Key Points
- 1.Explains five levels of building an AI second brain out of markdown files and folders that both you and your agents can recall.
- 2.The value is getting knowledge out of your head into a system, so your data becomes your moat and agents find facts instead of hallucinating.
- 3.The five levels: exact-word lookup, topic grouping, semantic search, relationship-chain tracing, and full autonomy.
- 4.Design backwards from how you will recall the data; the simplest level that fits the need is the right one, not the most advanced.
- 5.It works with any agent harness (Claude Code, Codex, others) because it is just files and folders.
Summary
What a second brain is
A place to save notes, recordings, and threads, organized as markdown files and folders so both you and your agents can find them again. If neither can, the folder architecture or routing is wrong.
The five levels
Level 1 is exact-word lookup; level 2 pulls everything on a topic; level 3 is semantic search (meaning, not exact words); level 4 traces relationship chains across topics; level 5 is full autonomy. Each level answers a different kind of question.
Design backwards from recall
How the data will be accessed later determines how to store it now. Start with the end in mind, the way you would design a basketball to fit the hoop.
Pick the lowest level that fits
More advanced is not better. Use the simplest level that solves your actual pain and do not add complexity without a reason. The system is portable across agent harnesses because it is just files.
Worth watching for
Knowledge workers and builders organizing their notes and data for reliable AI recall.
- second-brain
- knowledge-management
- claude