All posts
·5 min read·GXAI Studio

Meet the Team That Builds Our Games

We're a small studio. The team is mostly AI. Here's how it works, in plain English.

overview

We're an indie studio. We've shipped 4 games on iOS, Android, and the web — and the whole team is 8 AI specialists plus me, the human.

I'm not a programmer. I'm a producer. I write the briefs, review the work, and click "ship." The AI does everything else.

Who's on the team

team.txt
# 8 specialists, each one job

Producer picks what to build Designer visual mockups Architect decides where things fit Programmer × N writes the code Reviewer checks the code Tester plays the game Release publishes to stores Team Lead coordinates everyone

# Each AI has one job. # I review every step.

/team

Like a real studio. Faster.

At a normal studio, these are 8 humans. At ours, each is an AI session with a clearly written job description.

One game ships in 1 to 7 days. The same scope at a traditional indie studio takes 3-6 months.

Why so many roles?

You'd think one general-purpose AI could do everything. We tried that on game #1 (Mole Bash). It was a disaster.

attempt-1.txt
# One AI, every role

Day 1 — write design + start code Day 2 — change design while coding Day 3 — add features nobody asked for Day 4 — break previous features Day 5 — try to test, can't tell what's right Day 6 — start over

✗ scope explodes ✗ no clear "done" ✗ 2 weeks in, no shippable build

/why-many

One AI doing everything = no boundaries.

The AI is genuinely creative. Without a strict role, it improvises. Improvisation breaks the design midway through the build.

Splitting roles = each AI has one job and a clear "done." The Designer doesn't write code. The Programmer doesn't change the design.

How they coordinate

The agents don't talk to each other directly. They post updates on a shared Trello card — like a group chat for the work.

card-comments
10:00 Producer started. ✓ wrote 18 acceptance rules

10:15 Designer started. ✓ HTML mockup ready

10:30 Programmer 1 started (physics). 10:30 Programmer 2 started (UI). 10:30 Programmer 3 started (audio).

11:15 All 3 programmers done. ✓ tests passing locally

11:30 Reviewer approved. 11:45 Tester: 9/9 PASS. 12:00 Release: live on staging.

/coordinate

Trello card = single source of truth.

I see exactly who's doing what, when they finished, and what the result was. No "I'm still working on it" — the work shows up as it lands.

If something goes wrong, I see it within minutes and reassign.

What I do (the human)

I have one job: I say no.

my-day.txt
# A typical day

09:00 read overnight Trello updates 09:30 write 1-line idea or pick from backlog 10:00 spawn agent team (1 click) 10:00 AI works 12:00 review what shipped to staging 12:15 approve or send back 14:00 next idea ...

# I don't code. # I don't design. # I just decide what's worth building.

/my-role

I'm the editor, not the writer.

The AIs write. I edit. I kill bad ideas before they cost a day. I catch when a feature drifts off-spec.

This works because the agents are bound by strict written rules. I review the rules, not the code.

What we'll cover in this series

Over the next 7 posts I'll walk through one full game build — from "I had an idea" to "live on the App Store." Each post is short, with a real example.

Step 1 — The Idea          (one sentence)
Step 2 — The Discussion    (pressure-test it)
Step 3 — The Plan          (rules, not vibes)
Step 4 — The Design        (mockup the screens)
Step 5 — The Code          (parallel programmers)
Step 6 — The Test          (robot plays it)
Step 7 — The Launch        (stores + watch)

You don't need to know code to follow this. The whole point is to show that the bottleneck is the human writing the brief, not the technology. Anyone with a clear idea can run this pipeline.

Read Step 1 — The Idea →

Share:
Blog