How it works
Your second brain
is a folder.
You've used AI for a year. It still forgets everything the moment you close the chat. Your notes live in four different tools. Slack threads. Granola transcripts. Voice memos. Nothing connects.
A second brain fixes that with one move: put everything in one folder, and let your AI read and write to it. That's the whole idea. The starter on this site is a working implementation of it.
01 — What you actually do
Three moves. Always the same three.
Drop something in. A meeting transcript, a voice memo, an article you saved. Drag it into the folder. Claude already sees it.
Tell Claude to ingest it. "Ingest this." Your second brain reads the file, pulls the people, the decisions, the things worth remembering, and writes them back as short linked pages. You approve the draft.
Ask anything. Not just the kind any AI can pull back — "what did our biggest customer say about pricing?" — but the kind only a second brain can answer: "what separates the deals we won from the deals we lost, across every call we've recorded?" Your second brain reads its own wiki — every transcript, every memo, every page you've added — and finds the pattern. With citations to where each insight came from.
Everything else — the patterns, the synthesis, the brain getting smarter — is what happens when you keep doing those three things.
02 — The structure
Three folders. That's the whole structure.
When you unzip the starter, you'll see three folders. Each does one job.
Stuff in
What you put into your brain. Drag and drop, no labels needed — your second brain figures out what's what.
raw/transcripts/
raw/articles/
raw/voice/
Processed knowledge
What your second brain writes back. One page per source, plus pages for the concepts, people, and patterns it sees across them.
wiki/sources/
wiki/concepts/
wiki/people/
wiki/syntheses/
Your business, in Claude's head
Filled in during setup. The starter prompt asks 5-10 questions about your work and writes the answers here. You don't edit it by hand.
_system/domains.md
_system/schema.md
_system/CLAUDE.md
03 — Where this came from
This isn't my idea
The pattern was first proposed publicly by Andrej Karpathy — a founding member of OpenAI and former director of AI at Tesla. He calls it an LLM-wiki: instead of having your AI rediscover knowledge from scratch on every question (the way most chatbots work), you let it incrementally build and maintain a wiki — a structured collection of linked notes that compounds with every source you add.
"The wiki is a persistent, compounding artifact. The cross-references are already there. The contradictions have already been flagged. The synthesis already reflects everything you've read. The wiki keeps getting richer with every source you add and every question you ask."
— Andrej Karpathy, llm-wiki.mdWhat's on paugrowths.com is a working implementation of that pattern. The folders, the helper skills, the starter prompt — they exist so a non-coder can run the whole thing without ever touching a terminal.
04 — Why this gets better over time
The flywheel
A brain with 5 sources in it isn't very useful. A brain with 50 starts answering questions you didn't know to ask. A brain with 500 is a moat your competitors don't have.
Every transcript, every article, every voice memo you put in compounds. The connections multiply. After a few weeks, your second brain knows your work the way you do — and can answer in plain English when you forget something.
05 — See it in action
Ten minutes to your first brain
Download the starter. Open it in Claude and Obsidian. Paste the starter prompt. You'll have a working second brain by the end of this coffee.
Get the brainFree · No email · ~30 MB