Spec-driven Development using Coding Agents
summary
Builds a 3x3 tic‑tac‑toe app from a 4,500‑line natural‑language spec, using a multi‑agent pipeline (scout → strategist → executor), Python + FastAPI + Pydantic AI, JetBrains IDE integrations and providers like Claude/Gemini; generates an implementation plan, 11 phases (6 done) and 600+ unit tests for those phases, runs CI checks and acceptance tests, and enforces test‑first gates while human reviewers correct agent drift. Uses agents.mmd for persistent context and agent skills to standardize commits, tests and repeatable subsections.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:12Vibe coding problems
- 6:25Spec-driven development benefits
- 10:35How spec-driven development workflow works
- 14:30Live demo
- 23:59Using agents.md to control coding agents
- 27:10Agent skills for consistency
- 35:37Agentic SDLC
- 42:24Testable specs
- 43:06Lessons learned
- 45:57Future tools
- 47:43Key takeaways
related talks
25mspec-driven development
28mHow To Use Spec-Driven Development for Production Workflows
47mSpec-Driven Development with AI Agents From High-Level Requirements to Working SW
39mAgents for the Sake of Happiness
01h 04mSpec-Driven Development: Agentic Coding at FAANG Scale and Quality
30mBuilding Tools with Tools: Modern AI-Driven Development
40mHow To Tame Your Agents?
32mSpec Driven Dev From Single Player to Multiplayer to Ecosystem