I guess there's a bit of a misunderstanding. The audience isn't an LLM. The purpose of StemJSON is to get a working native feature/module/app from just a prompt. An LLM can produce code, but that code can't be delivered directly to a published mobile app to extend it. Instead, the LLM produces output in a structured format describing the UI and behavior. The runtime in the app then deserializes it and builds native UI and logic from it, and that output can be reused over and over as standalone functionality.