We Built 2 FULL Video Games In 7 Prompts (Claude Code vs GPT-5 Codex)

Creator: Riley Brown
Source video: https://www.youtube.com/watch?v=aEdRB2yVK-I
Published: 2025-09-22
Views: 248,615
Duration: 1:05:59
Intro ends: transition into body (e.g. 'Let's dive in')

Intro

The two best coding agents in the world are Claude Code and Codex with ChatGPT. Today, I'm going head-to-head with a pro developer to see who can vibe code a better video game. In this video, we're both going to build a video game with AI, and you are going to be voting on which game is better. Vishal is going to be using Claude Code. I'm going to be using Codex with GPT-5, which many people are saying is the best coding model in the world right now. We're each going to get seven prompts to build our game and we're going to be selecting five people who vote below in the comments whose game is better and we're going to be giving $1,000 worth of credits to Claude Code and Codex. Or if you want to build mobile apps, you can use your credits there. Last time Vishal and I went head-to-head, Vishal got over 1,000 votes and so he won by about 35%. And we're going to see this time if I can take him down. I'm feeling really confident. I've improved a lot. Let's see. Let the games begin.

Rest of video

Vishal scopes a 3D Star Wars-style space shooter where the player destroys a Planetron while dodging enemies and lasers.
Riley scopes a multiplayer Crossy Road-style game, choosing a more web-friendly path with shared rooms and real-time play.
The first prompts establish each game direction, with Riley discussing how Codex can be more literal while Claude Code can be more imaginative.
Vishal fights 3D movement, camera, reticle, lasers, enemy fighters, star fields, and weak-point targeting.
Riley adds gameplay polish, arcade music, multiplayer rooms, player movement, live shooting, and InstantDB-backed shared game state.
The middle section shows classic AI-game bugs: controls breaking, too much context, unclear movement, multiplayer syncing problems, and overcomplicated generated code.
Riley explains how to prompt database-backed apps by repeatedly giving documentation and keeping the AI focused on the current system.
The final prompts simplify both games, test whether each one is actually playable, and end with a vote-driven comparison of the two finished game demos.

Up: Riley Brown | Previous | Next