The Core Claim: Company as Agent
Block CEO Jack Dorsey has advanced a radical proposition: every company can be compressed into an agent. This is not the now-familiar pattern of giving each employee an AI copilot so the existing structure runs a bit smoother. It is using AI to take over the coordination function that hierarchy itself performs — the management layers, the information routing, the context-passing that consumes so many careers. Block is building the company as an intelligence agent, a kind of small AGI with a business attached.
Two Thousand Years of Organizational Evolution
To understand how radical that is, trace where the current form came from. The Roman legion invented the nested hierarchy. Its structure — the contubernium of eight soldiers sharing a tent under a decanus, the century of roughly 80 under a centurion, six centuries forming a cohort, ten cohorts a legion of about 5,000 — was, at its essence, an information-routing protocol. Its basis was a human limitation discovered through centuries of war: one leader can effectively manage roughly three to eight subordinates. The 8 → 80 → 480 → 5,000 cascade exists to multiply around that constraint.
The Prussian General Staff invented middle management. After Napoleon shattered the Prussian army at Jena in 1806, reformers rebuilt around a hard truth: never depend on individual genius at the top; depend on the system. They created a professional officer corps whose job was not fighting but planning operations, processing information, and coordinating units — precisely what corporate middle management does today: routing information, pre-computing decisions, keeping large organizations coherent.
The American railroads brought hierarchy into business. In the 1840s-50s, West Point-trained engineers moved to private rail companies carrying military organizational thinking. By the mid-1850s, Daniel McCallum of the New York and Erie Railroad — managing 500-plus miles of track and thousands of workers — drew the world’s first organization chart, formalizing the Roman logic of layered authority, explicit reporting lines, and structured information flow. It became the blueprint of the modern corporation. Frederick Taylor then optimized inside the hierarchy — decomposing work into specialized tasks, assigning them to trained experts, managing by measurement rather than intuition — producing the functional pyramid.
The Manhattan Project made the first great cross-functional attempt. Robert Oppenheimer organized Los Alamos into functional divisions but insisted on open collaboration between them — and when the implosion problem became critical in 1944, he reorganized the laboratory around it, creating cross-functional teams American business had never seen. It worked, but as a wartime exception. Postwar corporations asked whether such collaboration could become routine, and McKinsey’s 1959 matrix organization was the answer — combining functional specialization with business-unit accountability. The matrix powered the postwar economy while breeding complexity, rigidity, and bureaucracy, along with an entire consulting industry of frameworks to manage the damage.
Tech companies experimented boldly and mostly failed. Spotify popularized cross-functional squads and short iteration cycles; Zappos tried Holacracy and abolished management titles entirely; Valve went flat with no formal hierarchy. Each revealed the limits of traditional hierarchy, and none solved the underlying problem: Spotify regressed to conventional management as it scaled, Zappos bled talent, and Valve’s model struggles beyond a few hundred people. When organizations grow into the thousands, they return to hierarchical coordination — because no sufficiently powerful alternative information-routing mechanism existed. That is the two-millennium dilemma in one sentence: shrinking spans of control adds layers, and more layers slow information down. Every organizational innovation from Rome onward has been a workaround that left the tradeoff intact.
Block’s Move: Replace the Hierarchy’s Function, Not Its People One by One
What distinguishes Block’s experiment is the target. Haier’s micro-enterprises, platform organizations, data-driven management — all real attempts at the same problem — lacked a technology that could actually perform hierarchical coordination. AI is that technology. For the first time, a system can continuously maintain a model of the entire business and coordinate work the way information used to travel through layers of human management. Block’s architecture rests on four foundations.
Foundation one: capabilities. Atomic financial building blocks — payments, lending, card issuing, banking services, buy-now-pay-later, payroll processing. These are not products; they are hard-to-acquire, hard-to-maintain components (some carrying network effects and regulatory licenses). They have no user interfaces of their own — only targets for reliability, compliance, and performance.
Foundation two: the company world model. Block is remote-first, and everything it does produces machine-readable artifacts: decisions, discussions, code, designs, plans, issues, progress. In a traditional company, the manager’s job is knowing what the team is doing and passing context up and down the chain. In a remote-first company whose work is already legible to machines, AI can continuously build and maintain that picture — what’s being built, where things are stuck, where resources flow, what’s working and what isn’t. The information hierarchies used to carry now lives in the company’s world model.
Foundation three: the customer world model. Block sees both sides of millions of daily transactions — buyers through Cash App, sellers through Square — plus operating data from merchant businesses. That grants a rare capability: per-customer, per-merchant understanding of financial reality based on honest signals, compounding over time. As the argument goes: people lie on surveys, ignore ads, abandon carts. But when they spend, save, send, borrow, or repay — that is the truth. Every transaction is a real record of someone’s life. Richer signals make a better model; a better model drives more transactions; more transactions enrich the signals.
Foundation four: the intelligence layer. The layer that decides on top of the world models. Example one: a Square merchant’s inventory-turn data shows a product line stalling; the intelligence layer composes a new Cash Flow loan offer, promotion tools targeted at that category, and inventory-optimization advice. Example two: a Cash App user’s spending pattern suggests a move to a new city; the layer composes new direct-deposit settings, Cash App Card offers for the new neighborhood, and savings goals recalibrated to the updated income.
The Question Every Company Will Face
Block is early in this transformation and says so — it will be hard, and some parts will fail before they work. The reason for discussing it publicly is the belief that every company will eventually confront the same question: is your company distinctive enough to be worth existing as an agent? If the honest answer is no, then AI remains a cost-optimization story — layoffs, short-term margin, eventual absorption by a smarter competitor. If the answer is deep, then AI will not merely enhance your company — it will reveal what your company actually is.
Block’s own answer is the economic graph: millions of merchants and consumers, both sides of every transaction, financial behavior observed in real time — an understanding that compounds every second the system runs. Sequoia’s observers like to say velocity is the best predictor of startup success; most companies treat AI as a productivity tool, while a handful are asking how it transforms collaboration itself. Block is demonstrating a fundamental redesign of organizational structure — using AI to convert speed into compounding advantage.
The Bottom Line
From the Roman legion’s nested ranks, through the Prussian General Staff and McCallum’s chart, to the matrix and Block’s AI-native architecture, organizational evolution has always circled one question: how to balance scale against speed. AI is the first technology that might actually break the span-of-control dilemma rather than route around it. Whether Block’s experiment succeeds remains to be seen. But the question it poses — is your company distinctive enough to deserve existence as an agent? — deserves an answer from every boardroom. In the AI era, the essence of the organization may be redefined: not hierarchy, not matrix — agent.
