The Company Model

In March 2026, Jack Dorsey and Roelof Botha published an essay titled From Hierarchy to Intelligence. It is one of the most important pieces of corporate writing of the last decade, and most of the people who read it drew the wrong conclusion.

The surface argument is crisp. For two thousand years, from the Roman contubernium to the modern corporation, organizational design has been an attempt to solve a single problem: how to route information through a group of humans too large for any one person to hold in their head. Hierarchy is the answer everyone eventually converges on, not because it is elegant, but because until now there was no other mechanism capable of carrying that information. Dorsey's claim is that AI finally is. Block is rebuilding itself around a company world model, a customer world model, and an intelligence layer that composes atomic capabilities into solutions in real time. The org chart collapses from seven layers into three roles. Middle management disappears, because the thing middle management did, routing information up and down, is now done by the system.

This is a serious claim, and I think Dorsey is substantially right. But the most important word in the essay is not intelligence. It is model.

The real move

In an earlier piece I argued that the foundational risk facing enterprises today is not AI disruption or competition but the latency of their own organization, and that the underlying cause of that latency is the fact that the company cannot read itself. It has no coherent representation of its own state. Every decision has to be reassembled from scratch, out of human memory and scattered documents, every time.

What Block is actually announcing, underneath the intelligence language, is that they have started building that representation. They are calling it a world model. It is the same object, viewed from a more ambitious angle.

The move from hierarchy to intelligence is the headline. The move from opacity to legibility is the mechanism. You cannot have an intelligence layer without a substrate for it to operate on. The world model is that substrate. Everything downstream of Dorsey's argument, the collapse of middle management, the three roles, the edge-based people topology, the end of the predetermined product roadmap, depends on whether the company model actually exists and actually captures enough of the business to be acted upon.

This is the thing most readers are missing. The Block essay is not really a proposal about management structure. It is a declaration that Block has begun construction on a new kind of corporate asset, one that almost no other company currently has, and that this asset, rather than the AI layered on top of it, is where the compounding happens.

What a company model actually is

A company model is a continuously updated, machine-readable representation of what the company is, what it knows, and what it is doing.

That sounds abstract. It is not.

At minimum, it contains the company's current strategic commitments, expressed in a form that a system can reason against rather than a form that only a human skimming a deck can interpret. It contains the live state of the work in flight, who is doing what, what is blocked, what the dependencies are, what was decided last week and why. It contains the company's current beliefs about its market, its customers, its competitors, its pricing, and its positioning, expressed not as quarterly narratives but as structured claims that can be updated, challenged, and tested. It contains the operating history of the company, not as anecdotes but as a traversable record of decisions, outcomes, and lessons. And critically, it contains the company's capabilities, the actual atomic things the company can do, represented in a way that the system can compose into responses to new situations.

Most companies today have none of this. Or rather, they have all of it, scattered across Notion pages, Slack threads, Salesforce instances, financial systems, individual managers' heads, and PowerPoint decks that were last updated before the most recent reorg. The raw material exists. A coherent, queryable, continuously updated model of the company does not.

This is the real asset Block is building. The intelligence layer is what you attach to it. The three-role org structure is what becomes possible because of it. The compounding competitive advantage is what it produces.

Why you should care

The reason every operator needs to take this seriously is that the company model is shaping up to be the primary corporate asset of the next decade. Not the AI. Not the data. Not the tools. The model.

Consider what happens when a company has one and its competitors don't.

When a new regulation lands, the company with a model can compute its exposure in an afternoon, because its commitments, processes, and contracts are all represented in machine-readable form. The company without a model convenes a working group, assigns a partner at a law firm, and delivers a memo in six weeks. Both companies face the same rule. Only one of them responds inside the reaction window.

When a new market adjacency opens, the company with a model can recompose its existing capabilities into a new offering, because the capabilities themselves are represented as composable units rather than buried inside product teams. The company without a model starts a new initiative, staffs a team, builds a roadmap, and ships in eighteen months. By then the adjacency has closed.

When a talented person leaves, the company with a model loses a person. The company without a model loses years of context, because that person was the model. They carried the state of their area in their head, and when they walked out, it walked with them.

