We need prompts to behave more like shared systems, not throwaway instructions. Based on the research demo LatentLit from Thariq Shihipar at GoodfireAI I explored a remix called Framework Based Documents: a space where company teams and LLMs co-define rules, roles, and terms together. It’s a way to make collaboration with AI visible, repeatable, and trustworthy. Today most AI prompting is still just... text. It works until you’re collaborating at scale with a model, and a team. Then it breaks: • No structure • No shared language • No visibility There's an amazing space for provoking how to treat structured intent instead of raw prompting. More details of the original content: