AI Game Dev
AI is a tool in your hands, not a developer you hired.
Used well, AI removes most of what used to make game development slow: boilerplate code, first-draft art, dialogue for fifty NPCs, translation into four languages. Used badly, it produces a pile of plausible code nobody understands. The difference is entirely in what you ask for and what you check.
Start from a minimum playable game
One rule One screen One goal
A minimum playable game — an MVP — is not a demo or a prototype. It is the smallest thing that a stranger can pick up, understand in ten seconds, and enjoy for two minutes. One mechanic, one screen, one clear goal. If you can't name the single moment that makes it fun, the game is not small enough yet.
Publishing that small thing is what changes everything. Real players do what you never predicted, quit where you thought they'd stay, and tell you in one day what six months of solo work would not. Shipping is not the end of development — it's the first honest feedback you get.
Decide your publishing date before you decide your feature list. Then cut features until they fit. A game that ships in three weeks and gets played beats a masterpiece that never leaves your machine.
A game runs in two parts — client and server
Everything a player sees and touches is handled by the client running on their device. Connecting players and keeping records safe is the job of a server somewhere on the internet. Thinking of them separately makes it clear what to hand to AI and what to check yourself.
Client
Drawing the screen, keyboard and touch input, sound, physics and collisions — the part that actually runs the game every frame. For a web game the browser is the runtime, so it opens from a link with nothing to install.
- Babylon.js
- Phaser
- Unity
- Godot
- Flutter Flame
- Unreal Engine
Server
Accounts and sign-in, save data, leaderboards, chat, real-time sync for multiplayer — the part that ties many people into one game. Most are run as Docker containers.
- Nakama
- SpacetimeDB
- Colyseus
- Docker
A single-player web game is complete with just a client. When you need cloud saves or a leaderboard, look at the SDK of a game platform such as Cadeplay before standing up your own server. Running servers can wait until your game has players.
What AI actually does in game development
Not "it makes the game". These are the specific jobs where handing work to AI genuinely pays off.
The four-beat loop of building with AI
- 1 Describe Write exactly what should move and how — keys, speed, how you win and lose.
- 2 Generate AI writes the code. Ask for one feature at a time, not the whole game at once.
- 3 Play it Run it and find what feels wrong. Copy any error message exactly as it appears.
- 4 Report back Pass on what you saw and the exact error, and ask for a fix. Repeat these four beats in small, frequent loops.
Six jobs worth handing to AI
What a good request looks like
A vague request gets vague code back. Include the behavior, the conditions and how you will check it, as in these examples.
Make a game in a single HTML file. Pressing Space makes the character jump, and obstacles come in from the right, getting faster. On a collision, show the score and restart with R. On phones, tapping the screen jumps.
If I press jump twice quickly, the character jumps again in mid-air and flies off the screen. Before fixing it, explain which code causes it.
Deploy this game to Cadeplay. Before uploading, check that the zip layout meets the rules, and give me the play address when you're done.
People who understand how a model picks its output get dramatically better results from the same tool. They know why the same request gives different answers, why a long conversation gets worse instead of better, and why adding one example changes everything. Without that, you are pressing buttons and hoping.
You don't need months. The beginner track on this site covers what a model is, why it answers in probabilities, and how to write a request that gets what you meant. Come back here after that and every page will read differently.
Start with AI basics What AI cannot do for you
An AI is a machine running logic over probabilities. It has never been bored, never felt a game click, never wanted to play one more round. Everything below depends on exactly that.
Once it's playable, put it on Cadeplay
Cadeplay is a web game platform where games play instantly in the browser, with nothing to install. Upload your game as a single zip and you get a permanent address, with cloud saves and leaderboards available too. It's free, and you keep the rights to your game.
Install the skill, and "deploy it" is all you say
The Cadeplay skill is a bundle of instructions that teaches an AI coding tool the entire publishing procedure. Once it's installed, the tool zips the build, uploads it, waits for the check and makes it live — in order — then reports back.
Deploy this game to Cadeplay
Zipped from inside the build folder Uploaded — the automatic check passed Promoted the new build to live Drafted the store description
What to do this week
- 1 Write down, in one sentence, the single moment that makes your game fun. If you can't, you don't have a game yet — you have a setting.
- 2 Pick a publishing date three to four weeks out. Write it somewhere you'll see it.
- 3 Ask an AI to draft the core loop as code, then run it. Not the whole game — just the one fun moment, playable.
- 4 Put it in front of one person who has never seen it. Say nothing. Watch where they hesitate.
- 5 Fix the one thing that made them hesitate. Then ship it — with the Cadeplay skill, "deploy it" gets you a play address you can send as a single link.
Build a game right here, right now
Reading alone won't give you the feel. Type the game you want below, and the AI tool writes the code — play it right beside the chat. To keep going on a bigger screen with the code in view, practise in the Vibe Coding menu.
Nothing built yet. Describe something or pick an example.
- Start from an example