Vibe coding a 3D multiplayer game in 15 minutes—with no game dev experience | Cody De Arkland
Cody De Arkland (guest), Claire Vo (host)
In this episode of How I AI, featuring Cody De Arkland and Claire Vo, Vibe coding a 3D multiplayer game in 15 minutes—with no game dev experience | Cody De Arkland explores vibe coding a 3D multiplayer flight game using AI tools fast Cody walks through how he starts with a blank Vite/React project and uses Claude Code (plus tools like Cursor/Windsurf) as “junior developers” to scaffold a Three.js flight simulator from a short prompt.
Vibe coding a 3D multiplayer flight game using AI tools fast
Cody walks through how he starts with a blank Vite/React project and uses Claude Code (plus tools like Cursor/Windsurf) as “junior developers” to scaffold a Three.js flight simulator from a short prompt.
They show the reality of AI-assisted building: rapid progress to a fun v0, followed by iterative debugging for camera orientation, inverted controls, and unintended features the model adds.
Cody explains how he chooses technologies (ask the LLM, then validate via Google/docs, then feed findings back) and how that mirrors building production apps: broad concept → feature-by-feature iteration.
They also prototype multiplayer quickly with a simple server + WebSockets/Socket tooling, emphasizing speed/velocity while acknowledging the need for resets, constraints, and tighter scoped prompts when the AI goes off track.
Key Takeaways
Start broad, then iterate feature-by-feature.
Cody avoids overly detailed “one-shot” specs; he prompts for a workable v0 first, then refines discrete parts (controls, camera position, art style, environment) in successive passes.
Get the full analysis with uListen AI
Use the LLM to pick tech, then verify externally.
He asks AI which browser-game technologies make sense (e. ...
Get the full analysis with uListen AI
Treat AI tools like junior devs with parallel tasks.
Cody keeps multiple AI coding environments open and can run “dueling” Claude instances—one fixing frontend gameplay while another scaffolds backend multiplayer—to compress time-to-prototype.
Get the full analysis with uListen AI
Expect “two steps forward, one step back.”
The demo surfaces common failure modes: reversed controls, misinterpreted camera direction, UI shifted off-screen, and extra features added without request—yet the net progress is still faster than manual building.
Get the full analysis with uListen AI
3D assets require explicit orientation and “forward” definitions.
Imported models (glTF/GLB from places like Sketchfab) may load sideways or rotated; you must specify what “nose/forward” means and apply consistent transforms you can reuse for additional ships.
Get the full analysis with uListen AI
Multiplayer is approachable as a v0 with WebSockets, but needs review.
AI can scaffold joins/chat/movement quickly, but Cody still validates design and correctness (race conditions, disconnect handling, optimization), using the LLM to review his own code.
Get the full analysis with uListen AI
When the model goes off-track, reset and reduce scope.
His main recovery tactic is to restart a problematic section with a short list of requirements, simplify layered solutions, and focus prompts on one broken area (e. ...
Get the full analysis with uListen AI
Notable Quotes
“I tend to look at the different tools as like little junior developers who are helping me work on different things.”
— Cody De Arkland
“I try to stay pretty wide… what’s the broad strokes… then I can tweak the individual parts as I go.”
— Cody De Arkland
“You wrote, like, 27 words into this prompt, and now you have a video game?”
— Claire Vo
“This is a good example… things don’t always work the way we expect.”
— Cody De Arkland
“Hey, we’ve been wrong for a while. Can we take a fresh look at this problem? Here are the main requirements…”
— Cody De Arkland
Questions Answered in This Episode
In your initial prompt, what “broad strokes” are most important to include so Three.js + controls + environment come out sane, and what details should you intentionally omit until later?
Cody walks through how he starts with a blank Vite/React project and uses Claude Code (plus tools like Cursor/Windsurf) as “junior developers” to scaffold a Three. ...
