Game Ideas for AI: 8 Small Games You Can Finish
The best game ideas for AI are small: one screen, clear rules, little art and no other players online. Here are eight original ideas of that size, and a quick way to pick the one you will finish.
What does an AI build well?
This is our own judgement from how these tools behave, not a measurement.
Usually goes well: anything you can state as rules in words. Grids, cards, turns, timers, menus, text, simple shapes that move.
Usually goes badly: playing with other people online, realistic physics, large 3D worlds, animated 3D characters, and anything that needs a server running all the time. Those are not impossible, but they are poor first projects, because your AI cannot play them to notice what is wrong.
Which ideas are small enough?
We invented these for this page. Take one and change it until it is yours.
| Idea | What you do | The tricky bit |
|---|---|---|
| Last Bus Home | Nine seats, twelve passengers. Decide who boards the night bus before the route ends | Passengers' moods need clear rules, or your choices feel random |
| Tide Clerk | Sort crates on a pier while the sea takes one row a minute | Dragging must feel good on a phone too |
| Seed Swap | Plant crops on a 6 by 6 plot, where each crop changes its neighbours | Teaching the rule in thirty seconds |
| Bell Ringer | Ring five bells in the order the village shouts. Wrong notes wake the baby | Timing has to feel fair |
| Ghost Landlord | Move furniture at night so that your tenants stay or leave | Tenants' reasons must be readable |
| Lost Property | Callers describe what they lost, and you pick it from a shelf | Writing forty clues that are not too easy |
| Rock Signal | Tap short and long flashes to steer boats past rocks | Keeping it fair when two boats arrive at once |
| Pocket Eclipse | Turn three discs until the shadow lines up before noon | Making sure each puzzle can be solved |
How do you pick one in five minutes?
Ask these three questions about each idea:
- Can you say it in one sentence? If you need "and also", it is too big.
- Would you play three rounds with grey squares? If it only sounds fun with art, the fun is not in the rules.
- What can you cut? A good small idea still works with half its parts. Tide Clerk survives without a shop. It does not survive without the tide.
If an idea passes all three, write it down and stop shopping for ideas. The one-page design document is your next step.
Your own idea should stay your own. A mechanic you saw elsewhere is a different matter from another game's code, art and names, so read can you copy game mechanics if you are unsure. For building blocks, browse the blueprints.
Paste this to your AI
Here is what I enjoy: [two or three things, not game titles].
Here is how much time I have: [hours per week].
Give me three game ideas of my own kind, each playable on one screen with no online features. For each: one sentence on what the player does, the smallest version, and the hardest part to build.
Be honest about difficulty. For every "this is easy", say what makes it easy, for example that this part is only rules and text. If you have no reason, write "not sure".
If an idea is close to a game that exists, say so plainly and change it until it is its own thing. I do not want a copy. If you do not know whether something similar exists, say you do not know.
For the idea I pick, draw the main loop as a Mermaid diagram before anything is built: what the player does, what happens, why they go round again. Walk it through. Show me where the loop does not close, or where a round can get stuck.
Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP
How to check it worked
- Read each idea out loud. Could a friend explain it back to you?
- Follow the loop with your finger. Can you get back to the start from every box in the drawing? If a box leads nowhere, the idea has a hole.
- Ask why the hard part is hard. A real answer names something specific. "It may be tricky" tells you nothing.
- Sketch one round on paper. Ten minutes with coins and a pencil will show whether the rules work before anything is built.