CrazyGames rejected Drift Lane, the game Claude Code built from a 7-line PRD
Drift Lane is a Three.js browser game I built with Claude Code from a 7-line PRD. The first playable version took about 30 minutes, and after that I spent a few hours iterating on it. The build is 4.9 MB. I submitted it to CrazyGames one evening, the same portal that accepted my arena shooter Bolt Pit, and the rejection was in my inbox the next morning.
Posts about AI-built games usually stop at how fast the game was made. Drift Lane was fast to make, and this post is about what happened after that, when a portal reviewer judged the finished game.
What the rejection email said
CrazyGames gave me one sentence of feedback that was specific to Drift Lane: “The overall quality of the game does not yet meet the expectations of our platform.”
The rest of the email listed the usual reasons CrazyGames turns a game down: low gameplay quality, broken builds, copyright, or missing integration requirements. The email didn’t say which of those applied to Drift Lane, so I’m not sure why it was rejected. My guess is that the game was too similar to other games and generally not built that well. That’s a guess, and CrazyGames didn’t confirm it.
Bolt Pit passed the same review
Bolt Pit is a twin-stick arena shooter that also runs in the browser on Three.js and was also built with Claude Code. I submitted it to CrazyGames on October 14, and it passed review and went live on October 15 as a Basic Launch. A Basic Launch is CrazyGames’ test phase: their team reviews the build by hand, and if it’s accepted it goes live to a limited audience with no ads and no SDK. Games that do well in that test get invited to a Full Launch, which is where ad revenue starts.
So I had two Three.js games built with the same agent, going through the same manual review, and one got in while the other didn’t.
Passing review also isn’t the finish line. By October 18 Bolt Pit had 664 plays from 628 players, it had earned $0 because the Basic Launch has ads off, and fewer than 2 in 100 players came back the next day. I wrote about that scorecard in a separate post. Getting accepted only means CrazyGames lets strangers try the game, and then their numbers decide what happens next.
The 7-line PRD
This is the whole PRD I gave Claude Code:
Core concept: An arcade drifting game where you navigate narrow, elevated tracks, avoid obstacles, and collect coins.
What stands out:
- Simple controls: Steering and drifting are the main mechanics.
- Risk/reward: Drifting close to the edges creates tension.
- Short levels: Ideal for quick mobile sessions.
- Visual feedback: Persistent tire marks make every turn feel satisfying.
- Minimalist 3D style: Relatively straightforward to recreate in Unity or Three.js.It describes a type of game and the feeling it should have, and Claude Code turned it into something playable in about 30 minutes. After the first build I played it myself and changed a few things. I found it very hard at first, so the tracks got wider and the early corners gentler. Grazing the edge killed you instantly, so now you get a moment to steer back. The left and right buttons were hard to hit, so they became a floating joystick.
Get Frederik Roessell’s stories in your inbox
Join Medium for free to get updates from this writer.
You can paste the PRD into Claude Code or Codex and see what you get.
I already knew it wasn’t ready
On October 21 I felt Drift Lane was okay but not launch-ready, and I submitted it that night anyway. The rejection lines up with that feeling and with my guess about the reasons. A drifting game on narrow tracks is a familiar idea, and seven lines don’t say much about what would make this one different from the others. A few hours of iterating fixed the things I noticed while playing, and it didn’t get the game to a level I’d have called launch-ready myself.
For designers and indie builders shipping with AI, the short PRD did its job. It told me quickly that the idea was worth playing. What it couldn’t do was make the game good enough for a portal that reviews every submission by hand.
Porting Drift Lane to Unity
I still like Drift Lane, so I’m porting it to Unity and publishing it on the App Store. I won’t resubmit it to CrazyGames, because for this game it isn’t worth it.
Unity is where I built Mowtopia, a cozy game where you mow tiny round planets. Mowtopia came from a PRD with seven milestones instead of seven lines, and Claude Code built most of it in two days with 60 passing tests.
The App Store has its own review, and Apple has already rejected one of my games, 2048 Meadow, under its design spam guideline for being too similar to other games in the genre. That’s the same weakness I suspect in Drift Lane, so the Unity version has to feel like its own game before I send it to Apple.
If you’re shipping AI-built games to portals, play your own game the way a stranger would before you submit it, and ask yourself whether you’d call it launch-ready. On October 21 I called Drift Lane okay but not launch-ready, and CrazyGames reached the same conclusion the next morning.
I send these by email too: https://froessell.substack.com
