.