This is a 20-minute talk for people already using AI agents day-to-day. It compares two ways of building the same feature: one with iterative prompting, one with a spec written first, and walks through the actual outputs of both.
From there, I cover the structure of a useful spec and how teams can share and co-write specs before handing off to an agent. I'll also cover what makes the spec-first approach worth the initial overhead: less drift, reviewable decisions before any code is generated (more cost effective), and context awareness over time. But, it’ll also be honest about where the approach breaks down.
The audience should leave with concrete practices they can apply to their next piece of work, whatever SDD framework they choose.
