AI Agents and Game Dev

Smart people suggest that I focus on "closing the loop". A closed loop is a set of tools, criteria, and output that allows autonomous planning, developing, and testing. I've closed the loop to varying degrees of success for a couple key moments.

The complexity is multiplied due to the requirements of multiplayer games. It's not just one Godot binary but a linux server and clients for windows, osx, and the linux for the steam deck. Externalizing the testing logic from the game logic through a tcp server enables timing specific actions across the clients, such as creating a room with one client and joining the room with the other.

Closing the Orchestration Loop

I'm new to multiplayer game development, but not to devops. There are probably better ways to do things

Since this is multiplayer game, I run three instances of it, two clients and one server with a --server parameter.

One experiment I want to try is AI reviewed footage. You would add the --write-movie parameter to your Godot call to create a series png files with a wav file of the play session. You'd have to assemble these pngs into a video using FFmpeg and pass the resulting to a video-to-text llm solution using a Codex skill.