For nearly a century, the theory of the firm — Coase forward — has described one kind of organization. The Constructed Enterprise: a firm whose founding required building internal coordination from scratch, and whose growth was bounded by what the founders could build, hire, document, and supervise. The substrate has changed what is possible. There is now a second kind of firm.
Internal coordination is built piece by piece. Strategy, controls, processes, governance, and the organization's own memory of why it does what it does — all assembled by the people inside the firm, then maintained against entropy. Growth is bounded by what those people can build and hold.
The substrate provides the computational backbone — strategy schema, control patterns, decision lineage, governance grammar — deterministically from the organization's own coordinates. Founding becomes the selection of intent, not the construction of infrastructure. The firm carries its reasoning forward by architecture, not by discipline.
GrytLabs extends the canonical theory of the firm to accommodate a coordination mode the literature has never had to name. State that move precisely, and a chain of consequences becomes unavoidable.
Coordination capacity that used to be built and held inside a firm is relocated into an external, shared substrate — one that holds authority, constraint, and decision lineage at the protocol level. We call this compiled governance.
When the substrate supplies the coordination backbone, founding a firm becomes the selection of organizational coordinates rather than the construction of internal infrastructure. The defining property of the substrate-native firm is not size but founding cost.
The disruption is not a single attacker that scales up and absorbs incumbents. It is atomization — incumbents giving way to substrate-native operators, each holding individual relationships, with little left that is operational to acquire.
The conceptual-model aggregator — a firm whose product is persistent, licensable, attribution-preserving structured expertise — becomes economically tractable only because the substrate exists.
Substrate-mediated lineage lets an organization author a successor organization with inherited governance — completing the scaling story the canonical theory cannot describe.
Authorship of governed artifacts is not restricted to institutions. A domain expert, working alone, can author a substrate that institutions must engage with.
Incumbents fail to reach this configuration not from lack of capital or talent, but through epistemic foreclosure: upstream decisions quietly encode a foreclosed definition of the problem, so more capability is spent rendering the wrong model at higher fidelity. It is why incumbent AI deployments tend to degrade rather than compound.
What is the firm, when coordination cost can be located in the substrate rather than in the firm itself?
What the GEM thesis stands on
The first thesis asked what an organizational world model is, and built the substrate grammar. What follows is that substrate — shown generally, six ways, then the turn to what it becomes.
We publish this preview deliberately at this stage. Stating a thesis together with the conditions that would falsify it — before the evidence is in — is the discipline we hold others to. This is a position and the program built to test it, not a claim of settled fact.
The thesis will be warranted — or constrained — by a program of eight studies, each posing a question the position has to answer: the theory-of-the-firm foundations, the mechanism of incumbent failure, conceptual-model markets, substrate sovereignty, holonic world-model architectures, reproductive infrastructure, and who actually holds the client relationship.
The studies are specified and underway, not yet complete. What the inquiry returns may sharpen the statement, narrow it, or — at a falsification condition — overturn part of it.
"AI will make the expert obsolete" was forecast for radiology by 2021; in 2026 radiology is among the fastest-growing professions. The thesis holds that workflow capture does not replace expert judgment — and treats the radiology record as a live falsification test it must survive.
The position assumes a single governance grammar projects across very different domains. We are testing that directly. If the grammar does not transfer, the universality claim is constrained — and we will say so.