Interesting - I have a thesis that some meta-summary layer is going to develop between natural language prompts and structured, hierarchical prompts. They may not be 1-to-1 deterministic, but would allow more fine-toothed management of LLM interactions. These 5 page specs are not sustainable or realistic to build on top of.