AI can already sketch out a working app from a chatty prompt. That “vibe coding” style is fun and fast—but the moment you try to ship it, everything hurts: inconsistent output, missing tests, surprise security holes, and code no one wants to own. In this talk we’ll walk through a concrete, repeatable workflow for using AI without giving up engineering standards: Spec-Driven Development.

Using an open-source, CLI-driven flow (Spec Kit-style), we’ll start by encoding the “rules of the game”: architecture guidelines, tech stack constraints, security policies, and definition-of-done. Then we’ll let AI generate specs, plans, and tasks, and finally propose code changes that go through the same review process as any human PR. You’ll leave with a practical alternative to “just ask the AI,” plus templates you can adapt with your own team.