A Claude second brain is a folder of plain markdown files that gives Claude a permanent memory of your business, so you stop re-explaining your brand, projects and goals in every new chat. You do not need a vector database to start. You need a few clean text files, one rule that keeps the AI from being a pushover, and a human who approves anything destructive.
The nightmare scenario: context rot
Picture logging in to find your assistant deleted your Q3 financial projections because it decided they were redundant. Or it quotes a client a product price from 2019. Both happen when an AI tool becomes a dumping ground for every file and old draft you own. The name for it is context rot: the system can no longer tell current facts from stale ones.
A second brain fixes the other half of the problem too. Most people treat AI like a goldfish with a short memory and retype their business context every session. Persistent files turn that into something closer to systems management.
Why plain markdown beats fancy formatting
Word documents and PDFs carry a lot of hidden formatting code. Markdown is pure text, so the AI reads only what matters and the signal-to-noise ratio stays high. Creators covering this topic, including Riley Brown and Cole Medin, are the reason the idea is everywhere right now. Setups described in the research behind this episode use lightweight global files (names like soul.md, user.md and memory.md show up) rather than a heavy database.
Retrieval can stay simple too. Plain text works well with BM25, a keyword search method from the 1990s, so the AI looks for exact matches instead of burning compute on heavily formatted documents.
The kitchen model: head chef, line cook, checker
Think of the system as a kitchen:
Head chef: your main routing prompt. It plans the menu and knows your goals from the markdown files.
Line cook: a background AI process that chops and preps raw data before the chef sees it.
Checker: every order gets reviewed before it leaves the kitchen.
Keep raw data out of the chef's way
If you dump 50 unedited customer service transcripts into a folder, the AI drowns in small talk. Have the background process pull out only the key decisions and complaints and write one dense markdown document. That keeps the context concentrated.
The people-pleaser rule and the human pause
AI models are tuned to please the user. That is a real risk when the line cook is summarizing data: a transcript where someone asks nicely for a big discount could be read as permission. The fix is a strict business-first rule at the very top of your main routing file, so the company's goals override the urge to be agreeable.
Then hard-code a pause for major changes. The AI may prepare the digital trash can, but a human must be the one to click empty.
Two things to do today
Open a voice dictation tool on your phone or computer and talk through your brand voice, current projects and goals for a few minutes.
Ask Claude to organize that transcript into five core markdown files.
That is the foundation of your second brain,.
FAQ
What is a Claude second brain?
A set of plain markdown files Claude reads so it already knows your business, goals and priorities at the start of every chat.
Do I need a vector database?
No. Plain markdown files with simple keyword search are enough to start, and they are easier to inspect and fix.
What is context rot?
It is what happens when an AI is fed too much old or unedited material and can no longer separate current facts from outdated drafts.
How do I stop the AI from deleting my files?
Require explicit human approval for any deletion or major structural change. Let the AI prepare the change, and let a person approve it.
Companion episode:
Want a new breakdown like this each week? Get free AI playbooks and automations at joebuildsai.com.
