Why
The game I always wanted to make
Making games was always the goal. I built enterprise software for nine years at PayPal instead.
I had zero game-development experience, so the question was simple: what could I build if I just built what I want? That is the sandbox. AI removed the part I was missing — an engineering team. I could describe a system, review what came back, and decide again. So I ran the project the only way I know how: as a product.
The process
Same process as enterprise SaaS, different deliverable
Nothing about the method changed. Only the thing at the end of it did.
| Building SaaS at work | Building the game |
|---|---|
| Product requirements doc | Game design doc |
| Epics | Systems — meadow, combat, creatures, progression, interface |
| User stories with acceptance criteria | Features and skills — each sword art, the mount, the intro card |
| Design system and component library | Painted UI kit — 26 chrome pieces, 16 skill icons, one brushwork style |
| Prototype, then usability test | Build, then playtest in the browser |
| QA and release notes | Test and QA each release before it goes on the site |
| Iterate on feedback | Iterate the models, the look and feel, the HUD |
The game doc
Systems were epics. Features and skills were user stories.
Each system got an epic; each feature or skill got a story with a done-state I could test. What is in the demo today is what passed.
| Epic | Example user story | Status |
|---|---|---|
| Meadow | As a player, I walk a painted meadow with tall grass I can cut. | In the demo |
| Combat | As a player, I swing the Nyxblade and use sword arts from the hotbar. | In the demo 9 of 16 arts |
| Creatures | As a player, I fight Jellings, tame the dragon by standing still beside it, then ride it. | In the demo |
| Progression | As a player, I level to 50 and spend points on STR, AGI, VIT and INT. | In the demo per session |
| Interface | As a player, I learn the controls from one card and can close it with Esc. | In the demo |
Who did what
I designed and decided. The AI built.
Me
- The game doc — epics, user stories, what counts as done
- Every design decision, and the briefs for every asset
- Reviewing what came back; keeping, rejecting, re-briefing
- Playtesting, test and QA of each release
- What shipped and what stayed gated
AI and tools
- Engine code from directed coding sessions — three.js, TypeScript, the in-browser server
- Meshes, painted UI chrome and skill icons — Meshy
- Foliage sway sheets and the goblin horde — AutoSprite
- Animation clips — Mixamo, NVIDIA MotionBricks
- CC0 fills — Kenney, Quaternius
Test & QA
Every release got tested before it went up
Same discipline as a SaaS release: a test pass against the stories, a QA pass on the build, then it ships — or it doesn't.
| Release | What I tested | Result |
|---|---|---|
| Meadow slice | Controls, the intro card, the hotbar, cutting grass | Playable in the browser |
| Keyboard layer | Every action reachable without a mouse; key hints on screen | Shipped as a separate layer over the build |
| Performance | Frame rate on a 2019 laptop with an integrated GPU | 20 → 60 fps, tall grass kept |
| Look | Pixel grid against the painted textures | 600p, hard pixels, painted look kept |
| Loading | Input during load; the first skill after loading | Input gated until the world is built; no hitch |