I built an AI runtime that live-patches DOM, Canvas and Three.js without page reloads

I’ve been working on a browser based creation runtime that allows an AI to modify an application while it remains active.

Rather than regenerating the whole project, the system can inspect the current environment and apply targeted changes to DOM, Canvas, and Three.js experiences while attempting to preserve application state, camera position, timers, and render loops.

The demo on the link shows you what i've been working on and includes live visual changes, interaction updates, queued instructions, and runtime repair. I’ve also added visible verification telemetry to make the continuity easier to inspect.

I’d appreciate feedback on the engineering approach, especially around safe patching and long running session stability as this was only an idea i had a 3/4 days ago that i put together using existing implementations i had. The idea came from my game developer engine tool which already had playtest, livepatch and camera functions.

I then thought about what livepatch actually does. I could update landscape, objects and even the game play via text input to one of the model i had wired up. I wondered if i can create something from scratch using the same concept and use it as a in visual editor that never lost context, the same way livepatch worked. The rest of the functions got developed during constraints i had during testing, such as the canvas vision

What i built above is a completely different unplanned direction that i've taken a huge interest in and would love to hear your thoughts about the idea, how it could be used and something you may even want to try. I guess the question i'm asking myself is that i've figured out the main users could be game developers / web developers but are their other uses this could be used for that i'm not even thinking about that i should be.

Any feedback is highly appreciated