Summary

From ad-hoc prompting to a structured Spec-Driven Development workflow — a practical look at how to make AI-assisted coding reliable and repeatable in real project work.

Abstract

Most developers start using AI coding tools the same way: throw a prompt at it and hope for the best. That works — until it doesn't. This talk walks through the journey from loose, ad-hoc prompting to a structured Spec-Driven Development workflow. What makes a spec actually useful? How do you define task patterns that an agent can follow reliably? Where do quality gates and verification fit in? And what does the process look like once all the pieces are in place? Honest lessons from real project work — including what didn't work and what we'd do differently.