How I AI: Speedrun a 3D Multiplayer Game in 15 Minutes with Cody De Arkland

Today, I’m talking to someone who might just vibe code harder than I do: Cody De Arkland, the Senior Director of Developer Experience at Sentry. Cody is one of the most prolific AI-powered builders I know, and he makes everything from personal productivity apps for his family to complex workflows for his day job.

While I've seen him build some seriously cool 'meme apps' with Bolt, what really floored me was a fully functional, multiplayer 3D space game he built. This is the kind of project that used to take teams months, if not years, to pull off. Cody, who had zero game development experience, built it by treating AI tools like Cursor and Claude as junior developer partners.

In this episode, Cody shows us exactly how he does it. First, we break down how he built his game, Spaceflight. Then, he does something wild: he speedruns building a brand new 3D multiplayer flight simulator from scratch, live on the show, in about 15 minutes. We go from an empty folder to a playable prototype, showing just how much speed and creativity 'vibe coding' can unlock.

Workflow 1: Deconstructing the 'Spaceflight' Game

Before we got into the live demo, I had to ask Cody about the project that first caught my eye: Spaceflight, a beautiful online multiplayer space flight simulator. When I first saw it, I just assumed he had a background in game dev, but he told me this was his first time ever trying it—which really shows you what’s possible now with AI assistance.

Cody got his inspiration from classic games like X-Wing and No Man's Sky. He started with a simple thought: "I bet we could make something like this." And that’s where the vibe coding began.

The Process: From Gray Cube to Multiplayer Mayhem

Cody didn’t start with a grand, detailed plan. He started with a simple prompt in Cursor to create a flight simulator in space. The first version was just a starry background with a gray cube for a spaceship. From there, he just kept iterating.

1. Adding Movement: He prompted the AI to add controls for movement. This immediately brought up the realities of 3D physics and user input. He had to start thinking through concepts like movement in three dimensions and how to handle pitch, yaw, and roll.
2. Importing 3D Models: The gray cube had to go. Cody found some cool spaceship models on SketchFab and asked his AI assistant to import them. This sent him down a rabbit hole of learning about 3D model formats like GLTF and GLB. He used the AI not just to write the code but also to explain the technology. He'd ask questions like, "How do these work? How are they actually brought into the game?"
3. Solving Orientation Issues: A tricky problem came up when the imported models had no inherent 'forward' direction. The game would load them in sideways or backward. This took a lot of back-and-forth prompting to get specific. He had to tell the AI, "The front of the ship is actually a 90-degree turn to the left horizontally." It was a great lesson in how precise you need to be with your prompts for complex visual tasks.
4. Implementing Multiplayer: Once the single-player experience felt good, Cody wanted to bring in other players. He asked the AI for the best way to do this and learned about WebSockets. The AI laid out a few options, including open-source libraries or building it from scratch. As he built it, he used the AI to review his implementation, and it even caught potential race conditions and optimization issues, basically acting as a code reviewer.

This whole project was a perfect example of treating the LLM as a collaborator. Cody would learn something from a Google search about a tool like Three.js, feed that new info back to the AI, and ask it to implement the concept in his game. This back-and-forth, conversational style is what vibe coding is all about.

Workflow 2: Live Speedrun - A New Game in 15 Minutes

After breaking down the theory, Cody put it all into practice. Live on the show, he built a completely new flight simulator from a blank slate. The goal was to go from an empty folder to a working multiplayer prototype in just a few minutes.

Step 1: Project Setup

Everything started in the terminal. No boilerplate, no templates. Just one command to create a new, empty React project.

npm create vite@latest boop-flight -- --template react

Next, Cody initialized his favorite new tool, a command-line interface for Claude, right inside the project directory. He used the --yolo flag to give it permission to make any changes it saw fit without asking for confirmation.

claude init --yolo

This command scans the project, creates a claude.md file, and gives the AI all the initial context it needs on the project’s structure and framework.

Step 2: The Initial Prompt

With the project set up, Cody gave Claude a broad, creative prompt. He didn’t hand over a detailed spec sheet; instead, he just described the vibe he was going for. This is a huge part of his process—giving the LLM creative freedom within a clear set of initial constraints.

I want to build a flight simulator game that uses three js. Let's use polygon style art for the planes. WSD for movement. I want turning to bank the plane and arrow keys to control pitch. Let's have it take from a green field with trees and other objects around.

Claude Code immediately started working, figuring out which libraries it needed (like Three.js), installing dependencies, and scaffolding the entire application structure.

Step 3: First Run & Iterative Debugging

After just a few moments, the first build was done. Cody ran the app, and just like that, we had a game. It had a low-poly plane flying over a green field. But, as you’d expect with AI-generated code, it wasn’t quite right.

• The Problem: The camera was positioned in front of the plane, so when you pressed 'forward,' the plane flew towards the screen. The flight controls were also inverted.

This is where the conversational part of development shines. Cody simply told Claude what was wrong in plain English:

It looks like my camera is facing the front of the model... we should be looking should be fixed to the tail of the plane.

He followed that up with:

All of my flight controls are reversed. Now we should fix that. I also want the camera to be slightly above and behind the model.

Each time, the AI rewrote the code to fix the problem, getting closer to the goal with each iteration.

Step 4: Adding Multiplayer in Parallel

And this is where his workflow gets really interesting. While one instance of Claude Code was busy fixing the camera and controls on the front end, Cody opened a second terminal window. In this new session, he created a server directory and tasked another instance of Claude Code with building the multiplayer backend.

His prompt for the backend was just as simple:

I want to start implementing multiplayer for this game in the server directory. Handle player joins and give me a chat interface on the top right that shows when people join the game.

The AI got to work setting up an Express server and integrating Socket.io for real-time communication. It built out the logic for player connections, disconnections, and movement—all while the front-end fixes were still happening.

Step 5: The Grand Finale

After about 15 minutes of this parallel, back-and-forth development, we had something real. Cody launched the game, opened it in a few browser tabs, and sure enough, multiple planes appeared in the same world. The chat window was there (though it was at the bottom of the screen instead of the top), and you could see other players join.

The AI even added some features he didn't ask for, like a cockpit view you could zoom into and some nice-looking mountains in the distance. It wasn't perfect—the controls were still a bit off and the UI needed work—but the speed was incredible. As Cody said, "Think about how long it would've taken to learn how to build this from total scratch... this would've taken several hours, if not days... And we did this in what, 15 minutes total?"

Conclusion: Your Turn to Vibe Code

Cody’s live demo is a fantastic look at this new way of building software. It’s less about writing every single line of code perfectly and more about being a creative director for a very fast, sometimes chaotic junior developer. The process is iterative, conversational, and it lets you move at an incredible pace. We saw how you can take a general idea, break it down into smaller pieces, and use AI to build a working prototype at a speed that was hard to imagine just a couple of years ago.

This workflow isn't just for games, either. Cody mentioned he uses these same principles to build enterprise software for his day job at Sentry. The main skill is learning how to explain your vision, iterate on what the AI gives you, and guide it toward your goal. The barrier to creating complex, ambitious software has never been lower. What will you start building?

Thank You to Our Sponsors

A huge thank you to our sponsors who make this show possible:

• Enterpret—Customer SuperIntelligence Platform for Product and CX teams
• WorkOS—Make your app Enterprise Ready today with WorkOS

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts