Markdown Is the Operating System. Everything Else Is a Render.

TL;DR

Markdown vs HTML cuts token usage 68–87%; Markdown vs DOCX XML is ~3x more token-efficient; RAG accuracy improves 35% when ingesting Markdown over raw HTML — the format is doing real work, not just aesthetics.

42% of companies have abandoned most AI initiatives not because the technology failed but because knowledge was stored in formats agents can't efficiently read, write, diff, or query.

Three authoring formats only: Markdown for prose, YAML for configuration, CSV for tabular data — everything else (PDF, DOCX, PPTX) is a rendered output that exists only at the moment of distribution.

Git is the universal write path; the read path varies (MCP for agent queries, RAG for NLQ retrieval, llms-full.txt for bulk ingestion, static site for humans) — all from the same plain-text source.