What this talk is about
AI agents write code fast — until nothing is written down, and they confidently build the wrong thing. You re-explain the same constraints prompt after prompt, quality drifts, and nobody agreed what “done” actually means. The agent isn’t the problem. The missing spec is.
This talk introduces Spec Kit — scaffolding and five commands that turn a plain idea into a spec, a plan, and a task list before a single line of code gets written. Volodymyr will run it live on a real project: constitution → specify → plan → tasks → analyze → implement, including a plan that gets rejected because it violates the architecture.
Meet the speaker
Volodymyr Paslavskyy is a DevOps Architect who builds products from idea to production, hardened to the latest security standards. He’s spent his career on the infrastructure side of “ship it reliably” — not just wiring up AI tools, but making sure what they produce holds up.
Spec Kit is his answer to a problem every team hits once AI agents touch a real codebase: speed without a memory for intent.
Three things you’re taking home
A spec-driven workflow you can reuse
Constitution → specify → plan → tasks → analyze → implement — five commands that structure how you brief any AI agent, on any project.
A live demo on a real project
Watch Spec Kit block a plan that conflicts with the architecture, and see exactly which layer to fix — the spec or the plan.
A clear read on when to skip it
Spec Kit earns its keep on well-trodden, multi-person work. Know when a plain PLAN.md is lighter and enough — before you adopt another process.
Get acsess to the recording
Previous sessions
19 June 2026
Context Engineering
3 July 2026
System Design
4 June 2026
