Making a small game with AI, without writing an engine

Games break the usual advice, because there are no forms to describe. What you have to communicate instead is the loop: what the player does over and over, what makes it hard, and what makes it stop.

"A game about a cat" produces nothing playable. "The player taps to make a cat jump between shelves, one wrong jump ends the run, and each shelf reached adds a point" produces something you can try in a minute. Three sentences: action, failure, reward.

Numbers help more than adjectives. Say how fast, how many lives, how far apart. You can tune them by asking for one number at a time afterwards.

Menus, level select, saved progress and shops are easy to add later and expensive to get wrong early. Get the thirty seconds in the middle to feel right first — it is the only part players judge.

Multiplayer with real-time sync, large open worlds, and anything that depends on a specific licensed look. Not because they are impossible, but because you will spend your credits discovering that the fun was never in the part you asked for.

Buildr has an asset library of sprites, models, audio and animation you can search and attach, and you can bring your own. Using placeholder art while the loop is still moving is not laziness — replacing art is cheap, rebuilding a game around art is not.