Ship a browser game with free AI tools

Free lesson · 1 of 8

Lesson 1. Idea and scope: pick a game you can finish

Goal: a one-page spec for a game small enough to finish this week, designed so people want to play it again and show it to friends.

Most AI-made games die in the same place: the first prompt asks for too much. "An open-world RPG with crafting, quests and a day-night cycle" produces something that half-works in ten directions, and every fix breaks two other things. The games people actually finish, and the ones that spread, are tiny. One-tap arcade games, daily word puzzles, merge games and single-screen shooters all share a short list of traits, and you can design for them on purpose.

What makes a small game spread

Use this as a filter for every idea. A strong idea ticks at least four.

• Playable in two seconds. One tap, one key, or one simple rule. No tutorial screen. If you need to explain the controls in more than one line, the idea is too big for a first game.
• Short rounds, instant retry. Rounds of 3 to 10 seconds, and a restart that takes one tap and under half a second. "One more try" only happens when trying again costs nothing.
• Brutal but fair. Players should lose fast, but always know why, and always feel they could have done it. Unfair deaths make people quit; fair ones make them retry.
• A result worth sharing. A score, a short emoji grid, a "Day 41: 23 points" line. Something a player can paste into a chat in one tap.
• Everyone plays the same challenge. A daily seed means today's layout is the same for every player, so scores can be compared and bragged about.
• A look you recognise instantly. Bold minimal shapes or chunky pixels, two or three colours, one sound per action. Distinctive beats detailed.
• Tiny and instant. Loads in a browser in a second, no install, works offline. You will do this in Lesson 6.

Make it original. Borrow the traits above, never another game's name, characters, art or look.

Steps

1. Write ten ideas in ten minutes

Don't judge yet. Each idea is one sentence in this shape:

The player [does one action] to [reach a goal] while [avoiding a danger].

Examples: "The player taps to jump a dot between orbiting rings while avoiding mistimed jumps." "The player swipes to slide a block across ice to collect gems while avoiding spikes."

2. Score each idea against the spread filter

Give one point per trait it ticks from the list above. Cross out anything under 4. Then cross out anything that needs more than one screen, more than two controls, or a story to make sense.

3. Pick one and cut it in half

Take the winner and remove everything that isn't the core action. No shop, no levels, no characters with names. You can add them later; you can't easily remove them once the AI has wired them into everything.

4. Write the one-page spec

This is the most valuable ten minutes of the project. The AI will fill every gap you leave with a guess, and guesses are where bugs come from. Fill in every line:

The "Not in v1" line matters as much as the rest. It is your promise to yourself.

5. Turn the spec into a scope prompt

Before building, ask your AI assistant to attack the spec. This finds gaps cheaply, in words, before you spend a generation on code.

Prompts

Brainstorm inside the constraints

I want to make a tiny browser game I can finish in a week. Give me 10 original game ideas. Each must be playable with one tap or one key, have rounds of 3 to 10 seconds, allow an instant restart, and produce a score that is easy to share as one line of text. For each idea give: one sentence (the player does X to reach Y while avoiding Z), the single control, and what makes a round end. Do not reuse names, characters or themes from existing games.

Stress-test your spec

Here is my one-page game spec:
[paste your spec table]
Act as a game designer reviewing it before any code is written. List: (1) every rule that is ambiguous or missing a number, (2) anything that would make the first 10 seconds confusing for a new player, (3) anything that could make a round feel unfair, (4) features I should cut from version 1. Be specific and short. Don't write any code.

Shrink an idea that is too big

This game idea is too big for a first version: [paste idea]. Cut it down to the smallest version that is still fun for 30 seconds: one screen, one or two controls, one way to score and one way to lose. Tell me what you removed and why, then write the reduced idea as a one-page spec with exact numbers.

Checklist

• My game fits in one sentence: the player does X to reach Y while avoiding Z.
• It uses one or two controls, and I have written both the keyboard and the touch version.
• A round lasts seconds, not minutes, and restarting is instant.
• Every rule in my spec has a number where a number is possible.
• I know exactly why a round ends, and a player would too.
• There is a one-line result a player could share.
• My "Not in v1" list has at least three items on it.
• The name, look and theme are my own, not borrowed from an existing game.

Exercise

Write your ten ideas, score them against the spread filter, and fill in the full spec table for the winner. Then run the "Stress-test your spec" prompt and update the spec with every gap it found. Keep this file: you will paste it into your first build prompt in Lesson 2.

Stretch: write the share line before anything else. If you can't think of a share line that a friend would understand, the score isn't interesting enough yet. Change the scoring until it is.