I Described a Space Game in One Sentence. 54 Seconds Later I Was Playing It.

Building Asteroid Dodge with Amplify, the AI game maker I'm making from Iceland, and what I learned from it

I'm building Amplify, an AI game maker, on my own from Iceland. The idea is simple to say and hard to get right: you describe a game in plain words, Amplify builds a playable browser game, you change it by chatting, and you publish it at its own address.

As the founder, I'm the easiest person in the world to convince that my own product works. So I set myself a small, honest test. I'd write one ordinary sentence, the kind anyone could write, and see what came out. No hand-written code, no fixing things afterward, no polishing the result before showing it to you.

This is what happened.

The prompt

I opened the Amplify studio, left the model on Standard (the default tier, which runs Claude Sonnet), and typed this:

"A small space game: steer a ship left and right with the arrow keys or by touch, dodge falling asteroids, and the score goes up the longer you survive."

That's the whole brief. There's no design document, no list of assets, and nothing about frameworks or game loops. I picked something small on purpose. A dodging game is a classic first game: easy to understand and quick to play, but it still needs real pieces to work together. You need movement, things spawning, collision, a score and a game-over state.

I also asked for touch controls in the same sentence. A browser game gets shared as a link, and plenty of people will open that link on a phone. If the game only works with a keyboard, they can't play it.

The first build: about 54 seconds

The first build took about 54 seconds and cost 31 credits.

Then I played it. The ship moves left and right with the arrow keys, and it works on touch too. Asteroids fall, you dodge them, and the score climbs the longer you stay alive. It's a small game. It isn't trying to be anything else, and that was the point.

I'll be honest about what surprised me, because I build this thing every day and I still notice it. It isn't that the AI can write code; we all know that by now. It's how short the gap is between having an idea and holding something you can play and hand to someone else. With the first version of most projects, you spend a long time in the "nothing works yet" stage. Here, that stage was under a minute.

One change: adding a pause

Every game needs a pause button the moment someone actually plays it. So I asked for one, again in plain words:

"Add a pause: pressing P or tapping the pause button stops the game and shows a Paused message."

That change took about 45 seconds and cost 29 credits. P pauses the game on a keyboard, there's a button for touch, and a "Paused" message appears on screen.

This is the part of Amplify I care about most. Generating a first version is the flashy bit, but a game is really made in the changes: the tenth tweak, the "make it a bit harder", the "the button is too small on my phone". If changing a game is as easy as describing it, people keep going instead of giving up after the first draft.

Nothing was fixed by hand

I want to be clear about this, because it's the claim that matters. I didn't open the code and fix anything. Everything in the game came from those two sentences: the first build and the pause change.

All in, that's two prompts, roughly a minute and a half of building, and 60 credits.

You can play the result here: https://small-space-game.amplify.is/

Keys: arrows (or A/D) to steer, P to pause, Enter to start. On a phone, use the on-screen buttons.

What this means (and what it doesn't)

I don't want to oversell this. Asteroid Dodge is a small 2D arcade game, not a commercial title. If you want to build a big game with a story, levels and original art, you'll still be doing a lot of directing, testing and iterating. AI changes how fast you can try ideas. It doesn't replace knowing what makes a game fun.

But I think the small end is where it gets interesting:

• Kids and first-time makers can go from "I have an idea for a game" to "play my game" without first learning a game engine.- Teachers can show game mechanics (collision, scoring, difficulty) in something students built themselves that afternoon.- Developers and designers can test whether a mechanic is fun before spending days on it.- Anyone can make a silly, personal game for a friend, which is honestly one of the best uses of a game. Every game you publish on Amplify gets its own address, so you can share it like any other link. It's also set up so search engines and AI tools can find and read it, with nothing to configure.

If you want to try it

Amplify is free to start: you get 100 free credits when you sign up, and you don't need a card. A first game on the Standard model starts at around 21 credits, depending on how big it is. Asteroid Dodge, with its pause change, used 60. If you need more, top-ups start at $5. There's no subscription.

If you'd rather not start from a blank page, https://amplify.is/games has AI-written starter templates on Amplify, such as Gem Swap, Snake and Bounce, that you can play and remix.

I'm building this in the open, and I'll keep writing here about what works, what doesn't and what I learn as Amplify grows. If you make something, even something tiny and ridiculous, I'd really like to see it.

Describe a game and see what comes out: https://amplify.is

I'm Gudmundur, the solo founder of Amplify, an AI game maker built in Iceland. You can reach me at [e-mail].