Most teams are getting value from AI coding tools, but they’re also accumulating a new kind of chaos: oversized diffs, missing tests, invented APIs, and “it looked right in chat” decisions that don’t survive code review. The hard part isn’t getting code to appear on screen, it’s getting changes you can trust, explain, ship, and maintain.

In this hands-on workshop, we’ll build a practical workflow for using AI tools and AI coding agents as part of real software delivery. We’ll use GitHub Copilot, Claude Code, Codex, and other AI tools. We'll focus on the mechanics that make them reliable including prompts that encode constraints, context, instructions, and reusable agents and skills for common engineering tasks.

We’ll finish by shifting from vibe-coding to spec-driven delivery using Spec Kit write. We'll write a small spec, generate a plan, implement with tests, and run deliberate review loops to catch the failures models are good at hiding. You’ll leave with templates, guardrails, and a workflow you can apply to your own codebases without changing your stack.