GitHub Spark vs. Copilot CLI: Making 3D Games

Table of Contents

I recently started using a Copilot Pro subscription, mainly for the Copilot CLI harness, to integrate AI into some of my workflows. Then I noticed GitHub also has a tool called Spark for generating web applications. Finally, I realized that having each generate a 3D game would make for a fun experiment.

What Is GitHub Spark?

GitHub Spark is a tool for developing web applications. The concept is pretty cool, allowing a user to enter a prompt and have a full application generated for them. It even has a feature to save and deploy the application directory in the web interface.

What Is Copilot CLI?

Copilot CLI is an AI harness. It allows us to provide domain-specific knowledge via skill files, or completely modify agent behavior with customized markdown files.

How They Were Judged

I picked an existing game that I have fond memories of playing: “Geometry Wars.” Then I gave each tool the same prompt to make me a game with a similar style. The winner was the tool that gave me the most polished game.

Which Tool Won

This is not a rigorous benchmark, but from my experience, Copilot CLI appeared to be the clear winner. It took way less prompting, and I only provided it context around general best practices for web technologies. It had several bugs around resetting game state, lack of mobile support, and issues with obstacles.

GitHub Spark took about 20 prompts to even make the game models visible on the page. At one point it frustrated me enough that I threatened to fire it, but not even my threats managed to make it resolve the issues.

If you’re wondering how good of a 3D game an AI can make in one shot, it is hosted at the moment at poly.todoprogramming.org.