Skip to main content

Preface

Software development has moved beyond code completion. Coding agents can inspect repositories, edit multiple files, run tools, delegate work, and continue asynchronously. That capability changes the engineering problem: producing code is no longer the only scarce activity. Teams must communicate intent, bound authority, isolate execution, and demand evidence.

This guide treats specification-driven development as one component of a complete agentic AI software development lifecycle. It does not assume that code is disposable, that an agent is correct because it finished, or that one framework fits every project. Code comprehension, debugging, architecture, security, and operational judgment remain essential.

The Four Dimensions​

  • Intent records outcomes, requirements, constraints, uncertainty, and decisions.
  • Authority defines the tools, data, environments, and actions an agent may use.
  • Execution makes work isolated, resumable, observable, and safe to integrate.
  • Evidence connects requirements to tests, evals, reviews, releases, and production outcomes.

Who This Book Is For​

The primary reader is a working engineer. Individual contributors can follow the implementation path; leads can focus on task boundaries and review; platform engineers can focus on harnesses and delivery; security teams can focus on authority and threats; modernization teams can follow the brownfield path.

The Running System​

TaskFlow is an existing multi-tenant SaaS product with a TypeScript API and web application, PostgreSQL data, background work, role-based authorization, incomplete tests, and production telemetry. Its undocumented behavior and technical debt are intentional: agentic engineering must work on existing systems, not only clean demonstrations.

The main change—organization-scoped audit export—is specified and delivered through GitHub Spec Kit, OpenSpec, and BMAD. Later chapters isolate parallel work, add evals, release behind a flag, simulate an authorization and latency incident, and feed the learning back into the specification.

Evidence and Currency​

Durable chapters teach principles. Framework and protocol chapters are marked Evolving. Commands, versions, support matrices, and model-specific facts are Snapshot material with a technical verification date. Primary sources are cited near factual claims.