AI Spec-Driven Game Development

What you'll learn

• Write a spec that works. Describe behavior precisely — Gherkin/BDD scenarios and design documents — so a person or an agent builds it correctly.
• Drive AI coding tools. Use Claude Code, Cursor, and OpenSpec to turn specs into working games — and learn where your judgment still decides everything.
• Generate assets with AI. Pixel art, 3D models, sound, and voices — with image generation, Meshy, ElevenLabs, and more.
• Real game-dev skills. Game loops, collision, input, state, reproducible seeded runs, lit 3D scenes, and a real engine (Godot).

No deep computer-science background required. You'll leave able to build far more than six games.

The six projects

1. 

   🤖Dock Bot

   browser · JavaScript

   A turn-based crate-pushing puzzle. One screen, nine sprites, then seeded boards on demand. The whole game is its spec.

2. 

   🐸Capybara Crossing

   browser · JavaScript

   Frogger, but chill. Add real time: a frame-rate-independent loop, moving hazards, collision, lives.

3. 

   🧱Brick Smasher

   2D engine · C

   Breakout on the Finch engine. Closer to the metal: pixel buffers, sprites, and deterministic playtest evidence to diagnose what you see.

4. 

   ⛏️Cave Runner

   Godot

   Your first game in a real engine. Build it from a spec and asset pack, then take the training wheels off and extend it.

5. 

   🔥Marshmallow Run

   Godot · 2.5D

   A one-screen platformer whose rules are simple on purpose — so the lesson is the 3D scene: camera, light, materials, an imported mesh, built gray-box first, then dressed with art.

6. 

   🏎️Low-Poly Cart Racer

   Godot · 3D

   Everything comes together: spec writing, AI-generated 3D assets, a full 3D engine, a checkpoint circuit with medal times.

How it works

Spec → drive the AI → test and prove → ship. You write the contract; the AI builds; you verify the actual behavior, not a convenient proxy. When something breaks — and it will — you won't guess: you build tools that let you and your AI see it, reproduce it, and prove the fix. The difference between asset soup and a game with a soul is your taste. That's the skill. It's forever.

Who it's for

Builders and aspiring builders curious about AI-assisted development. If you can code a little — or want to — and you care about making good things, you'll fit. You'll install a small, staged toolkit as you go: Git, one coding agent, and Node to start; the C and Godot tooling arrives when the course reaches those projects.