Nothing much, if the brain is a folder you own. I pointed a second model at my own business brain a couple of weeks ago and it picked up where the first one left off. It sounded like me and it followed the same rules, because the thing it was reading hadn't changed. That's the whole argument for keeping your firm's AI knowledge in files you control rather than inside whichever product you happen to be paying for this year.
Where does your firm's AI knowledge live right now?
Most of the firms I work with already have somebody doing good work with AI, which is the step that matters and the one that opens everything else up. Somebody has worked out a solid way to get it to draft a standard client letter, and that way of doing it lives in one chat window.
The next person opens a blank chat and starts from scratch. So the firm has done the thinking more than once, and that thinking is worth keeping.
The question I'd put to a partners' meeting is a simple one. If you cancelled your AI subscription this afternoon, what would you still have? If the answer is a folder, you've already got the thing that matters. If it's a chat history, that's the piece worth moving into files.
What happened when I pointed a second model at the same folder?
I run my own business off a folder. Who I am, how I write, what we sell, how the work gets done, and what's been decided and why. Claude has been reading it for a good while now.
A couple of weeks ago I connected a different model to the same folder, mostly to see what would happen. It read the same files and carried on. It had the same voice and the same rules. I didn't have to re-explain anything to it.
That told me something I had half assumed and never actually tested. The models will keep changing. One is fast this month and slow the next. My brain doesn't have to change with them, which is the bit I keep coming back to. Whichever model suits the job in front of me gets pointed at the same folder, and the business it's working for is the same business.
Why does this matter more for a firm than for one person?
Because a firm has to hand things over. A firm has more than one person needing the same answer, so it has to live somewhere other than one person's head. People join, people leave, partners retire, and eventually somebody buys the place or inherits it. Every one of those moments is the same thing, and what gets passed on is either written down somewhere or it's gone.
I've written before about how no knowledge walks out the door once a firm captures what sits in its senior people's heads. The portability question is that same argument one layer further out. Your knowledge should outlast the person who holds it, and it should also outlast the software that reads it. A folder does both of those, which is most of the reason I keep mine in one.
What belongs in a brain you own?
Five layers, and I'd build them in this order. Instructions are how you want it to think. Guardrails are how it has to behave and what it must never touch. Knowledge is the firm itself: who you serve, what you sell, how you write, how the work gets done, and what you've decided and why. Connectors are the apps and data it's allowed to reach. And skills are the jobs written down once, so they run the same way every time rather than being explained again.
The knowledge layer is the bulk of it. The skills are the bit of AI that compounds, because each one is a job written down once and your competitor can buy the same subscription tomorrow afternoon, at the same price, and the way your firm does the work stays yours.
Mine sits in a synced folder on my own drive, so the AI reads it as files. The brain and the hands are two different problems, and the brain is the half you can start on this week without asking anyone's permission.
What does a firm get back by moving this into files?
Two things, and neither of them shows up on an invoice.
You keep the thinking, so the second request is better than the first and the hundredth is better again. And you keep the freedom to change your mind about a vendor, because the real switching cost is every hour of context somebody would otherwise rebuild from scratch, and that cost grows the longer an arrangement has been working well.
There's a third one that only shows up later. The firm becomes easier to hand over, because what a buyer or a successor is really acquiring is how the work gets done.
Generic output is the symptom you notice first. AI gives you generic answers when it knows nothing about you, and context kept in a folder is what moves it off generic.
What would I do about it this month?
I'd start with one folder and five files. Pick the five things your AI would need to know to be useful on a real job. Who you are, who you serve, what you sell, how you write, and how you want it to behave. Put them in a synced folder wherever the firm's files already live, so your IT people have nothing new to approve. Point your AI at it and run one real piece of work through it.
You'll know inside a week whether it's working, because the output starts sounding like you.
And then there's the test I'd run every so often. Point a different model at the same folder and see whether the work holds up. If it does, the brain is yours and the model is just the thing reading it. That's the whole test.
Want the map of what a firm's AI needs to know before it can be useful? The Firm's AI Brain Template is a one-page layout of the five layers, including who in the firm can see what. It's a free download.
If you're putting AI on the program for a partners' day or a conference, the keynote and workshop formats are here, and the session covers where a firm's brain sits before it covers a single tool. I've seen this land differently in a multi-office group, where one office is well ahead of another.
Frequently asked questions
What is an AI business brain?
It's a set of files that tell an AI who you are, who you serve, what you sell, how you write, how the work gets done, and what it must never do. Put those files somewhere the AI can read them and a general tool starts doing work that looks like your firm did it. The files are the brain, and the AI is whatever is reading them today.
Will my AI brain still work if I switch from one AI tool to another?
Yes, as long as the brain is kept as files in a folder you control. I pointed a second model at my own folder a couple of weeks ago and it read the same files, kept the same voice and followed the same rules with no re-explaining. A brain kept inside one product's chat history or custom settings stays there, so the folder is the part that travels.
Where should a firm keep its AI brain?
In a synced folder on the drive the firm already uses, so it sits under the firm's own storage, backups and permissions. The folder structure is what carries access, so who can see what is set by where a file sits, and the skills live in the team tier that everybody can read. Keeping it in ordinary files means any AI you connect later can read it.
Does setting this up need IT approval?
The folder half doesn't, because the AI is reading files that already sit in the firm's storage. Connecting the AI to email and calendar is the part an administrator has to approve, and in a lot of firms that administrator is the outside IT provider. So the brain, the skills and the written-down processes can all start while that request sits in a queue.
What should go in the brain first?
Five files. Who you are, who you serve, what you sell, how you write, and how you want the AI to behave. Those five carry most of the value, and the rest gets added as you go. Running one real job through them in the first week is what tells you whether they're any good.
How is this different from writing better prompts?
A prompt is one instruction for one task, and it leaves with the person who typed it. A brain is the context sitting underneath every prompt the firm ever writes, which is why the second request is better than the first. The brain is what the firm keeps.
