Make a game with AI without giving up the game

Not "can AI make a game" but "which part of making a game do I want handed off". Those have very different answers.

Most people who want to make a game already know roughly what they want. They stop because between the idea and the thing you can play there is a programming language, an engine with a thousand menus, and about two hundred hours of learning that has nothing to do with their idea.

That gap is the thing to remove. Not the design, not the taste, not the deciding — those are the reasons to make a game at all.

It starts with questions, one round at a time, each with a recommendation, and then a brief: what gets built first, what comes after, what is not in this version. Nothing is built until you approve it. Ask for something enormous and you get the truth and a smaller first version, not a fake yes.

Then it builds. The world goes in with world tools, as real nodes and resources in the scene tree. The behaviour goes in as GDScript inside your project: movement, interaction, inventory, enemy behaviour, save points, whatever the game needs. Then Flockbay runs the game, sends input events, reads stdout and the debugger, and captures the viewport, to check its own work before you see it.

The code is real code in real files in a real Godot project. You do not have to read it. Nothing is hidden from you either — it is not a black box format that only this app can open.

What you do instead of reading is play. You are the one who decides whether the result is right, and you will often decide it is not. Saying so and getting it changed is how the tool is meant to be used.

More on this: A 3D game maker for people who bring their own Claude or ChatGPT plan

What Flockbay builds depends on the game. In 3D it is terrain, sculpted and painted, water, sky, weather, plants, the player controller, characters, people and props. In 2D it is a sky colour and a band of scenery behind everything, tile ground, and sprites at an exact position, rotation and scale. The tiles, the edges and the collision all fall out of which cells are ground, so painting the same cells in a different order draws the same world, and a rebuild never reshuffles an edge you already accepted.

Generating a world from a description usually gives you someone else's guess at your game. The brief is what takes the guess out: you have said what goes in this version before anything is built. What is left is taste, and taste is the one thing a sentence is slow at. Moving an object is a drag; describing where to move it is a paragraph.

So the same world tools are buttons around the Godot editor's own 3D and 2D view: Sculpt, Paint, Plants, Place, Water, Sky, Weather, Player controller, NPCs and Camera in 3D, and Paint, Place and Player controller in 2D. Nudge the hill, move the tree, repaint the path, and Godot's undo takes it back. You never have to touch them. A sentence gets you a valley; your hand can get you the exact valley.

Art comes from a library of ready-made objects, from Godot's own shapes when that is the straightest line to something playable, from generated pictures and sound, or from your own files.

More on this: A 3D game generator, and what actually gets generated

A desktop app for Windows, macOS and Linux, downloaded and installed. Open-source Godot bundled inside, and the editor is Godot's own, with the chat in a dock. Games run locally on your hardware. Nothing streams.

Building needs an internet connection — the AI, generation and world data are all network services. Playing does not.

Free to download, and every world tool is free. Flockbay is free, AI included: free hosted AI every day, or your Claude or ChatGPT plan, your own agent, or an API key. It is free for one person or a team.

More on this: Make a first person game where the world holds up close

The first hour is the brief and the first build, and it is fast and slightly unnerving, because things you expected to take days work immediately. The second day is slower, because you start caring about specifics and specifics take iteration.

That second phase is the real work and it is the good part. You are playing your own game and finding it not quite right and fixing it, which is what making a game is, and which no amount of AI removes or should.

It will not decide what your game is. Ask for something vague and you get questions back, each with a recommendation, because a plausible generic answer would be worse than useless: it looks like progress.

It will not tell you the game is boring. That verdict only comes from you playing it, and it is the most valuable piece of information in the whole process.

And it will not build a game nobody approved. Nothing goes into the scene tree until the brief says it should, and a too-big ask gets a smaller first version rather than a starter world dressed up as your idea. The tool is deliberately narrower than the marketing category it sits in, and the narrowness is what makes it work.

More on this: AI 3D game maker for people who want scale