Module 1 of 6 · About 30 min
Why a second brain that maintains itself
Understand the difference between taking notes and building knowledge, and what AI changes.
Most notebooks die the same way. You write a lot in the first weeks, then filing becomes a chore, notes pile up unlinked, and one day you no longer find anything. The problem is not writing. It is maintenance: summarising, linking, tidying, checking. That is exactly the work AI does well and tirelessly.
You bring the documents and ask the questions. The AI summarises, links, files and keeps the records.
Three layers
| Layer | What goes in | Who writes |
|---|---|---|
raw/ | Your originals: PDFs, articles, meeting notes, exports | You. Never modified. |
wiki/ | Notes: one per source, per topic, per person, plus links between them | The AI, following your rules |
| Questions | What you ask, and the answers filed in wiki/qa/ | You ask, the AI answers and files |
The key word is compounding. Each document added enriches the notes that already exist, and each answer filed is one less question to recompute. After three months, the brain knows things that no single document says, because it has connected them.
Where this method comes from
The skeleton (a raw/ folder, a wiki/ folder, a rules file and an index) is the one Andrej Karpathy made popular: a Markdown wiki maintained by an LLM, read in Obsidian. This course takes it further on two points that matter in daily use: what each session costs (module 5) and how to find a note among hundreds (module 6).
What you need
- Obsidian, free, to read and browse your notes.
- Claude Code, which needs a paid Claude plan (Pro or above). It reads and writes the files in your folder directly: no connector or plug-in to install.
- A terminal, for exactly two commands. Module 4 shows them.
Everything stays on your computer, in plain text files. If you stop using Claude tomorrow, your notes remain readable by anyone.
To do in your vault
Tick as you go: it is kept in this browser.