I’ve been vibe coding games for the past few months, and I’ve reached a conclusion that surprised me: frontier language models can now design fun gameplay. This isn’t just about generating code or assets anymore. Claude Opus 5.5 in particular understands what makes a game enjoyable to play.
I know this sounds hyperbolic. Models have been able to create something that looks like a game for quite a while. The real challenge has always been game design itself. Creating a compelling gameplay loop, balancing difficulty curves, and producing something that’s genuinely satisfying to play is a craft I’ve always respected too much to think machines could handle it. Game designers are artists.
Then I tried telling Claude Opus 5.5 to “Make it a LOT more fun” while working on Battle Drones, my Command & Conquer-style RTS project. It worked. The model genuinely improved the game design.
The Four Games That Changed My Mind
Over the past few weeks, I’ve vibe coded four different games that I’ve actually enjoyed playing. Tree Appreciator is a charming spot-the-difference game built in a single prompt using three.js. The RPG demo evolved from a simple engine into a full 15-minute story experience that emotionally affected me by the end, despite every asset being AI-generated. Battle Drones became a multiplayer RTS with seven maps, a random generator, and AI opponents that can compete against each other. The Half Moon Bay Pumpkin Farming Simulator is still in development but represents something genuinely novel: a weird, personal game that wouldn’t have been built in the before-times.
None of these projects required more than four hours of my creative input. The rest was Claude Code churning away, iterating based on my feedback about what wasn’t working. I didn’t debug complex AI pathfinding or design intricate balance mechanics myself. I described what I wanted and the model handled the craft.
I attended Foo Camp recently and discovered I wasn’t alone. Multiple developers reported similar experiences. Building games that actually play well is shockingly easy now. The new bottleneck isn’t technical execution. It’s conceiving of novel ideas that are worth building in the first place.
What This Means for Game Development
This creates an interesting split in the industry. Building a computer game used to be one of the most challenging projects in all of software development, requiring diverse skills and intense focus over months or years. Now I can spit out a playable multiplayer RTS by prompting on my phone.
Some players and developers will reject AI-generated games entirely. I understand that position, even though I don’t hold it myself. The games industry already has aversion to AI-generated assets through tools like Midjourney. We now have games where everything—code, assets, dialog, game design—comes from models. The history of tooling in creative fields suggests this will result in a genuine split: some will refuse AI involvement entirely, while others will embrace it as democratizing game development to a much wider audience.
I certainly won’t submit vibe coded games to traditional competitions. I don’t feel I’ve earned that right. But I expect new venues will emerge specifically for AI-assisted game development, the same way new categories emerged when 3D engines democratized game development decades ago.
The Real Opportunity
The exciting part is imagining what happens when someone takes this seriously. I’ve gotten interesting results from minimal creative input. What could someone achieve who invested significant time in game design, applied rigorous playtesting, iterated until their vision was polished, and maintained high standards for taste and novelty throughout?
The barrier to entry for game development has dropped dramatically. That’s either terrifying or liberating depending on your perspective. I lean toward liberation. Modern browsers are incredibly capable, frontier models know three.js and Phaser, and static hosting is free. The technical infrastructure has never been more accessible.
Vibe coding games works precisely because it’s low stakes. The damage caused by bugs is minimal compared to mission-critical systems. This makes it the perfect testing ground for exploring what AI can do in creative domains. Building something fun and personal without needing a team of specialists for the first time in history changes what’s possible.
The games I’ve built so far are mostly proof-of-concept projects. But they demonstrate something important: the creative bottleneck in game development has shifted from execution to conception. The question isn’t whether AI can build a game anymore. The question is whether you can imagine a game worth building.