Why an AI agent needs a brain
An AI agent needs a brain for the same reason a new hire needs time with you. Knowing your clients, prices and files isn't enough. It also has to know how you make calls, how you talk to each person, and when to stop and ask you. The Brain is where both live.
Here's how that played out with my own agents.
Facts were the easy part
For about a year I kept every decision and meeting in one notes folder. My agents could read it and write to it. In August I wrote that the context had become the main source of my agents' success. On October 15 I moved that folder into the WeaveKit Brain, more than 13,000 files.
So my agents knew what I had decided and why. What they didn't have was one page on how I decide. It was there in pieces, across meetings and emails, but nothing put it together.
A page on how I think
In November I had my agents read my side of about 250 meetings from this year, my founder notes and a couple hundred emails I sent.
What came back reads like a portrait, written for an agent that has to make a call for me. It's organized by the kind of call: what I take on and what I turn down, how I price and treat clients, how I think about money and bets, and how I talk to a new client versus a founder asking for advice.
It also keeps my contradictions. I'm a product person who has lived on services. I'm generous with clients and I care about margin. The page doesn't pick a side for me. It tells the agent those calls come to me.
The agent reads only what the task needs
The full page is too long to load on every task, so there's a short card. It says who I am in one paragraph, which section to open for which kind of call, and when to bring me in: money moves, a price gets set, a promise is made to a person, or it's about to say something in my name I've never said before.
My chief of staff is told to read the card before any call it makes in my name. For a reply to a client it opens one section, "How he talks, by audience." For a pricing question, a different one. For mechanical work like code and data, the card tells it to skip the page.
It keeps learning, with my OK
Every night my chief of staff starts a job that reads the day before: new meetings, email, my notes. If something adds to the portrait, sharpens it or contradicts it, the job writes a suggestion to an inbox page, with links to where it saw it. Nothing changes until I approve it, rewrite it in my words, or delete it.
If you're about to hire someone
Think about what you'd teach a new hire in their first weeks: which customers get some slack, and when to call you instead of just handling it. Mine wasn't written down anywhere until this.
On day one we load your email, Slack, contracts and meeting notes into the Brain, and after that the agent learns from each new email and meeting. A page on how you work can sit on top, built and approved the same way mine was.
If that's the hire you're weighing, bring it to a 20-minute call. We'll tell you what an agent could take on with what you already have, and what would still need a person.