From the Roman Legion to the Org Chart: What Dorsey Gets Right — and What He's Overstating
The Origin Story: Why Hierarchy Exists at All
Dorsey opens with a line that is more precise than it sounds: "Every org chart you have ever seen is a Roman military formation wearing a suit."
He is not being rhetorical. He is being literal.
Roman legions were organized around a hard constraint: a commander could directly supervise, communicate with, and make decisions about a fixed number of people. Research on human cognitive limits — later formalized by Robin Dunbar and studied extensively in military science — put that number between 3 and 8. The Roman solution was to build nested layers: a decurion commanded 10 soldiers, a centurion commanded 8–10 decurions, a tribune commanded cohorts, a legate commanded the legion. Each layer existed not because it added value, but because communication technology — the human voice, a rider on horseback — could not reach further.
This structure endured because the constraint endured. For two thousand years, the fundamental problem of large organizations was the same: how do you get the right information to the right person in time to make the right decision? Every organizational innovation in that period was an answer to this question within the same physical limits.
The first major translation of military hierarchy into commercial use was the American railroad in the 1840s and 1850s. The railroads were the first organizations in history that required real-time coordination of people and assets across hundreds of miles — schedules, signal management, accident response, freight routing. Daniel McCallum at the Erie Railroad drew what is widely recognized as the first modern org chart in 1854, explicitly borrowing the military hierarchy's chain of command. His innovation was to formalize reporting lines, define responsibility, and create a system where information could travel up and decisions could travel down — fast enough to prevent trains from colliding.
From the railroads, the structure entered corporate life. Alfred Sloan codified it at General Motors in the 1920s. McKinsey formalized it as a consulting product in the 1950s. Peter Drucker named the manager's job. By the mid-20th century, hierarchy was not just how organizations were built — it was how people understood what an organization was.
But the hierarchy was always solving a specific problem: the bandwidth limitation of human beings as information conduits. The org chart was not a theory of how people should relate to each other. It was an engineering solution to a communications constraint.
The First Experiment in Breaking It: The Manhattan Project
The first major attempt to build a large organization that deliberately suspended hierarchical boundaries for a defined purpose was the Manhattan Project, beginning in 1942.
The project required physicists, chemists, engineers, military officers, industrialists, and procurement specialists to work in real-time collaboration across dozens of sites with overlapping authority. J. Robert Oppenheimer at Los Alamos ran something closer to a research university than a military command structure — scientific staff could challenge decisions, teams formed around problems rather than reporting lines, and information flowed laterally rather than up-and-down. The military hierarchy existed at the perimeter (General Groves managed procurement, security, and political coordination) while the internal structure was deliberately flat and cross-functional.
The Manhattan Project succeeded in three years at a scale and speed that traditional hierarchy almost certainly could not have matched. It also only worked because the coordination problem was bounded and temporary — everyone understood the mission, the timeline was defined, and when it was over, the structure dissolved.
The corporate world has been trying to replicate that model ever since, with mixed results. Skunkworks teams. Cross-functional squads. Matrix organizations. Agile. Holacracy at Zappos. Valve's flat structure. Every experiment that achieved genuine cross-functional speed did so by creating a bounded exception to the surrounding hierarchy — not by eliminating hierarchy entirely. The attempts to eliminate it wholesale, including Valve's famous flat structure, consistently hit a ceiling: somewhere around 300 people, the coordination costs of a truly flat organization exceed the coordination costs of a well-run hierarchy.
This is the history that Dorsey is entering. And his argument is that AI changes the underlying constraint that made every prior attempt fail.
What Dorsey Gets Right
The core of Dorsey's argument is structurally sound.
Hierarchy was built for information flow. Managers historically exist to translate strategy into tasks, aggregate status upward, and maintain shared context across teams. AI does all three better, faster, and more accurately than a human relay chain. If the world model that Block is building — synthesizing Slack messages, code commits, product decisions, and transaction data into a real-time organizational context — actually works, then a significant fraction of what middle management does today becomes technically redundant.
His three-role structure is also more thoughtful than it first appears. Individual Contributors do deep specialist work. Directly Responsible Individuals own a defined problem for a fixed term — roughly 90 days — with authority to pull cross-functional resources. Player-Coaches develop people while doing the work themselves. What this eliminates is not management. It eliminates administrative management — the layer of people whose primary function is running status meetings, producing alignment decks, and managing up. That layer is genuinely the lowest-leverage activity in most large organizations, and AI can replace most of it.
