I turned Markdown into a protocol for generative UI (fabian-kuebler.com)

Eric Schmidt's prediction about the future of user interfaces is realized in a new prototype that transforms Markdown into a protocol for generating dynamic user interfaces on the fly. The prototype utilizes an AI agent to create React UIs by interleaving text, executable code, and data within a single streaming context. This innovative approach features three key components: Markdown as a universal protocol that LLMs (Large Language Models) understand, streaming execution of code where LLM-generated instructions are processed immediately, and a `mount()` function that enables the seamless creation of reactive UIs that communicate effectively between client, server, and LLM. This work is significant for the AI/ML community as it simplifies the development of agentic UIs by leveraging existing knowledge of Markdown and common coding practices, enhancing the user experience with more responsive interactions. The innovative design also introduces various patterns for data flow, allowing for real-time updates and interactions, while modular 'slots' can be used to progressively inject complex UI components without delay. By aligning the system with LLM ergonomics and familiar frameworks like React, the prototype demonstrates a promising direction for future UI development, circumventing the need for new training methodologies and maximizing the efficacy of current AI capabilities.