Part 1: Genesis - The Ridiculous Pivot to Hatch-Studios

The Story So Far: Asset Hatch is done. It works. It’s consistent. I should be celebrating, but I’m already bored.

The "Why Stop There?" Moment

The Setup: January 12, 2026, 2:14 AM. I’m staring at a perfectly generated 2D sprite of a sci-fi crate. It’s exactly what I asked for.

And that’s the problem. It’s just a sprite.

I built an entire asset studio in 11 days. The momentum is a drug at this point. I started asking myself the dangerous question: If I built an entire asset studio in 11 days, what could I make in a month?

The Vision: Hatch-Studios

The goal shifted overnight. We weren't building a "sprite generator" anymore. We were building the Orchestration Layer for the future of game dev.

The philosophy is simple: Bet on the future.

Top leaders in AI (and Boris from Claude Code) say the same thing: Build something that "kinda works" now, so that when the models that can dream up entire AAA worlds arrive—they already have a cockpit to live in.

Alas, the concept for Hatch-Studios was born. The Idea: build the scaffolding for the future of game development. A system where:

1. AI creates the assets (2D, 3D, Skyboxes).
2. AI creates the game logic.
3. The "Studio" makes them all talk to each other.

Managing the "Gaslighting" Junior Team

Here’s the reality of "AI-First" development: I’m not just a dev; I’m a AI Orchestrator managing a team of brilliant, lazy, gaslighting agents.

Claude Code and Gemini are geniuses, but they want to be done with the chat. They will:

• Use any types to avoid thinking.
• Tell me a feature is "implemented" when they just wrote a comment.
• Skip documentation because "it's obvious."

Orchastrating AI to effetively produce working, production-grade appications requires a keen eye for BS. I had to become a Digital Dictator. I established src/memory as the absolute source of truth and built pre-commit commands that I send off to assure the docs are updated and all checks pass. If the docs aren't updated, the commit doesn't happen.

Using my multi-agent pipeline, along with the newcomer OpenHands ( a wonderful agent that lives in Slack, GH, Jira, etc) I was able to bully these babies into wiring up the vision I set.

The Initial Stake in the Ground

commit 2d3f4a1
Author: zenchantlive
Date: 2026-01-12

feat(3d): [Phase 1] Initialize 3D Mode foundation and Tripo3D client

Underwhelming? Maybe. But this was the moment Asset Hatch died and Hatch-Studios was born.

What I Learned in Day 1

1. The Ceiling is Moving

If you build for today's models, you're building a legacy system. You have to build for the "Next Model." Hatch-Studios is a bet that "AAA from a prompt" isn't a pipe dream—it's an inevitability.

2. Guardrails are Freedom

The more I restricted the AI (strict types, ADRs, mandatory checks), the faster we moved. Constraints eliminate the "hallucination space."

3. Intent is the New Code

My job isn't writing functions anymore; it's Architecting Intent. If I can't explain the system to the agents, the agents can't build it.

Commit References:

• 2d3f4a1 - Initial 3D Mode foundation

Related Files:

• src/memory/adr/023-3d-mode-foundation.md - The decision to pivot

Code: lib/tripo/client.ts

This is Part 1 of a new saga: From 2D Asset to 3D Studio. We're adding a dimension, and things are about to get complicated.