When the company itself needs to be restructured, reorganized, or sold, the company with a model can represent itself coherently to an acquirer, a board, or a regulator. The company without a model is what it always is under pressure: a set of slides that were accurate last quarter, a set of systems that disagree with each other, and a set of managers with conflicting accounts of what is actually happening.

The model is not a productivity feature. It is the corporate equivalent of a central nervous system. A company without one is effectively operating blind and coordinating through shouting, which is how every company has always worked, and which was tolerable only because everyone else was doing the same thing. That equilibrium is ending.

Why this is hard, and why that matters

If building a company model were easy, every enterprise would already have one. They don't, and the reasons are structural rather than technical.

The first reason is that encoding the state of a company into a machine-readable model surfaces every incoherence the company has been carrying. Conflicting OKRs, commitments no one remembers making, processes that contradict each other, ownership gaps, authority gaps. In a slide-deck company, these incoherences are absorbed by human judgment, papered over in meetings, or simply forgotten. In a modeled company, they show up as inconsistencies that the system cannot reconcile. Fixing them is political work, not technical work.

The second reason is that the people who hold the most context today are the people whose authority derives from holding that context. Encoding it into a shared model is, from their perspective, a transfer of power from themselves to the system. This is the real reason most knowledge management initiatives have failed for thirty years. The problem was never the tool. The problem is that the hierarchy runs on private knowledge, and asking it to build a public model of itself is asking it to dismantle its own basis for existing.

The third reason is that the model has to be built against a moving target. The company keeps operating while the model is being constructed, and every decision made during construction either gets captured by the model or escapes it. Companies that do this badly end up with a stale model that captures the company as it was eighteen months ago, which is worse than no model at all, because it creates the illusion of legibility without the substance.

The companies that succeed at this will do so because they treat the company model as a first-class engineering problem with executive sponsorship, a clear owner, and the authority to change how work is done in order to make it capturable. The companies that fail will treat it as an IT project or a knowledge management refresh, and they will produce the same scattered artifacts they already have, but with a more expensive dashboard on top.

What Dorsey is really signaling

Block published this essay now because they believe the window to build a company model is closing.

Not closing in the sense that it will become impossible. Closing in the sense that the companies that start now will accumulate a compounding advantage that late starters will not be able to close. Every day the model operates, it learns. Every transaction deepens the customer world model. Every decision enriches the company world model. Every failure of the intelligence layer to compose a solution becomes a signal about what capability to build next. The whole system is reflexive. It gets better because it runs, and it runs because it got better.

A competitor who starts two years later does not simply have two years of catching up to do. They have to catch up to a moving system that has been improving for two years, on a substrate they have not yet built, while their existing organization, which is still run on hierarchy and tribal memory, continues to consume all the oxygen in the company. This is why Dorsey is willing to publish the architecture in public. The architecture is not the moat. The compounding operation of the model is.

This is the part every CEO should sit with. If your competitor begins building a company model in the next twelve months and you do not, you are not behind on AI adoption. You are behind on the corporate asset that will determine whether your company is still a going concern at the end of this decade.

Where to start

A company model is not something you buy. It is something you build, in situ, against the real operations of the business. The companies that succeed at it will generally start the same way.

They will pick one slice of the business where the cost of opacity is highest, typically customer operations, revenue, or a specific product line, and they will build the model there first. They will encode the state of that slice, the capabilities that serve it, the decisions that shape it, and the signals that drive it, into a shared substrate that both humans and agents can read and write. They will run the slice against that substrate for long enough to learn what the model gets wrong and what it misses. And then they will extend it, one adjacency at a time, until the scope of the model matches the scope of the company.

This is unglamorous work. It does not produce impressive demos. It does not make for good all-hands slides. It is the corporate equivalent of laying fiber. Most of the companies that attempt it will underestimate how political, how technical, and how culturally destabilizing it is. A few of them will do it anyway, and those are the companies that will be operating at a fundamentally different clock speed five years from now.

The rest will still be running on hierarchy, reading Dorsey's essay, and wondering whether it applies to them.

It does.