AI coding tools can generate features fast, but teams hit the same wall when it’s time to ship: architecture drift, missing tests, surprise security gaps, and PRs nobody wants to own. This session shows a practical alternative: Spec-Driven Development, a lightweight workflow that keeps AI output anchored to your engineering standards.

We’ll start by writing a minimal “constitution” for the repo: constraints, definition-of-done, and the checks that must pass. Then we’ll walk through an AI-assisted, CLI-first flow that generates a spec, turns it into traceable tasks, and proposes code changes that remain reviewable and testable like any normal PR.

You’ll see a small feature built end-to-end, plus patterns for adopting spec-driven workflows in existing codebases without turning them into waterfall. Leave with templates, a checklist, and a clear rule: when vibe coding is fine for exploration, and when the spec must drive delivery.