I Dared Claude to Build a 4th of July Game in One Shot. Play It Here.

I have been putting Claude Fable 5 through real work all week, and it keeps clearing bars I expected it to miss. So with the 4th of July coming up, I decided to stop giving it work and start giving it a dare.

The prompt was simple to write and, I thought, brutal to satisfy. Build the most amazing web-based game you can to celebrate the 4th of July. And do it under these rules.

And then the part that made it a real test: write the article, publish it, and merge it, so that when you are done I can go to the website and play it right there. No staging link. No draft for me to fix up. The finished thing, live.

The game Fable invented is called Grand Finale, and the idea is that you are not watching the fireworks show. You are running it.

Shells launch themselves from the skyline and climb into the night. Your only job is to burst each one at the right moment, with a click or a tap. The higher the shell is when you burst it, the more the crowd loves it. Burst it inside the blue sweet band near the top of the sky and it is a PERFECT worth 500 points. Catch it mid-climb and it is a GREAT worth 250. Panic-pop it low and you get 100 and a polite golf clap.

Three systems stack on top of that. Burst shells back to back while the last bloom is still in the air and you build a combo multiplier, up to five times your points. Every good burst also fills the crowd hype meter, and when it maxes out the game tips into FINALE mode: the sky fills with shells and every burst is worth double. And if you let a shell fall without bursting it, it becomes a dud, the crowd groans, your hype drops, and your combo dies.

A show lasts sixty seconds. When it ends, if you earned a spot, you punch in three initials arcade-style, and the all-time board next to the game keeps your score for good. The first name on that board is FBL with 7,500 points, because it felt right that you should have to beat the model that built the game. The share buttons are right there when you do.

Here is what happened between me hitting enter and this page existing, with nobody touching a file in between.

Fable started by reading this website's own codebase to learn how our blog works: how articles are defined, how pages get their social cards, how every post automatically ships a Markdown twin for AI readers. Then it designed the game, wrote it as a single self-contained component of about 950 lines, and wired it into the article system so it renders right at the top of this post.

The graphics are hand-rolled. There is no game engine and there are no libraries: an HTML canvas, a particle system that throws around a few hundred sparks per burst, a twinkling starfield, a silhouette skyline, and screen shake on a perfect hit. Even the sound is synthesized in the browser with the Web Audio API, so the pops and crackles you hear were never recorded by anyone. It works with a mouse on desktop and bigger touch targets on phones.

On rule four, honesty matters. Every game borrows physics from somewhere, and tapping things in the sky is as old as arcades. But this exact design, altitude bands for scoring, combos chained through still-blooming bursts, a crowd meter that unlocks a double-point finale, and duds that punish neglect, was invented for this page. It is not a reskin of anything you have played.

Then it wrote this article, generated the title card, opened the pull request, merged it, and verified the page was live in production. The article you are reading is part of the deliverable, which is a strange sentence to type inside of it.

I did not run this dare because Masset is pivoting to video games. I ran it because a game is the hardest kind of one-shot: it needs design taste, working physics, sound, state, persistence, mobile support, and a reason for anyone to care, all at once. If one prompt can clear that bar, it can clear the bar your team actually cares about.

Think about what this same motion looks like on your website. The ROI calculator you have wanted for two quarters. The interactive product picker. The assessment tool that turns readers into leads. Those are all easier than a playable fireworks show, and they have been sitting in your backlog because interactive content used to mean a developer, a designer, and a sprint.

One more thing made this possible, and it is worth saying out loud: the site itself was built to be worked on by AI. The blog knows how to give any new article a social card, an AI-readable twin, and a sitemap entry automatically. Fable did not fight the website. It read it, understood it, and extended it. That is what we mean when we say your content should be a home your AI can actually work in.

Now go set a high score. FBL is sitting at 7,500, and I have already lost to it twice.