Get the full analysis with uListen AI
When the camera/controls were reversed, how would you systematically diagnose whether the issue is model orientation, camera transform, or input mapping—before asking the LLM to change code?
They show the reality of AI-assisted building: rapid progress to a fun v0, followed by iterative debugging for camera orientation, inverted controls, and unintended features the model adds.
Get the full analysis with uListen AI
You mentioned importing ships from Sketchfab: what concrete normalization steps (scale, pivot, forward axis, collision bounds) would you standardize into a repeatable pipeline?
Cody explains how he chooses technologies (ask the LLM, then validate via Google/docs, then feed findings back) and how that mirrors building production apps: broad concept → feature-by-feature iteration.
Get the full analysis with uListen AI
For multiplayer, why choose raw WebSockets/Socket-style approaches over alternatives (WebRTC, hosted realtime services, game networking libs), and what trade-offs showed up immediately?
They also prototype multiplayer quickly with a simple server + WebSockets/Socket tooling, emphasizing speed/velocity while acknowledging the need for resets, constraints, and tighter scoped prompts when the AI goes off track.
Get the full analysis with uListen AI
Claude added unrequested features (cockpit view, mountains, zoom behavior). What rules, constraints, or “definition of done” checks would you add to prevent scope creep in future iterations?
Get the full analysis with uListen AI
Transcript Preview
we're talking about games, and we're talking about building games here, but the same thing translates really well to, like, when you're building actual applications, too. A lot of times you're starting with this blank framework, and you're giving it, like, a broad idea of the thing you wanna make, and then you're diving into these individual features.
What are your sources for figuring out how to scaffold with existing technologies?
A lot of times I'll ask the AI, "If I wanted to build a game, and I wanted it to run inside of a browser, which technologies make the most sense?" But then picking those things out and starting to do deep dives on it, like traditional Google, then feeding that back in the LLM, so it almost becomes like a conversation with another developer, where you're like, "Hey, I learned this thing from the internet. Can you implement this in the game?"
I just think that's a really interesting process that nets out net positive.
We're in this time period where everyone can go and do this. My kids have sat down and started playing with building games and things like that.
So we're gonna speed-run vibe coding this game.
We are going to speed-run. I want to build a flight simulator. I want turning to bank the plane, and arrow keys to control pitch. And there's our game.
Hold your horses. You wrote, like, 27 words into this prompt, and now you have a video game? [upbeat music] Welcome to How I AI. I'm Claire, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, we have a very fun conversation with Cody De Arkland, senior director of developer experience at Sentry. Cody is one of the most prolific vibe coders I know, doing everything from building personal to-do apps for his family to automating just about everything you could automate at work. But today, we're doing something extra fun. Cody's going to speed-run building a 3D multiplayer game live on the show. Let's get to it. This episode is brought to you by Enterpret. Enterpret is a customer intelligence platform used by leading CX and product orgs like Canva, Notion, Strava, Hinge, and Linear to leverage the voice of the customer and build best-in-class products. Enterpret unifies all customer conversations in real time, from Gong recordings to Zendesk tickets to Twitter threads, and makes it available for your team for analysis. What makes Enterpret unique is its ability to build and update a customer-specific knowledge graph that provides the most granular and accurate categorization of all customer feedback, and connects that feedback to critical metrics like revenue and CSAT. If modernizing your Voice of the Customer program to a generational upgrade is a 2025 priority, like customer-centric industry leaders Canva, Notion, and Linear, reach out to the team at enterpret.com/howiai. That's E-N-T-E-R-P-R-E-T.com/howiai. Okay, Cody, I hate to admit when people vibe code harder than I do, but I believe [chuckles] that you are one of a very few set of people who do vibe code more than I do. So tell me the, the truth. What is up on your screen right now? What are you working on?
Install uListen to search the full transcript and get AI-powered insights
Get Full TranscriptGet more from every podcast
AI summaries, searchable transcripts, and fact-checking. Free forever.
Add to Chrome