Well, they want to be able to parse and render on the fly. Their chat example is a good example of this, linked from the navbar where the LLM is instructed to use their DSL and they can render UI from it which is an interesting application.