Your Company Has Amnesia. Here's the Cure.

Build an AI second brain for your business from plain text files you own. The folder structure, the agent rulebook, and the weekly rhythm, explained simply.

Let me paint you a picture you already know.

A client emails asking if you can rerun the hoodie order from last spring. Same color, same decoration, same everything.

So you go digging.

You search your inbox. You find three threads and none of them have the art file. You Slack the person who ran it and they left in January. You open the CRM and the deal is there but the notes just say "sent proof." You find the PO in a shared drive folder called "2025 FINAL FINAL v3."

Forty five minutes later you have most of it. Not all of it.

That's not a you problem. That's an amnesia problem.

Your company knows the answer. It just has no idea where it put it.

The dumb simple fix nobody expects

Ready for the cure?

A folder of text files.

That's it. That's the whole thing.

Not a new CRM. Not a $40k platform. Not a six month implementation with a consultant in a quarter zip.

A folder. On your computer. Full of plain text files. And an AI agent pointed at it that reads them, writes to them, and keeps them tidy.

Andrej Karpathy, one of the most respected AI researchers on the planet, published this idea in April 2026 as a single markdown file he called an "idea file." Not code. Not a product. Just a pattern written in plain English, meant to be pasted into an AI agent so the agent could go build the thing for you.

He wasn't the only one who landed there either. Karpathy, Google, and Y Combinator president Garry Tan all arrived at markdown for agent memory in the same quarter. Three groups, three completely different problems, one answer: a folder of markdown files.

When three groups solving three totally different problems land on the same answer, pay attention.

So let's break it down like you're nine years old and I'm explaining it at the kitchen table.

First things first. What is a .md file?

Okay. You know a Word doc?

Forget it.

A .md file is a text file. Like Notepad. You can open it on any computer made in the last forty years and it works. No subscription. No login. No "your trial has expired."

The .md part just means "markdown," which is a fancy way of saying you can make text bold or make a bulleted list by typing a couple of symbols.

Want a headline? Put a # in front of it. Want a bullet? Put a dash in front of it.

That's the whole language. You could teach it to a fourth grader in ninety seconds.

Here's the part that matters: a .md file weighs almost nothing, it opens anywhere, and no company on earth owns it.

Your notes in Notion belong to Notion. Your notes in a .md file belong to you.

Hold onto that. We're coming back to it.

The structure. This is where it clicks.

A pile of text files is just a pile. What makes it a brain is the shape.

Here's the layout, and it's only five things.

1. The raw/ folder. The stuff that happened.

This is your evidence locker. Emails. Meeting notes. Supplier quotes. Call transcripts. Screenshots of a spec sheet.

Rule for this folder: nothing gets edited, ever. You drop it in and you leave it alone.

Think of it like the receipts drawer. Messy is fine. Messy is honest.

2. The wiki/ folder. The stuff you learned.

This is where the AI does its job.

For every supplier, client, product, and process you deal with, there's one page. One file. SanMar.md. GetDX.md. screen-print-vs-dtg.md.

Each page is a clean summary written from everything sitting in raw/.

So instead of nine email threads about SanMar, you have one page that says: here's their turn time, here's our rep's name, here's the three times they blew a deadline and what happened, here's their decoration minimums.

You don't write these. The agent does. You just read them.

3. index.md. The table of contents.

One file that lists every page in the wiki with a one line description.

This is the front door. It's how a person finds anything in four seconds, and it's how the agent knows what already exists so it doesn't create SanMar.md and San Mar.md and sanmar-notes.md.

4. log.md. The diary.

Every time something gets added or changed, one line goes in here with the date.

Why does this matter? Because six months from now when somebody asks "when did we switch decorators on the Eccles account," the answer is in the log. Not in somebody's memory.

5. AGENTS.md. The rulebook.

This is the sneaky good one.

AGENTS.md is a text file that tells your AI agent how to behave. What a page should look like. Where new stuff goes. How to name things. What to do when two sources disagree.

It's the config file that turns a generic AI into a disciplined wiki maintainer. It defines how pages get structured, where new stuff goes, and how answers come back.

Here's the wild part: you change how your AI works by editing a note. Not code. Not a settings panel behind a login. A note.

Your ops manual and your AI configuration become the same document. Write the process down once and both the humans and the robot follow it.

That's the moment most people's eyebrows go up.

Wiki links. How the brain actually connects.

Here's the piece that turns a filing cabinet into a brain.

Inside any page, you can write a link to another page by wrapping its name in double brackets. Like this: [[SanMar]].

Now the GetDX page links to the SanMar page. The SanMar page links to the screen printing page. The screen printing page links to the three clients who use it.

Ask about one thing and you can walk to everything connected to it.

It's Wikipedia. For your company. About your accounts.

And here's the kicker. A person would never keep up with this. You'd have to remember to link fifteen files every time you added a note and you would quit by Wednesday.

An agent doesn't get bored. It'll touch fifteen files in one pass and never complain about it. That's exactly the kind of tedious cross referencing work these things are absurdly good at.

Wait. Won't the AI just make stuff up?

This is the question I get every single time, and it's the right question.

Here's the answer.

When you ask ChatGPT "what's SanMar's turn time," it's guessing based on the entire internet. It might be right. It might confidently tell you six days when it's twelve. That's a hallucination and it'll cost you a client.

When you ask an agent pointed at your brain, it's not guessing. It's reading wiki/SanMar.md, which was written from your actual emails sitting in raw/.

