← GR Work Tech · Beta
Analysis — brain protocol

Emergent Memory

A company's brain isn't built top-down as a central system. It emerges from the bottom up, from the union of the brains of the people who work there. We tested it on a real deployment.

giobi.com · June 2026 · working document

The thesis

A brain belongs to a person. A company is something else.

A brain — in the protocol sense — is a set of files that give an assistant memory and continuity: who the user is, what they work on, what happened. It is by definition single-user: one voice, one diary, one point of view.

The instinct, when you think about a company, is to build a company brain: a single central mind everyone pours knowledge into. Our thesis is the opposite: the company layer isn't built, it emerges — from the union of individual brains, plus a thin layer of shared rules. There is no central consciousness to maintain. There is a constellation that, read as a whole, is the company.

The question isn't "how do I build the company's brain." It's "the company's brain already exists — distributed — and can I read it?"

The test

Not talk. On a real deployment.

We took a live installation at a recruitment firm: three people — the CEO/founder, the co-founder (operations), the COO — each with their own brain active for months. Hundreds of diary entries, thousands of catalogued entities. Read-only access, purely structural analysis: who contains what, not the personal substance.

The test question: reading the three brains together, does the company emerge as a coherent entity — without anyone ever having written a "company brain"?

What emerged

Three findings, no central system.

1 — Values live in a shared config file, not in a brain

All three brains share the same read-only domain file. Inside there's no consciousness: there's the company's identity and an operating posture — truth before output, never fabricate corporate facts, declare your sources, separate fact from assumption. The governance that elsewhere is sold as a proprietary differentiator is here 2.5 KB of markdown, written once and injected into every brain.

2 — The org chart is the sum of individual profiles

Nobody drew an org chart. Yet roles and hierarchy emerge cleanly from the union of profiles: CEO-founder, co-founder in operations, COO. The structure isn't a central document: it's the union of the user.md files.

3 — The map of the work is the graph of shared entities

By measuring which clients and projects appear across more than one brain, the shape of the company appears on its own. One account touched by all three. Others shared in pairs. The founder as the central node touching everything; the other two connected on their respective domains — delivery and execution.

account A  → CEO + co-founder + COO  (all three)
account B  → CEO + co-founder
client C  → CEO + COO
· the overlap graph is the map of the work

Patterns, not averages

This is where the bet is won.

A major account appears in two brains with two different natures: in the founder's brain it's a company record — the strategic account; in the COO's brain it's an operational project — the placement in progress. They are the same reality seen from two sides.

The temptation would be to average them — merge them into one version. That would be a mistake: it destroys information. The right move is to union them: the strategic view plus the operational view give the complete relationship with that client — a relationship no single brain holds in full. An LLM reading the constellation doesn't compute an average: it recognizes a pattern and composes it.

Averaging flattens. Union composes. Company intelligence lives in composing partial views, not in averaging them.

Where emergence ends and the tool begins

The same test draws the line, honestly.

The same data shows the limit. The same real-world entity lives as uncoordinated nodes: the same client with different spellings, different schemas, a record here and a project there. For emergent intelligence that's fine — an LLM reconciles it. But if you need deterministic canonical truththe record for that client, its identifier, the exact pipeline value — the constellation of brains is ambiguous by construction.

That is, and remains, a different function: a structured layer — a deterministic app, a CRM — that holds the single record. It's not a company brain: it's an operational tool. The two coexist and complete each other.

Bottom-up · brains

Values, org chart, map of the work, patterns, context. They emerge. It works.

Structured · Hub/CRM

The deterministic canonical record. The one thing worth centralizing.

Consequence

A company brain isn't sold as a thing to build.

You don't build the company's brain. You read it, you activate it, and you give it a record where it's needed.