workshop · AI Accelerator
Stop Prompting, Start Specifying: Ship AI-Built Software You Can Trust with GitHub Spec Kit
- github-copilot
- developer-productivity

This week I had the pleasure of coaching at the Volaris Group AI Accelerator in Austin and presenting two hands-on demos.
1. Stop Prompting, Start Specifying: Ship AI-Built Software You Can Trust with GitHub Spec Kit
We took a brownfield .NET app and adopted spec-driven development live. Constitution, spec, plan, then implement. Along the way, the analyze step caught a tenant-isolation gap before any code was written. We also covered how teams can run Spec Kit across many developers.
2. Your Legacy App Just Got a Brain: Build, Test, and Trust AI Agents with Microsoft Foundry
We put two agents on top of the same legacy API without rewriting it. They had read-only tools, grounding in the company's manuals, memory per user, and a human approval gate before any booking. Then we deployed agent versions live and ran evaluations in Foundry. Every built-in evaluator scored the naive agent as perfect, and our own business-rule rubric caught a prompt injection they all missed.
Thank you to the Volaris team Trey Drake / Noah Collins and every participant for the energy and great questions.
Both repos are public if you want to try them yourself:
https://lnkd.in/g5UhUbw6
https://lnkd.in/gNNRMp8q
#AI #MicrosoftFoundry #GitHubSpecKit #AgenticAI #SpecDrivenDevelopment #MicrosoftMVP
