カンパニーブレイン converges on Markdown bodies with YAML labels

The author's argument builds on earlier work of their own: an instruction to write procedures into SKILL.md, and mdschema, a tool that validates Markdown structure against a YAML schema. Both, the author notes, favored Markdown with YAML added alongside — the pattern this essay now spells out.

The concept has precedent. The author says the term company brain began appearing in English-speaking circles around 2026, and cites Andrej Karpathy's LLM Wiki and Garry Tan's GBrain, both of which use a Markdown repository as the knowledge store. Anthropic and OpenAI formats already lean this way: CLAUDE.md is a Markdown instruction file Claude Code reads per project, AGENTS.md is OpenAI's Markdown-style agent description placed next to a README, donated to the Linux Foundation's Agentic AI Foundation in December 2025, and SKILL.md was published as an open standard in December 2025.

Databases still have their place for large records, frequent updates, simultaneous writes and strict filtering, and vector databases are useful as an index built from Markdown rather than as the source of truth. The author's own caveats are practical: YAML lets anyone write anything, so a typo like cf0 for cfo goes unnoticed, cross-cutting queries such as listing every document past its review date need a script, and branching in Git for review remains unfamiliar to non-engineers.