My (new) AI-First Knowledge Base

Why your company wiki dies, and the one change that keeps it alive.

I think I just solved something I’ve been losing to for years. And it’s the most boring-sounding win I can imagine: a knowledge base that doesn’t die.

Stay with me…

Every operator has built one: a wiki, a Notion system, a folder of SOPs. And every single one I’ve built has died. Useful for a month, stale by the quarter, abandoned by the year. I’d made peace with it and learned to manage around it.

This week I built one I actually believe will survive. It fixes every reason the old ones died.

And the whole difference comes down to one idea…

I’m not the librarian anymore.

Knowledge bases were never low-value. They’re one of the highest-leverage things you can have: how the business actually runs, what the levers are, the stuff that lives in one person’s head. The value was never the problem. The upkeep was.

Every time you learn something, you have to decide where it goes, whether it duplicates something you wrote six months ago, whether you now have to update it in three places. That tax never stops. So eventually you stop paying it.

It gets worse…

The moment your team finds something in there that’s stale and no longer accurate, they learn to not trust the knowledge base.

And they’re right back to tribal knowledge and DM-ing you for info.

The new plan: I don’t organize anything anymore. I give it information. Drop in a file, paste a link, or just tell it what I learned in a sentence. An AI librarian takes it from there: it decides where the fact belongs, checks it against everything already in there, and keeps it current. When I need something back, I don’t go searching. I ask, and it retrieves.

Under the hood it’s split in two.

First, the hard numbers and structured data live in a Postgres database: costs, vendor terms, bill-of-materials for products, etc.

Everything else…

The SOPs, tribal knowledge, background info on vendors and processes, story of how the business runs, etc. This piece lives in plain text the librarian writes, rewrites, and maintains over time. That’s plumbing.

The point is that I feed it and I ask it, and it does the organizing I used to abandon. And when it’s filing new info, it checks it against anything related to ensure things are cascaded to where they belong.

The result: a knowledge base FREED from the tyranny of tedious, meticulous maintenance.

And the best part…

When I need to know some detail about the business (like who’s this person in my inbox? Or where was that how-to process?), I just ask Claude CoWork to retrieve it.

I’m running the first one on one of my brands now. And I’ll be straight with you, because that’s the deal here: capturing what’s true this week is easy. The real test is whether it’s still true and still used a year from now. That’s the exact place every knowledge base I’ve built has failed. Ask me again in six months.

How many dead knowledge bases have you started? Reply and tell me. I’m collecting them.

— Deacon