How long does it take to make a game with AI?

Realistic: a stretch of terrain or a row of platforms, a character that walks and jumps on it, and one thing that happens when you reach a particular place. Two to four hours including the download and finding out where everything is.

The front of that is the brief. The app reads your first message and preselects 2D or 3D from it, and Flockbay asks questions one round at a time, each with a recommendation, until there is a brief: what gets built first, what comes after, what is not in this version. Nothing is built until you approve it, so those minutes aim the first build at the right game. Then Flockbay builds the world, writes the code and plays it before you do. A flat game reaches the character-that-moves moment sooner than a 3D one, because there is less world to build.

Either way, that is not a game. It is the moment where the project becomes real to you, and it matters mostly because it happens at all — the same milestone by hand means learning an engine first, which is weeks.

What eats the time in that first session is not the AI. It is you deciding what the space should be, and then deciding again after you have walked through it.

More on this: A game maker for students with a deadline

A small thing with a beginning and an end. One room or one short area, one mechanic, one reason to be there. Maybe five to fifteen hours spread over evenings.

Most of that time is playing your own game and disliking parts of it. The jump is wrong. The room is too big. The thing you thought was clever is tedious the second time. Each of those is a decision that could not have been made in advance, because it depended on something that did not exist yet.

The code changes in that week are minutes each. That is the whole shift: what used to be the expensive part is now the cheap part, and the ratio of thinking to typing goes up sharply.

More on this: Make a game without coding, and without a ceiling

Something you would put in front of a stranger. Several areas, a reason to keep playing, art that holds together, sound, a menu that is not embarrassing. Call it a few months of evenings for one person, and be suspicious of anyone quoting less.

The world is a large share of that, though not in the way you might expect. Flockbay builds it — terrain, water, weather and props in a 3D world, tile ground and sprites in a flat one — so the hours are not in the building. They are in walking through it, deciding the valley is too wide, and saying so, and that is the part AI does not compress, because it is taste rather than typing. When one call is pure taste, the same tools are buttons in the Godot editor, and nudging one hill yourself beats describing it.

Polish is the other share, and it is the part everyone underestimates in every medium. The last ten per cent of quality is not ten per cent of the time.

More on this: Making a game with no experience: what you skip and what you do not

The learning curve before you start. That is the largest single saving and it is invisible in any comparison that starts from "an experienced developer would take".

The implementation cost per idea. Trying a variation stops being a decision about whether it is worth an evening, so more ideas get tested, which usually improves the game more than the speed does.

And the debugging you would have done on your own mistakes. Some of it comes back as debugging the model's, which is a real cost and a smaller one.

The decisions about the world. Flockbay sculpts the terrain, lays the water, sets weather and sky and places the props in a 3D project, and lays the tile ground and sets each sprite at an exact position, rotation and scale in a 2D one. What that world should be is still yours: which way the river runs, how big the room is, where the ledge sits. The brief settles the first version; every later one is a round of playing and saying what is wrong.

Judging the result. You play the game and say whether it is right, and being handed work back is the normal loop rather than a failure. Not writing the code does not mean not having an opinion about the game.

Practically: it is a desktop app for Windows, macOS and Linux with Godot inside it, the game runs locally, and building needs an internet connection while playing does not. The download is free so finding out how long your game takes costs nothing but the evenings.

More on this: An AI game maker for solo devs: one person, several jobs