Can AI make a game?

Code. A language model can write the movement, the collision handling, the inventory, the save system, the dialogue state machine, the enemy behaviour. This is not a marginal capability — it is the majority of the typing in most small games, and it is the part that stops non-programmers from making games at all.

It is good at variations, too. Five versions of how a jump feels is a sentence each instead of an evening each, so you end up testing ideas you would otherwise have skipped.

And it is good at the tedious middle: menus, settings screens, the fourth variation on a UI panel. Nobody ever gained anything from typing those out by hand.

Given the right tools, it is good at the world too. With world tools instead of a blank scene, a model can sculpt terrain, lay water, set the sky and weather, and place a player controller and props as real Godot nodes. It can also play what it built: run the game, send input events, read the debugger and capture the viewport, which catches the broken script before you do.

More on this: How building a game with AI actually goes

Deciding what the game is. A model will happily produce a game if you ask for one vaguely, and what you get is the average of everything it has seen — competent, coherent, and nobody's idea. It has no opinion, so it produces the middle.

Judging the result. It cannot tell you that the corridor is boring, that the enemy is annoying rather than threatening, or that the jump feels wrong at this height. There is no test for those and no amount of model capability changes it, because the question is about a human experience.

And the last metre of a world. You can describe a valley and get a valley. Getting the valley you actually meant, by description alone, takes longer than shaping it yourself, and the intermediate versions are almost right, which is the most expensive kind of wrong.

More on this: Do you need to code to make a game?

It means a person decided what to build, described it in pieces, played each version, said what was wrong, and did that repeatedly until the game was right. The model did the typing. The person did the deciding.

That is a real and large change. It moves game development from a skill you have to acquire first into a thing you can start on Saturday. It is not the same as the game appearing from a sentence, and any tool implying otherwise is selling the demo rather than the product.

The word for the person in that loop is not "prompter". They are making design decisions constantly, they just are not typing the implementation.

More on this: How long does it take to make a game with AI?

Flockbay starts with a brief. It asks questions one round at a time, each with a recommendation, and writes down what gets built first, what comes after and what is not in this version. Nothing is built until you approve it, which is how the deciding stays with you.

Then it builds: the world with world tools (terrain, water, weather, buildings and characters in 3D; tile ground and sprites in 2D), the GDScript and the scenes, and a play-test of its own work. You play it next and say what is wrong. When a call is pure taste, the same world tools are buttons around the Godot editor's view, so you can shape the valley you meant by hand instead of describing it again, because almost right is worse than yours.

The engine underneath is Godot, open source, bundled in a desktop app for Windows and Linux. The game runs locally on that copy of Godot. Building needs an internet connection because the chat and world data come over the network; playing what you built does not.

More on this: The best AI for game development depends on which part you mean

Yes, if you mean that a person who has never written code can now end up with a real, playable game they designed. That was not true a few years ago and it is true now.

No, if you mean that you can type one sentence and receive something worth playing. Nothing does that, and the honest version of this answer is more useful than the exciting one.