Your company knows things it can't reach.
Connect Claude to everything and something strange happens. Each person gets a genuinely smart assistant, and the company gets nothing. Your Claude knows your threads, your docs, your meetings. Your colleague's Claude knows theirs. Neither knows what the other decided last Tuesday.
The knowledge was always scattered; AI just made the scattering obvious. A decision made in a Slack thread never reaches the person who needed it. A customer objection heard on three calls never becomes something the team can point at. Context stays private by default, and private context does not compound.
A company brain is the fix. What follows is the most basic version of one, and that is the point. It is deliberately small enough to stand up this week.
The smallest thing that works.
Don't buy anything. Don't design a taxonomy. The first version of a company brain is a shared drive, or a Notion space, with a handful of folders in it. That's the whole architecture.
A single Drive or Notion space every person and every Claude can read.
Company-wide, plus one per function. Sales, product, engineering, support.
Decisions, patterns, and answers written once, so nobody has to reconstruct them.
What makes it a brain is not the storage. It's that things get written into it deliberately, in a form both people and models can read, instead of staying trapped in whichever tool they happened to be said in.
The bar for what belongs is simple. If it would still be useful to someone else in three weeks, it goes in. If it's only useful to you today, it stays where it is.
One task a day, running for you.
Here's the part that makes it actually happen. Set up a daily Claude Cowork task. Each morning it goes over yesterday's material across everything it's connected to, your Slack, your docs, your Notion pages, your meeting notes, and does one job: separate what is only your context from what the company needs.
Then it drafts. For each item worth keeping, it writes a short note in your words, proposes which folder it belongs in, and says why. Company-wide, or sales, or product, or nothing at all.
That is the entire mechanism. One task, once a day, per person. The work you were never going to do by hand now happens by default.
Nothing lands without a human.
The daily task proposes. It does not publish. Every morning you get a short list of drafts and you approve, edit, or drop each one. Five minutes, most days.
Keep that step. It's tempting to automate it away once the classification looks good, and it's the wrong thing to remove. It's what keeps sensitive material out, what stops half-formed decisions from being recorded as final, and what keeps the writing in a voice your colleagues trust. A brain nobody trusts gets ignored, and an ignored brain is just a folder.
A prompt to start with.
Give the daily task something like this. It's intentionally plain; you'll sharpen it once you see what it gets wrong.
Every weekday, review everything I touched yesterday across Slack, Drive, Notion, and my meeting notes. Your job is to decide what belongs in our shared company brain and what stays private to me.
For each item worth keeping, write a short note of five sentences or fewer: what was decided or learned, why it matters, and who it affects. Use plain language and my voice. Do not summarize the whole thread; capture the durable part.
Then classify each note into one folder: Company for anything that affects everyone, or the relevant function folder for anything narrower. Say in one line why you chose that folder.
Skip anything personal, sensitive, half-decided, or only relevant to me. When in doubt, leave it out and tell me you did.
Return the drafts as a single list for me to approve. Only after I approve, write each approved note into its folder in the shared drive.
Start with two people.
Do not roll this out to the company. Pick one or two people whose work touches the most surfaces, usually someone in sales and someone in product, and run it for two weeks.
You will learn three things fast. Which folders you actually need, which are always empty, and which one keeps collecting things that don't fit. What the task over-shares, since the first week it will propose far too much. And what a good note looks like in your company, which is almost always shorter than the first drafts.
Fix those, then add the third person. The structure that survives contact with two people is the one worth giving to twenty.
Add complexity later, and only when it's earned.
You will be tempted to build the good version first. Tagging schemes, nested folders, automatic routing rules, a review workflow, embeddings over the whole thing, a second task that summarizes the first. All of it sounds obviously right on a whiteboard, and all of it is guessing about problems you have not had yet.
The simple version has a real advantage: when it fails, it fails visibly and you learn something specific. Notes nobody reads mean the bar is wrong. A folder that fills with junk means the classification prompt is loose. People searching manually means it's time for retrieval. Each of those is a signal that buys you one upgrade.
When one keeps collecting things that don't belong in it.
When search inside a folder stops surfacing the right note.
When the daily review is consistently boring and you trust every draft.
So ship the folder, the daily task, and the approve step. Then add exactly one thing at a time, in response to something that actually broke. A basic brain people use beats an elegant one they don't, and the elegant one is much easier to build once you know what it's for.