Skip to content

AI agents and game development

By Eduardo Orellana · Updated

Agents are good at codebases that describe themselves. Game projects do not describe themselves, and that — rather than model quality — is why agentic game development mostly disappoints. Free AI to start: Flockbay's AI and your own Claude, ChatGPT, agent or API key are free for your first build; after that, a plan keeps you building. It runs on Windows, macOS and Linux, on the official Godot engine.

AppImage, x86_64. Ubuntu 22.04, Debian 12, Fedora 39 or newer.

Intel or AMD Chromebook, with Linux turned on.

Download for macOSDownload for WindowsDownload for Linux

The disappointment has a cause

Agents got good at software fast. They are noticeably worse at games, and the usual explanation — games are harder — is not quite it. Plenty of game code is simpler than the web services agents handle daily.

The real difference is that a game project carries most of its meaning outside the code. What a scene is for, which node is the player, how the terrain relates to the water, what "the village" refers to: none of that is written anywhere an agent can read. It is in a person's head and in a set of conventions that were never stated. So the agent infers, and inference on missing context is where the confident, expensive mistakes come from.

Watch a long agent session on a bare game project and this is what you see. Not incompetence. A model reconstructing the world model from scratch every few turns and getting a slightly different answer each time.

Making the game legible to the agent

An agent pointed at a bare project has to infer how the game is put together from its files, and every inference is a chance to be wrong.

Flockbay's answer is to make the game legible — to give the agent operations at the level a game is actually described in, so there is less to invent. That is what Flockbay is: Godot underneath, and a capability layer on top that turns the parts a game is made of into things an agent can operate rather than files it must interpret. In a 3D world those parts are scenes, nodes, scripts and signals: the player controller, characters and props, with terrain, water or a sky from a Godot plugin. In a flat one they are the TileMapLayer, the Parallax2D layers behind it, the player controller, and the sprites placed in it.

Every capability that exists is a chunk of code no model writes from nothing and a class of failure that stops happening.

More on this: How do I use Flockbay with my own AI agent?

Flockbay's AI, or your own agent

Flockbay's own AI ships inside the app. You describe the game, it agrees a brief with you, builds the scenes, writes GDScript into your project, and plays the result to check its work before you play it. Nothing to connect, nothing to maintain, and it is aimed at the capability layer already.

Or you bring your own agent. Flockbay ships one MCP server inside the desktop app, and an MCP-capable agent on your machine can drive it: the capability layer, the assets, and the game itself, which it can edit in the open Godot editor, run, look at and play. On Windows, Claude Code, Codex and Gemini CLI can add it with the Flockbay plugin, published on GitHub as flockbay/flockbay-plugin; any other MCP-capable agent connects to the same server. We have not tested every client and behaviour varies.

Whichever AI does the building, it builds the scenes and writes the code, and you can make any single change by hand in Godot's own editor. Either way, you judge the result by playing it.

More on this: What is an AI game maker?

The person is not being removed from this

Godot's own editor is there in the Standard and Full views: the 3D view in a 3D project, and the 2D view with Godot's TileMap tools in a 2D one, where sprites sit at an exact position, rotation and scale. The agent builds the scenes. Your hand is for the calls that are pure taste, where one drag beats a paragraph: exactly where the crate sits, which way the path bends. Godot's undo takes it back.

And the judgement stays with you. You play the game and say whether it is right. Not writing the code does not mean not having an opinion about the result — the opinion is the part you supply, and the agent revising after you say what is wrong is the normal loop rather than a sign something failed.

This is also why the pages promising an agent that produces a finished game from a paragraph tend not to survive contact with a real project. Somebody has to decide the game is good, and that somebody is playing it.

More on this: Plugins: where the manual controls live

What it runs on, and what it costs

A desktop application for Windows, macOS and Linux, with Godot bundled inside it. Your project is a real Godot project in a folder on your disk. When you press play the game runs locally, at your machine's framerate — no streaming, no cloud rendering, no browser preview.

Building needs an internet connection, because the AI and asset generation come over the network. Playing what you built does not.

Building by hand is free: the editor, your own assets, plus play, playtesting and exporting. Free AI to start: Flockbay's AI and your own Claude, ChatGPT, agent or API key are free for your first build; after that, a plan keeps you building. Flockbay builds from a library of ready-made objects, or with Godot's own shapes. Connecting your own agent over MCP is free for your first build too.

More on this: What is an AI game engine?

Questions

Why do AI agents struggle with game projects?
Because a game project carries most of its meaning outside the code, so the agent has to infer it, and inference on missing context is where confident mistakes come from.
Can an agent make an entire game for me?
Not without you. An agent can build the scenes and write the code from a brief you approved, and it can play-test its own work. A person judges the game by playing it, says what to change, and can make any taste call by hand in the Godot editor.
Do I have to use Flockbay's own AI?
No. Flockbay's AI ships in the app and is the shorter path, but an agent of your own can drive the same capability layer over MCP.
Do you integrate with specific agent tools?
On Windows, Claude Code, Codex and Gemini CLI can install the Flockbay plugin, flockbay/flockbay-plugin on GitHub, which registers the server in one command. Every other MCP-capable client reaches the same server directly. We have not tested every one.
Does using a better model fix quality problems?
Partly. What fixes most of them is the model checking its own work: running the game and reading Godot's parse and runtime errors, which Flockbay gives every model.
What does it cost to use my own agent?
Connecting your own agent is free for your first build, and a plan keeps it building after that. You pay your agent's provider for the model.

Free to download for Windows, macOS and Linux. Flockbay in the box, or bring your own agent.

Download Flockbay free