3.2.1. How Matt Built CoD With One Prompt (and a Blind Judge)

Matt asked an AI to build a AAA first-person shooter with one prompt. The game isn't the interesting part. The prompt shows a new way of working: you stop doing the work yourself and design the system that does it.

Here are three ideas, two quotes, and one action for founders building with AI.

THREE IDEAS:

I. Split the big goal into small pieces.

"Build Call of Duty" is too big to work on directly. "Make this one texture look AAA" is small enough to finish. The prompt sends out sub-agents and gives each one a single item: textures, physics, lighting, weapons. When a big goal turns into many small tasks running at the same time, you can attempt things that used to need a studio. The skill is in where you draw the lines between the pieces.

II. The critic matters more than the builder.

The key instruction isn't "build." It's that a separate agent acts as a harsh judge and compares each piece side by side, blind, against the real game. Builders tend to approve their own work, so you need someone else to check it. Most people tell the AI to make something good and then stop. Matt adds a judge who doesn't accept "good enough."

III. Loop until the bar is met instead of stopping after one try.

"/loop until it's utterly perfect" changes when the work counts as done. It's no longer finished after one attempt. It's finished when the judge is impressed. The first draft is never the result. Quality comes from repeated rounds against a clear standard, and a clear standard plus enough rounds will beat raw talent.

TWO QUOTES

"Have a separate sub-agent check it visually to ensure it looks triple A. That separate sub-agent should be a really harsh critic, and if it doesn't look triple A, it should keep going." - Matt Shumer

"It should literally compare them side by side blind and say which one looks better." - Matt Shumer

ONE ACTION

Prove one small, real thing to yourself today.

Don't start with the whole game. Pick one piece, such as a single weapon model, a lighting setup, or one screen of your app, and run the full loop on it:

A builder agent makes it.
A harsh critic agent compares it blind against the best real example you can find.
Loop until the critic picks yours, or can't tell them apart.

If you can get one piece past a strict judge today, you know the method works. Tomorrow you run it on everything.

The question to ask yourself: What one piece could I take to "indistinguishable from the best" today, so I trust the system enough to build something meaningful tomorrow?

🎬 Watch Matt's video with the original prompt, see the prompt run, and decide which parts you'd use in your own work.

Ludovic

3× Entrepreneur, 2× Unicorn Investor, 1× IPO — Founder of BOBIC Generational Wealth, Author of Atomic Scaling

P.S. Here is Matt's full prompt:

I want you to build a first-person shooter at the level of the most recent Call of Duty games. It should be utterly perfect, visually beautiful, with every single thing done at AAA quality—from textures to physics to anything you could think of.

Fan out sub-agents and have sub-agents tackle each one individually so that the game is utterly perfect. You should /loop on each item and have a separate sub-agent check it visually to ensure it looks triple A. That separate sub-agent should be a really harsh critic, and if it doesn't look triple A, it should keep going.

Don't stop until each sub-agent is utterly wowed with the quality when compared with the actual Call of Duty game. It should literally compare them side by side blind and say which one looks better. Do this in ThreeJS. /loop until it's utterly perfect. Fan out sub-agents and ultracode.