automated game dev via ai is yet to have its watershed moment. there are just way too many moving parts for a simple agentic loop to tackle. a full fledged ai powered game dev engine will require tons of harness.
the comparison, for me at this point, comes against the current state of coding agents that are munching through feature after feature. landing prs in production. start-ups like rosebud are largely building simple three js scenes.
based on my own experience of creating games, here are the things such an engine will have take care of:
1- core story and mechanics, based on user input
2- character design
3- level design
4- animation
5- input controls
6- sound effects and wiring
7- progression
8- splash screen
9- scoring mechanism
in theory we can have a subagent handling each aspect mentioned above, and then another agent which does integration.
another agent plays the game, capturing image, video, and console logs and passing it to the orchestrator which decide what changes to make.
this runs in a loop for either a fixed number or till an agent judges that the game is finally built.
this is part 1 of series with no fixed schedule. a full fledged game dev engine that can build a playable game based on user input is one of dream. i may build it at some point.
Top comments (0)