Andy Grove identified this problem decades ago. A manager's output, Grove argued, is the output of their team. The highest-leverage activities a manager can perform are training people, giving feedback that compounds over time, detecting problems early, and making directional calls under uncertainty. Almost none of those activities require a status meeting. Almost all of the time most managers spend is not on those activities. AI removes the reason to spend time on the rest.
What Dorsey Is Overstating
The jump from "AI reduces coordination overhead" to "there is no need for a permanent middle management layer" is where the argument runs ahead of the evidence.
Management is not coordination. That is the central conflation in Dorsey's essay, and it matters.
The highest-value work of a manager — judgment under ambiguity, coaching individuals through difficult growth edges, motivating people in the face of uncertainty, making calls where the data is incomplete and the stakes are real — is not an information-routing problem. AI does not know what a specific person needs to hear at a specific moment in their development. It does not know when to push and when to protect. It does not carry the trust that makes difficult feedback land rather than wound. These are not gaps that will close as models improve. They are functions that depend on human relationship and human accountability in a way that is structural, not technical.
Block's own employees are already pushing back. According to reporting from the Guardian and covered in Bloomberg, roughly 95% of AI-generated code changes at Block still require human modification. Regulatory constraints in financial services — which is Block's core business — significantly limit how much AI can autonomously own decisions. And the FT's coverage of the announcement noted what the essay does not: Block's headcount tripled between 2019 and 2022, from 3,800 to over 12,000 employees, during the zero-interest-rate era. A significant portion of what AI is "replacing" is the organizational excess that accumulated during that period, not permanent management value.
Dorsey himself acknowledges it: "Block is in the early stages of this transition. It will be a difficult one, and parts of it will likely break before they work." That is the most honest sentence in the essay, and it deserves more weight than it gets.
What Is Actually Going to Happen
The realistic arc is not a binary between "hierarchy stays" and "hierarchy disappears." It is a phased shift in what management is for.
In the near term — the next two to three years — AI removes the administrative burden that consumes most managers' calendars: status updates, coordination overhead, reporting chains, meeting preparation. Managers gain time, better information, and faster decision cycles. The shift is from meeting-heavy operator to higher-leverage leader. This is already happening at the companies that have committed to it seriously.
Over the medium term — three to seven years — fewer pure middle-management layers survive. Not because the work disappears, but because the span of control expands. A manager augmented by AI can effectively oversee more people, more work, with better information than a manager without it. The role evolves toward what Grove called the highest-leverage activities: training, performance reviews, early problem detection, direction-setting. Player-coaches replace pure administrators. This is Dorsey's DRI model, and it is credible at this timescale.
In the longer term — seven to fifteen years — organizations genuinely rewire. Structures built around information-routing constraints become structures built around human judgment and relationship. AI handles coordination and workflow orchestration. Managers become system designers, culture carriers, and judgment authorities — the people accountable for outcomes that require trust, ethics, and wisdom that no model can provide. This is the version of Dorsey's future that is plausible. It is not zero layers. It is fewer, better layers doing different work.
The Key Reframe
The story Dorsey is telling is not "AI is replacing managers." The story — if you read it carefully — is "AI is removing the lowest-leverage parts of management so that managers can finally operate at their highest leverage."
Grove said the best managers allocate their time to the activities that multiply the output of everyone around them: training that compounds, feedback that lands, early signals caught before they become crises, decisions made in time to matter. Most managers today do not operate that way. They can't — the administrative gravity of a large organization pulls everyone toward the work that is urgent and visible, not the work that is high-leverage and slow-burning.
AI breaks that gravity. Not by eliminating management, but by eliminating the reason to spend time on everything else.
The managers who understand this — who use AI to clear the calendar, then use the calendar on the work that actually develops people and moves organizations — will operate at a level that was genuinely difficult to sustain before. That is the real transformation Dorsey is describing, even if he sometimes overstates it.
One-line takeaway: AI won't eliminate managers — it will finally allow great managers to operate at full leverage.
Sources: Jack Dorsey & Roelof Botha, "From Hierarchy to Intelligence" (Block, April 2026) | Sequoia Podcast: "Every Company Can Now Be a Mini-AGI" | FT coverage (April 2026) | Andy Grove, High Output Management (1983) | Daniel McCallum's Erie Railroad org chart (1854) | Paul David, "The Dynamo and the Computer" (1990) | Robin Dunbar on cognitive limits and span of control