Puzzle games need a prototype in an hour, not a fortnight

Puzzle games are unusual. There is no story to carry a weak core, no combat feel to fall back on, no world to explore while the mechanic warms up. Either the rule is interesting or the game is not.

And you cannot know from the description. A mechanic that sounds brilliant written down often turns out to have one obvious solution that applies everywhere. A mechanic that sounds trivial sometimes has forty good levels in it. The only way to find out is to play it.

So the thing that matters most in this genre is how long it takes to get from an idea to something playable. If that gap is two weeks of coding, you will test three ideas a year. If it is an hour, you will test three ideas a week.

You describe the rule. Blocks slide until they hit something. Two blocks of the same colour touching merge. The player can only move when the light is on. Flockbay writes the GDScript in your project and you play it, using rough shapes if that is the fastest honest route to something playable.

Then the interesting thing happens: you discover what the rule actually does, which is never quite what you thought. You add a condition, remove one, invert one. Each change is a sentence.

That iteration rate is the real product. Not writing the code has never meant not designing the game.

Once the mechanic works, the game becomes a sequence of levels, and that sequence is a teaching curriculum. Level one shows the rule. Level four shows an implication of the rule you did not state. Level twelve combines two implications. Level twenty is the one people talk about.

That ordering cannot be generated, because it depends on a model of what the player has understood so far. You build it by hand, playing each level as if you did not already know the answer.

The builder is built for this: place, play, move something, play again.

More on this: AI stealth game maker: the guard has to be readable

Which one you are making is the first question the app asks: 2D or 3D, shown as two pictures, with no default and nothing created until you pick. A grid-based puzzler picks 2D and gets a flat world's own tools, playable from the moment the project opens.

Ground there is a tile layer painted from a coverage map. You drag to lay cells and drag to erase them, and the tiles, the edges and the collision all fall out of which cells are ground, so the board you drew is a board the player can stand on. Paint the same cells in a different order and you get the same board; a rebuild never reshuffles the edges you already accepted. Whether a material stops a walker at all is a separate declaration, so a wall and a floor tile can be drawn with identical strokes and only one of them blocks anybody.

The pieces on the board come from the Catalogue, which is mostly sprites: 55,073 of its 60,648 objects, searchable by name with a picture of each. You place one at an exact position, turn and size, and it sits on its own drawn bottom-centre rather than the middle of its file, so a block sits on its tile instead of floating above it.

Pick 3D and you get that side's tools instead, which suit spatial puzzles: perspective tricks, rooms that reconfigure, a puzzle built into a landscape with real terrain and weather. Same app, same builder, the same chat writing the rule underneath either one.

Art comes from generation, the web, or your own hand, and clean geometry built directly is often the right answer for this genre, where legibility beats detail.

More on this: AI game prototyping, on an engine the prototype can survive on

Flockbay is a desktop app for Windows with Godot bundled inside. Your game runs locally when you press play. It is a download, not a browser tool.

Building needs an internet connection, since the chat and asset generation come over the network. Playing does not. The download is free and the builder costs nothing to use. Flockbay is free on your own AI: your Claude or ChatGPT plan, your own agent, or an API key. Flockbay builds from a library of ready-made objects, or with Godot's own shapes.

More on this: AI strategy game maker for people with a rule set in mind

You cannot judge the difficulty of your own puzzle, because you know the answer and cannot unknow it. Every puzzle you make will feel obvious to you and some of them will be impenetrable to everyone else.

So the essential and unglamorous step is to put it in front of someone and shut up. Do not hint. Watch where they hesitate, what they try first, and which rule they assume exists that does not. Twenty minutes of that is worth a week of thinking about it.

What the tool can do is make acting on what you learned cheap. You saw that everyone tries to push the block instead of pulling it, so you either change the level or change the rule, and either is a short conversation followed by playing it again.

More on this: AI logic puzzle maker: the rule is small, the consequences are not