25 million plays. That's what three browser games built by someone with zero coding experience three months ago have accumulated.
The creator's starting point was a natural language description of a color memory game, submitted to Claude through the Cursor IDE. The prompt described the mechanic: show a color, randomize some sliders, let the user try to recreate it, score the result side by side. Six hours later, a working game existed. Twelve hours in, it was polished enough to ship. Two more games followed the same pattern.
Two of the three games are single HTML files around 8,000 lines long. Not because that's good engineering practice - it isn't - but because that's what emerges when you describe what you want and let an AI write everything into one document. For context, 8,000 lines is roughly the scale of a moderate professional web application. All in one file.
The 8,000-Line File Problem
Professional developers write modular code: separate files for different functions, reusable components, organized structure. AI coding assistants often skip this. They produce working code, but it tends to sprawl - everything packed into as few files as possible, with minimal organization beneath the surface.
For a side project that just needs to run in a browser, this is fine. For anything that needs to be maintained, extended with new features, or handed to another developer, it creates real problems. The 8,000-line file works until you need to change something structural, at which point you're asking Claude to navigate a document the size of a short novel.
A developer with experience knows when to push back on these architectural choices. A first-time coder doesn't yet know what to push back on - and in this case, it didn't matter.
What 25 Million Plays Actually Tells Us
The play count matters more than the code quality here. Building something technically functional is one thing. Getting 25 million people to play your games requires the games to actually be fun - good feedback loops, satisfying mechanics, a reason to replay. The creator had the instinct for what would be engaging. Claude handled the syntax.
This is the real division of labor in AI-assisted coding for non-developers: you supply the idea, the taste, and the judgment about what works. The AI handles implementation. The people who do well with this workflow tend to have clear opinions about what they're building and can describe it precisely enough to get useful output on the first pass.
The practical ceiling is complexity. Self-contained browser games can live forever as single HTML files - nobody needs to maintain or extend them. Anything requiring a backend, user accounts, or regular updates will push against the limits of the "describe and ship" approach faster than most first-time builders expect.