The answer is sourced. You can click through and see exactly where it came from.

The AI isn't the source of truth. Your notes are. The AI is just a very fast, very patient librarian.

Which means the whole thing lives or dies on one habit: capture the details. Log the call. Drop the quote in the folder. Save the email.

Garbage in, garbage out. But good stuff in? Compounding advantage.

The RAG thing, explained without the jargon

You'll hear people say "RAG" a lot. It stands for retrieval augmented generation, which is a mouthful for a simple idea: before the AI answers, it goes and looks something up.

Most RAG setups work like this. You dump a hundred files into a tool. You ask a question. The tool grabs a few random paragraphs that look related, hands them to the AI, and the AI writes an answer. Then it forgets everything and starts over on the next question.

It's a research assistant with no memory. Every day is day one.

The wiki approach flips it. Instead of rediscovering everything from scratch on every single question, the wiki compiles what you know one time and keeps it current. The cross references are already sitting there. Contradictions have already been flagged.

You do the thinking once. You keep the result. Every question after that gets faster and better instead of starting over.

That's the difference between renting an answer and owning one.

Why coding agents do the heavy lifting

Now, could you build all this by hand?

Sure. And you'd get through about four files before you remembered you have actual clients.

This is where coding agents like Claude Code come in, and I want to be clear about why, because it's not what people assume.

You're not using a coding agent to write code. You're using it because it's the only kind of AI that can read and write a hundred files on your computer without you babysitting it.

Point it at your folder and tell it:

• Read everything in raw/ that came in this week
• Figure out what each thing is about
• File it under the right wiki page or make a new one
• Link it to everything related
• Update index.md
• Write a line in log.md

That's automatic triage. It sorts, summarizes, categorizes, and links while you're on a call.

Fifty emails in becomes twelve clean pages out. Every week. Without you.

That's the whole reason this works at company scale instead of just being a cute personal notes hack.

Now the part I actually care about most

Let's talk about ownership, because this is the argument I'll die on.

Every tool you use right now is renting you your own information.

Your deal history lives in your CRM. Your notes live in Notion. Your files live in a drive. Your conversations live in Slack. Every one of those has a monthly bill, an export button that half works, and a company that could change its pricing, its API, or its entire product next quarter.

You've felt this. Everyone in this industry has migrated platforms at least once and everyone has lost something in the move.

Text files don't have that problem.

• They open everywhere. Obsidian, VS Code, Notepad, your phone, a computer from 1998.
• They're searchable with the basic tools already on your machine.
• They sync with anything. Dropbox, iCloud, Git, a USB stick.
• They're future proof. Every AI tool that comes out over the next ten years is going to be able to read a folder of markdown. That's the safest bet in technology.
• Nobody can raise the price on them.

You want to switch from Notion to something else in two years? Cool. Your brain comes with you. Untouched. Zero migration.

You want to swap AI models? Point the new one at the same folder. Done in ten seconds.

The tools are temporary. The notes are permanent. Build on the permanent thing.

What this looks like in the promo world specifically

Alright, let's get out of theory. Here's what actually goes in the brain for a promo shop.

Supplier pages. Real turn times, not catalog turn times. Which rep actually answers. Decoration minimums. The time they shorted you 40 units and how it got resolved.

Client pages. Brand colors and Pantones. Who signs off on art. Their reorder rhythm. The fact that their marketing director hates the color orange and you found that out the hard way.

Product pages. What's worked for what industry. Real landed costs. What died on the shelf.

Process pages. Your margin formula. Your quoting steps. Your art submission checklist. How you handle a rush.

Decoration pages. Screen print vs DTG vs embroidery, with your actual notes on when each one makes sense and what it really costs.

Now a new rep starts. Instead of six months of shadowing, they read the index and ask the agent questions. They're dangerous in three weeks.

Now your top rep leaves. Their knowledge doesn't walk out the door with them, because it's been getting written down automatically the whole time.

That right there is the ROI. Not "AI efficiency." Institutional memory that survives turnover.

How you start. This week. For real.

You don't need permission or a budget for this.

1. Make a folder. Name it whatever. brain works.
2. Make the five pieces. raw/, wiki/, index.md, log.md, AGENTS.md.
3. Dump one thing in. Pick your five biggest accounts and drop in every email you can find.
4. Point an agent at it and tell it to build the wiki pages.
5. Read what it made. Fix what's wrong. Fixing it teaches the system.
6. Do it again next Friday.

That's a Saturday afternoon. Total cost, roughly zero.

Six months from now you'll have something no competitor can copy, because it's built out of your actual experience.

The bottom line

Your company already knows everything it needs to know. It's just scattered across inboxes, Slack threads, and the inside of a few people's heads.

A brain is just a folder that catches all of it and an agent that keeps it organized.

Plain text. Locally owned. Portable forever. Sourced from your real work instead of guessed from the internet.

It's not fancy. That's exactly why it works.

Want the full blueprint?

I put together The Company Brain Blueprint, a free walkthrough that gives you the complete folder structure, the starter AGENTS.md rulebook you can copy and paste, page templates, and the weekly rhythm that keeps it running.

No fluff. Just the build.

Logan Stout runs Stout Intel Agency, a boutique AI consulting and automation shop out of Salt Lake City. He builds systems for sales teams and agencies who are tired of losing information. Say hi at logan@stoutintel.com.