Writing Use Cases for AI

For forty years, use cases have been the best tool we have for describing what a system must do. Ivar Jacobson invented them, Alistair Cockburn taught a generation to write them well, and Use-Case 3.0 connected them to modern agile delivery. All of that work assumed one thing: the reader of a use case is a human being.

That assumption no longer holds. With AI code generation, a use case has a third reader: an AI agent that turns the specification directly into code and tests. Implementation has become cheap; expressing intent precisely is the new bottleneck. The specification is now the asset, and the code is its output.

This guide teaches you to write use cases that work for all three readers: the stakeholders who validate them, the engineers who review against them, and the AI agents that implement from them. You'll learn:

• What actually changes when an AI reads your specification
• The classical fundamentals that still carry the weight: actors, goals, scope, and complete behavioral stories
• A practical specification template with a fully worked example
• How to write main scenarios and alternative flows an AI cannot misread
• Precision calibration: what to say, what to leave out, and how tests protect the gaps
• How to slice use cases for iteration, and how to handle change requests, bugs, and brownfield systems
• Checklists and common mistakes to keep open in a split pane while you write

Part of the AI Unified Process, and a companion to Spec-Driven Development: From Specs to Code with AI Agents (Apress, 2026). Written for requirements engineers, business analysts, and software engineers who know use cases from Cockburn or Use-Case 3.0 and want the delta, not a rehash.