Organizational AGI
The ones that do not reject it often fare worse: the tools go in, and coherence quietly rots. Either way the diagnosis is the same. The organization was not healthy enough to absorb what it bought.
Organizational AGI is that health. Not a smarter model — a company able to put artificial cognition to work across everything it does, coherently, at the tempo its environment demands, without spending its own survival to do it.
It is a long-horizon attractor, not a product you purchase on day one.
Leaders move toward it through strategy craft (QCEA / Strategy Engineer), then Metabolic Organizational Design, then System of Cognition and Judgment Infrastructure — toward long-horizon Organizational AGI capacity.
The practical discipline
Without it, AI adoption remains a series of local transplants that the organization either rejects or absorbs at the cost of coherence.
The laptop superworker is a successful transplant that still costs coherence. One person plus a local model can replace a department. Generation delivered a local gain. The organism did not receive a shared representation it can act on. Intelligence that consumes more than the organism earns is not intelligence — locally brilliant, globally fatal.
The unlock
The word does the most work and causes the most confusion. In the machine sense, general means one system that handles anything. That definition is why nobody can say when it has been achieved.
The borrowed meaning
Capability located in a single system, measured by what it can answer. Undefinable in practice, and when imported into an organization it quietly argues for centralization — which is the opposite of what an adaptive system needs.
The meaning we use
The ability to keep re-allocating sensing, inference and action as conditions change — carried across many nested loops held coherent, with the irreversible decisions reserved for the humans who own the intent. Measurable, and located in the organization rather than in a vendor’s model.
Why this question has an answer
The organizational version of the question is more tractable than the general one, and the reason is structural rather than commercial. Three things an organization has that general intelligence lacks:
You can say what is inside the system and what is environment. Without that cut, there is nothing to evaluate — every claim about the whole becomes unfalsifiable.
An organization can name what it intends to persist — this business unit, this capability, this obligation. That designation is what turns viability from a slogan into a testable condition.
Signal to inference to decision to action and back to signal. You can time it, find where it breaks, and compare it against how fast the environment moves.
General AGI has no boundary, no designated level, and no closable cycle — which is precisely why the debate about it never resolves. The organizational question is the one a theory can actually evaluate, and it is the one that decides whether your AI investment returns anything.
The landscape
We are not proposing an alternative to these. Two of them describe the destination accurately and the third supplies the physics. What none of them provides is the instrument that tells you whether a given organization can get there from where it actually stands.
Ismail, Malone and van Geest
Scaling output by leveraging capability you do not own — external assets, community, algorithms — rather than by adding headcount. The observation that legacy structures attack new ideas like an immune system is diagnostically exact.
What it leaves open. The prescription is to build at the edge, away from the immune response. That treats the rejection without treating what caused it, and nothing in the model tells you whether the organism can survive the scaling it prescribes.
Microsoft
The AI-first enterprise where agents orchestrate core operations and feedback loops run continuously. Correct about the destination, and about agents becoming participants rather than tools.
What it leaves open. It is a description of the destination and of the operating disciplines that get you there — the human-to-agent ratio, the employee who directs and answers for the agents they deploy, human-led and agent-operated. What it does not supply is a way to tell whether a given organization is currently able to run that way. It describes the target state, not the readiness to hold it.
Prof. Clayton Williams
A formal account of what makes an adaptive system viable at all: it must close the loop from information to action faster than its environment drifts, and the cost of its cognition and operations must fit inside the resources actually available to it. Reading "available" as what the organization earns is our operational narrowing, not his formulation.
What it leaves open. It is a theory of viability, not a product. Turning that theory of viability into Metabolic Organizational Design practice — and into something measurable from an organization’s own telemetry — is the work we take on.
Guardrails
Every one of these rules something out. They exist because the failure mode of a framework is flattening — turning a set of simultaneous constraints into a menu you pick from, which is how rigorous ideas become consulting decks.
Rules out
A single unified model that thinks for the whole company
The organism is many nested loops held coherent, not one mind. Centralizing to fix fragmentation destroys the variety that let the system perceive its environment in the first place. The cure for fragmentation is coherence without control.
Rules out
Certification, maturity levels, and anything you can put on a slide permanently
Organizational AGI is a predicate re-evaluated against current conditions, never a milestone reached. An organization that qualified last quarter may not qualify today, and the instrument has to be able to say so.
Rules out
Cognitive capability measured without reference to what it costs to run
The system has to fund its own cognition and survive the load it carries. Intelligence that consumes more than the organism earns is not intelligence — locally brilliant, globally fatal.
Rules out
Borrowed machine-learning notions of generalization
The relevant capacity is not answering questions it was not built for. It is editing its own architecture faster than the landscape moves, without breaching the constraints that keep it alive.
The instrument
The viability predicate is eight admissibility clauses that must hold simultaneously, over an eighteen-law conjunction, evaluated at the level the organization has designated it wants to keep alive.
A conjunction, not a menu
The laws are not a catalogue you select the relevant ones from. They hold simultaneously or the predicate does not clear. Treating them as a menu is exactly the flattening the guardrails above exist to prevent.
Measured, not asserted
Fourteen laws are bound to a real signal — among them the information-action cycle, inference through knowledge, goal-maintaining behaviour, survival and resolution. Four (Laws 9, 11, 13 and 18) need data sources that do not exist yet and each names the telemetry it awaits. None is assumed true because it would be convenient. Bound is not the same as resolved: a bound law still returns no reading on a day its data is absent.
You do not get there by adopting a topology or buying a platform. The path is strategy craft (QCEA / Strategy Engineer) → Metabolic Organizational Design → System of Cognition and Judgment Infrastructure → long-horizon Organizational AGI capacity.
System of Cognition is how the organism reallocates sensing, inference, and action across nested loops. Judgment Infrastructure is rights, verification, and liability around irreversible calls. The system of record is what keeps both organizational rather than personal: intent, consent, judgment class, entitlements, dissolution.
Judgment infrastructure is a system of record for shared cognition. Organizational AGI remains the attractor. It is not a SKU.
You begin by finding out whether your organization can currently metabolize what you already bought.