An AI agent that knows your business is exponentially more useful than one that doesn't. The thing that gives it that knowledge is a second brain: your company's facts, decisions, voice, customers, and processes, written down in one place the agent reads before it does anything.
Without one, every agent you run starts each job as a stranger. With one, it starts like someone who's been on the team for a year.
What a second brain is, for a company
You may have heard "second brain" used for personal note-taking. Tiago Forte's Building a Second Brain made the idea popular for individuals: a system outside your head for the things you want to remember.
A company second brain is a different job. It isn't your notes. It's the written-down version of everything a good employee learns in their first year.
That means what you sell and what it costs. Who your customers are and what they usually ask. How you write. What you decided last spring and why. How a job gets quoted, scheduled, and closed out.
Most of that lives in your head today. Some of it is scattered across old emails, a price sheet, and a Google Doc nobody's opened since it was written.
A second brain puts it in one place, in plain words, where an AI agent can read it.
Why it makes AI agents so much better
An AI agent only knows what it's been given. Anthropic's documentation for its coding agent says it plainly: "Each Claude Code session begins with a fresh context window."
Some assistants now remember bits of past chats. That still isn't your price sheet, your service area, or the call you made last spring.
So an agent with nothing to read does one of two things. It asks you, again. Or it guesses.
The guess is the bigger problem. A guess sounds confident, and you only find out it was wrong after it went out.
A second brain replaces the guess with your answer. Here's what that looks like on three ordinary jobs.
Drafting a client email
Without it: the agent writes a polite, generic reply. It doesn't know this client is on a monthly plan, it signs off the wrong way, and it promises a fix "by Friday" because that sounds helpful.
With it: the agent reads that the client is on a monthly plan, what you did for them last, and your rule never to promise a date. The draft sounds like you and says only what's true.
Quoting a job
Without it: the agent asks for your rates, or makes up a reasonable-sounding number.
With it: the agent reads your price sheet, your minimums, and the note that weekend work costs extra. It drafts the quote and flags the one thing that doesn't fit your usual pattern.
Answering a website chat
Without it: a visitor asks if you service their area, and the assistant says yes because yes sounds friendly.
With it: the assistant reads your service area, your hours, and what you don't offer. It gives a straight answer and hands real leads to you.
The model didn't change in any of these. What it read first did.
How to build your company's second brain
This is the high-level version. It works with any AI tool, and you can start this week with what you already have.
1. Pick one home
Choose a single place: a folder of plain text files on your computer, a Notion workspace, or a Google Drive folder.
One home matters more than which home. An agent can't use a fact it can't find, and two places means two versions that disagree.
2. Start with six pages
You don't need to write everything. Start with these, and add more when an agent gets something wrong.
- The company. What you sell, what it costs, where you work, and your hours.
- Customers. Who buys from you, what they ask most, and what each current client is working on.
- Voice. How you write, words you never use, and two or three real emails you were proud of.
- Decisions. What you decided, when, and why. This stops an agent from reopening a question you already settled.
- How we do things. The steps for the jobs you repeat: quoting, onboarding a customer, handling a complaint.
- Right now. This quarter's goals and what's in progress, so the agent knows what matters this week.
Write the company page first. It's the one every agent needs, and you can finish it today.
3. Write facts, not descriptions of facts
"We're customer-focused" tells an agent nothing. "We call back every missed call the same day" tells it exactly what to do.
Real examples beat rules. Paste in the email you liked instead of describing your tone.
4. Connect your agents to it
How you connect depends on the tool. A few common ways:
- Claude Projects. A project has its own knowledge base and instructions. Anthropic's help page says you can "upload relevant documents, text, code, or other files to a project's knowledge base" and add instructions for how Claude should work.
- Context files for coding agents. Claude Code reads a file called CLAUDE.md at the start of every session, and its docs describe these files as "instructions you write to give Claude persistent context." Put the most important facts there, and point it to the rest of the folder.
- Connectors. The Model Context Protocol, or MCP, is "an open-source standard for connecting AI applications to external systems." With an MCP connector, an assistant can read a workspace like Notion directly instead of relying on copies you uploaded.
Whatever the tool, the test is the same: before the agent acts, has it read the pages that job needs?
5. Keep it current
When an agent gets something wrong, fix the page, not just the chat. A correction in a chat helps once. A correction in the second brain helps every agent from then on.
Replace old facts instead of adding new ones beside them. If two pages disagree, the agent has to guess again.
Put a date on decisions. "We stopped offering weekend installs" means more with "as of March" next to it.
Give it one owner. A shared document nobody owns goes stale fast.
6. Decide what stays out
Passwords and logins never go in. Keep them in a password manager.
Be careful with customer details. Put in what an agent needs to do the job, and leave sensitive personal information in the system it already lives in.
Two ways to get yours built
Our own agents read a second brain like this before they act: our facts, goals, voice rules, and past decisions. We'll build yours the same way.
Let Praetorium set it up for you
Praetorium is Athena's AI workforce. It's new, it's still being finished, and it's open now to founding members: Athena as your AI chief of staff, a director for each part of your business, and the specialists under them, 75+ agents in all. It's $1,000 a month, flat.
The first 13 founding members get onboarding free: a live session where we set up your workforce with you and connect your apps. Building your second brain will be part of that session, so your agents start out knowing your business.
Build it with Riley
If you'd rather build a second brain for the AI tools you already use, book a call with Riley, our founder. You'll work through what goes in it and how your tools read it, together.
FAQ
What is a company second brain?
It's your business written down in one place, in plain words, so AI agents can read it before they act. It covers what you sell, who your customers are, how you write, what you've decided, and how your jobs get done.
Do I need special software to build one?
No. A folder of plain text files, a Notion workspace, or a Google Drive folder all work. What matters is keeping it in one place and connecting your AI tools to it.
Why do AI agents need a second brain?
An agent only knows what it's been given. Some tools remember bits of past chats, but that isn't the same as knowing your prices, customers, and decisions. Without written context, an agent either asks you the same questions again or guesses. A second brain gives it your real answers before it starts.
How do I keep a second brain from going out of date?
Fix the page every time an agent gets something wrong, replace old facts instead of adding new ones beside them, date your decisions, and give one person ownership of it.
What shouldn't go in a second brain?
Passwords, logins, and sensitive customer information. Keep logins in a password manager and customer records in the system they already live in.