Every company is about to need a brain.

You can't extract a company brain. You design one.

By Chris Huber Reitz

The "company brain" is the natural next step after the personal second brain, and the instinct is right: an organization that can't remember itself can't put AI to work. No argument there. The argument is with how almost everyone is building one. The standard blueprint guarantees failure. And the reasons it fails are nameable.

Look at how the brain is being built. Pull knowledge out of people's heads and old chat threads. Scrape the wikis and the tickets. Structure it into a store. Put retrieval on top. Call it a brain.

That isn't a brain. That's organizational memory — and memory isn't intelligence. Every build on that blueprint breaks in the same three places.

Where it breaks

• The 80% that matters won't extract. The knowledge that actually runs a company is tacit: the why, the judgment calls, the "ask Dave, he knows why we never touch that account." None of it is written down. A brain built from your documents knows the company that exists on paper. It's blind to the one that lives in people's heads.
• We ask retrieval for truth. It can only return what's there. Your real documents contradict each other. Point retrieval at that and it serves the contradictions back with total confidence, quietly turning confusion into official policy. Storage without governance manufactures confident wrong answers.
• Nobody owns keeping it true. Every dead wiki and three-year-stale doc teaches the same lesson: the maintenance burden is the product. The same complaint has recurred, almost word for word, in practitioner forums for over a decade. Tooling never fixed it, because it was never a tooling problem. It's a human-incentive problem.

Storage without governance
manufactures confident wrong answers.

Build past all three and you still end up with a brain calibrated to the company you used to be — the slowest-decaying knowledge, inside the fastest-moving orgs.

The reframe

That means three things the storage approach skips.

Surface the invisible operating system first.

Every org runs on two. The visible one — docs, org charts, SOPs. And the invisible one — the tacit shortcuts and unwritten rules where the real judgment lives. You can't train an AI on a layer you've never rendered. Before you can train your AI, you have to teach your organization how to remember itself. (the Substrate Model)

Relay context into action — don't just retrieve it.

The hard part isn't finding the right chunk. It's moving from situational context to coordinated action across human and machine actors — and preserving the human contribution in the handoff instead of dropping it. (the Context-to-Action Relay)

Calibrate trust instead of assuming it.

"Can it?" and "will it?" are different questions. Grant an agent exactly as much authority as its capability and its integrity have earned, and revise as that profile changes. A highly capable agent acting without integrity is the most dangerous thing in the building. The org trusts it precisely because it's capable. (Ability/Integrity Trust → Calibrated Authority)

The discipline that does all of this — that designs the architecture around the agent so the agent can be trusted to participate — is Organizational Intelligence Design.

It is not a diagnostic of your team's collective IQ, and it is not org-chart redesign. It's what organizational intelligence becomes when your people are working alongside an agent, not only each other: the layer above org design, not a redo of it.

Who's building this

I've spent the last several years on exactly this seam — across healthcare, higher education, and workforce systems, helping more than 15,000 people through the shift to working alongside AI. Organizational Intelligence Design is the name for the architecture my work kept pointing at: the infrastructure that has to exist before agents can be trusted with anything that matters.

"How" is the whole game.
It's the part nobody's naming yet.