Making a Game Using Only AI Prompts
By Prateek Sharma and Jana Diamond
I Tried Making a Game Using Only 7 AI Prompts — Here’s What Actually Happened
You know what sounds like a fun weekend project?
“Hey ChatGPT, make a space shooter game for me.”
You know what isn’t a fun weekend project?
Actually waiting for that AI to make a space shooter game that works.
This is the story of how I tried to build a browser-based game using only 7 prompts. No hand-written code at first. Just me, ChatGPT, and a growing sense of:
“…what have I done?”
Stage 1: The AI Demo High
Prompt 1:
“Create a simple browser-based space shooter game with HTML5 canvas, where the player controls a ship and shoots incoming asteroids.”
Result:
A working foundation! Ship movement, basic shooting mechanics, and asteroids spawning randomly. But no collision detection, no scoring, and the asteroids just... floated there menacingly.
Me: Wow, AI game dev is easy.
ChatGPT: Heh-heh. Give it a minute.
Stage 2: Reality Lands Like a Meteor
Prompt 2:
“Add collision detection, scoring system, and make asteroids explode when hit.”
Result:
Things got interesting. The collision detection worked, but the asteroid explosion animations look like pixel confetti from a 1998 PowerPoint transition. The scoring system displayed but didn't increment properly.
AI is great at building features — terrible at remembering what those features should do after 0.4 seconds.
Me: AI excels at structure but struggles with state management.
ChatGPT: Were you talking to me?
Stage 3: Tiny Wins, New Problems
Prompt 3:
“Fix scoring bug and make asteroid explosions look realistic using particle systems.”
Result:
Scoring fixed! Explosions looked decent. But now I had a new problem—the game was too easy. No difficulty progression.
Me: I accidentally built the world’s first meditation-based asteroid shooter.
ChatGPT: OHMMMMMM.
Stage 4–5: Adding Difficulty, Power-ups & Mild Emotional Damage
Two prompts later, I had:
✔ Difficulty scaling
✔ Power-ups
✔ A game-over screen
❌ Readable code
ChatGPT just duct-taped new functions to the bottom of the script like a ransom note.
Stage 6: The Audio Apocalypse
Prompt 6:
“Add background music and sound effects.”
Result:
ChatGPT added audio.
It did not add audio loading logic.
The game froze every time an MP3 loaded too slowly.
Me: Aim! Shoot!
ChatGPT: “Autoplay is illegal. Who do you think you are?”
Stage 7: Debug Mode & Accidentally Learning Stuff
Prompt 7:
“Fix the audio loading issues and add a start screen with instructions.”
Result:
Shockingly… it worked.
But only after I manually introduced two things ChatGPT didn’t:
1. A Finite State Machine (FSM)
const GameState = {
MENU: 'MENU',
LOADING: 'LOADING',
PLAY: 'PLAY',
GAMEOVER: 'GAMEOVER'
};
let state = GameState.MENU;
function setState(next) {
state = next;
}
2. Asset Loading Before Starting the Game
async function loadAudio(urls) {
const ctx = new AudioContext();
const promises = urls.map(async url => {
const res = await fetch(url);
const arr = await res.arrayBuffer();
const buf = await ctx.decodeAudioData(arr);
audioBuffers[url] = buf;
});
await Promise.all(promises);
}
What This Taught Me About AI & Development
| AI Is Great At… | AI Is Terrible At… |
|---|---|
| Starting projects | Finishing them |
| Writing functions | Managing state |
| Making demos work | Making everything work together |
| Explaining code | Understanding consequences |
| Confidence | Shame |
Things I Now Tell ChatGPT Before It Touches My Code
Build a game state machine first.
Keep score, lives, and objects in a single world object.
Preload audio and assets using Promises. No blocking.
When adding features, update existing functions — don’t duct-tape new ones to the bottom.
Difficulty should scale with time, not vibes.
Final Thoughts
Would I do this again? Absolutely.
Would I trust AI to build the whole game unsupervised? Absolutely not.
AI isn’t the game developer.
It’s the world’s fastest, most confident intern — brilliant at generating code, but you better tell it where to put things, or it will staple them to the ceiling. Or the wall. Or the dog.
Thinking about trying this yourself?
Do it. You’ll learn a ton — mostly about debugging AI’s emotions.
Originally published on Protovate.AI
Protovate builds practical AI-powered software for complex, real-world environments. Led by Brian Pollack and a global team with more than 30 years of experience, Protovate helps organizations innovate responsibly, improve efficiency, and turn emerging technology into solutions that deliver measurable impact.
Over the decades, the Protovate team has worked with organizations including NASA, Johnson & Johnson, Microsoft, Walmart, Covidien, Singtel, LG, Yahoo, and Lowe’s.
About the Authors
Share article
