AI coding assistants made it trivially easy to prompt-and-pray your way to a feature — and just as easy to end up with code nobody can explain, reproduce, or extend. That approach doesn’t scale past a toy. Spec-driven development flips the model: the specification becomes the durable, version-controlled artifact, and code becomes a build output you can regenerate. GitHub’s open-source Spec Kit makes this concrete with a structured workflow that any AI coding assistant can drive.

In this session I’ll show spec-driven development the way I actually use it every day. We’ll walk the full Spec Kit workflow live — establishing a project constitution, specifying a feature, clarifying ambiguity, planning, breaking work into tasks, and implementing — and see how the quality gates catch the mistakes that prompt-and-pray ships to production. You’ll learn when to take the lean path and when the full set of gates earns its keep, and why this approach is agent-agnostic: it works with whatever assistant you already use. You’ll leave able to start spec-driving your own work the same week.