An idea, put up for scrutiny
A meta-model for operating a company — the structure its knowledge takes, so people and AI can both rely on it.
Robert Blust · Software Engineer & Architect
The problem
A company's knowledge lives everywhere — and nowhere.
Purpose, strategy, how work flows, who owns what, what gets measured, which rules hold.
It is written down. No two people would look in the same place.
Why it is not solved
Writing it down is easy. Keeping it true is expensive.
The wiki pagewritten once, true for a month
The decktrue on the day it was shown
The spreadsheettrue for whoever was in it last
So you ask a person instead. That is the tax.
The idea
One structured knowledge base — the brain of the company.
The facts as Markdown, structured by a meta-model, versioned like code.
Not new documentation but a projection of how the company runs. Built inside one company — and a company of one needed the same shape.
The shape
Every role, every rule, every metric — written down once.
- Plain words, not a databasenothing to install, nothing to log into
- Stored and versioned like codeevery change has an author, a date and a reason
- Its title is its nameyou refer to it the way you say it out loud
- It grows with the companya thing that gains parts of its own keeps them with it
What holds it together
Nothing points at nothing.
Pages refer to each other by the name of the thing — not by a link that points somewhere.
A rule with nobody accountableshows up the moment it is written
A role nobody holdsshows up the moment someone leaves
A metric with no targetshows up before anyone reports it
You cannot quietly lose anything.
The breadth
Everything that makes a company — in one vocabulary.
Roles and teamswho does what
Processes and gateshow work flows
Strategy and objectiveswhere we invest
Metricswhat we measure
Ruleswhat we enforce
Conceptswhat we define
Decisionswhat we decided, and what we rejected
Valueswhat we believe
Peoplewho we are
Not every company fills all of it in. The kind stays in the model anyway.
Concrete
Governance as data — not as a slide.
- Who signs it offa named role, not a habit
- What has to be truethe criteria sit where they apply
- Which rule sits behind itnothing gets built before the sign-off
Why it pays
For people, and for AI.
For people
Onboarding by role. An answer to “who owns this?”
And one place where the answer is the same for everyone.
For AI
The same page, with nothing in between.
It acts inside the rules you wrote.
It does not make an AI right. It shrinks what it has to guess.
The thesis
Markdown, checked by agents.
A schema language
A second format nobody reads, kept beside the one everybody does.
A schema in prose
A table says what a role has. An agent reads it and checks every page against it.
Not a stage on the way to a real schema. With the right meta-model, the prose is the schema.
What would make it usable
Three things before anyone can really use it.
A real companydescribed in this vocabulary, in the open — my own, one person
A tool like spec-kitset up an instance, add a page from its schema, check, upgrade, package as a skill
A validatorafter that — not before
Which of these exist, the roadmap says. Not this slide.
github.com/companygraph/meta-model#roadmap
What I am asking
Tell me where this is wrong.
- 1 · Where does your company not fit this shape?
- 2 · Would you keep this as Markdown checked by agents — or do you need a real schema language before you trust it?
- 3 · Would you set up your own instance with a tool like spec-kit — or is copying a folder of schemas enough?
- 4 · Would you pay for help building one — time and material, nothing hosted — or is that something you only do yourselves?
A no is worth more to me than a polite yes.
Apache 2.0 · companygraph.io