Stop Vibe Coding Everything: The Case for Spec-Driven Dev
Spec-driven development with AI coding agents could change how you
build software. In this GitKon 2025 talk, Erik Hanchett, Senior Developer
Advocate at AWS, breaks down why AI coding assistants perform dramatically
better when they start with structured specifications instead of raw prompts.
If you've been vibe coding your way through complex features and wondering
why your AI keeps going off the rails, this is the video for you.
In this session, you'll learn:
- What spec-driven development actually is (and why it's not just
- waterfall with a rebrand)
- How to create a requirements doc, design doc, and task list that
- AI coding agents can actually follow
- When to use spec mode vs vibe coding for different types of work
- How MCP servers like Context7 can pull live documentation and
- project management context directly into your specs
- How property-based testing ties back to your original requirements
- to catch edge cases your unit tests miss
This workflow works with any AI coding assistant. Erik demonstrates it
live inside Kiro (AWS's AI IDE), but the principles apply across Cursor,
VS Code, Windsurf, and GitHub Copilot workflows too.
Spec-driven development is quickly becoming how serious teams work with
AI. Watch this to get ahead of it.
00:00 Introduction to Spec-Driven Development
01:45 What Are Specs and Why Developers Stopped Writing Them
04:10 How to Build Specs Manually with Any AI Coding Assistant
07:30 Requirements Doc, Design Doc, and Task List Walkthrough
11:00 Live Demo Inside Kiro
17:20 Property-Based Testing Against Requirements
20:00 MCP Integration for Live Docs and Project Context
23:00 Spec vs Vibe: How to Decide
#SpecDrivenDevelopment #AICodingAgents #GitKon
GitKraken Desktop:
http://tr.ee/GKDYT
GitKraken CLI:
http://tr.ee/CLIYT
GitLens for VS Code:
http://tr.ee/GLYT
Git Integration for Jira:
http://tr.ee/GijYT
Git Blog:
http://gitkraken.com/blog