One-Shot Game Prototypes with Claude: A Half-Hour Workflow
I made two playable browser games in about an hour total. Neither will win any awards. Neither was meant to win any awards—they were jokes, ideas I'd never have bothered with when prototyping took actual effort.
But they exist now. Playable, in a browser, with art and everything. That still feels like getting away with something.
The trick is constraints. I asked Claude for five game concepts, all top-down, all 2D vector art—because vector art is something Claude can actually generate. Five options means I'm picking, not negotiating with whatever hallucination appears first. Two looked promising in a "this is dumb but would be funny" kind of way. I asked Claude to explain how they differed, and suddenly I'm making real design decisions about games I have no business making.
From there it's almost disappointingly straightforward. PixiJS v8, single HTML file, fullscreen responsive. The single-file constraint eliminates an entire category of bikeshedding. Claude generates the first pass—code and assets. I play it. Things break.
The breaks are the interesting part. Rendering didn't trigger until gameplay started. Fullscreen scaling was wrong. Performance cratered after a minute—turns out recreating graphics objects every frame is a fantastic way to make the garbage collector question its life choices. One prototype was so aggressively bad at memory management that it reliably crashed my entire MacBook Pro. Not the tab. Not Chrome. The whole machine. A browser game. I wasn't even mad, just impressed.
Each fix is a quick conversation. The fixes compound into something that actually runs.
Then you add personality, which is where the real fun lives. Radio chatter with inside jokes nobody will get. Enemies that hunt through your tunnels like they have rent to pay. Environmental hazards that crush them—or you, if you're dumb enough to stand under a collapsing ceiling. Bore Runner turned into a Dig Dug riff with falling rocks and questionable physics. Space Burger Bar became... exactly what it sounds like.
Neither is a good game. The balance is rough. The collision detection has opinions I don't share. But that's not the point. The point is that ideas not worth investing in a year ago can now be visualized for fun and whimsy in the time it takes to watch a sitcom episode. The barrier to "let's see if this is fun" has effectively collapsed.
And the same workflow applies to serious ideas.
I've used this approach to prototype dozens of game concepts—the ones I actually care about. Most went nowhere, which is the point of prototyping. But one of them was actually fun. Fun enough that I kept going. Fun enough that it stopped being a throwaway spike and started being a real project.
That project is Blueshifted—a physics-based retro throwback inspired by Asteroids, Time Pilot, and Gyruss, updated as a modern roguelite. It's now over 110k lines of TypeScript across 1,600+ commits, entirely AI-authored under my direction. A scope that would've been impossible before I found Beads for long-horizon project management.
Will it be a great game? Probably not. It's my first serious game and I'm new to game dev. But it's the ambitious realization of one of these many throwaway prototypes—proof that sometimes the joke grows legs.
Currently in private preview. More on that soon.