Will middle management be rendered obsolete by AI? Jack Dorsey: Future organizations will only need three types of people.

Author: Jack Dorsey, co-founder of Twitter, founder and CEO of Square

Compiled by: Peggy, BlockBeats

Editor's Note: While most companies still view AI as an "efficiency tool," Jack Dorsey takes the question a step further: Is AI rewriting the very logic of how organizations operate? As the co-founder and former CEO of X and the founder of Block, he has long focused on the relationship between technology and organizational structure.

This article takes a historical approach, re-examining why enterprises have evolved into their current form and why this structure has begun to loosen. From the Roman legions to the modern corporation, the evolution of organizations over the past two millennia has consistently revolved around the same constraint: achieving information transmission and collaboration within a limited "span of control." Hierarchical structures, middle management, and matrix systems are essentially different solutions to this problem.

The emergence of AI has challenged this premise for the first time. When information can be modeled, understood, and distributed in real time, do organizations still need coordination mechanisms centered on "people"?

Similar changes are already emerging in reality. Recently, *People* magazine reported that the AI company "Dark Side of the Moon," with a team of over 300 people, operates without departments, job titles, or OKRs/KPIs. Collaboration relies on direct communication rather than hierarchical reporting; each of the five co-founders directly reports to 40-50 employees. Simultaneously, agents are embedded in daily workflows, capable of quickly completing information processing, product design, and even code generation. This structure is not simply about "de-managing," but rather about proactively addressing complexity in recruitment, employee turnover, and tool systems.

Taking Block's practices as a starting point, this article further proposes a more radical idea: moving from a "hierarchical organization" to an "intelligent company," replacing the traditional information routing system with a "company world model + customer world model + intelligent layer," and even reconstructing middle management itself. This is not just a matter of efficiency, but more likely a complete rewriting of organizational structure.

The following is the original text:

Sequoia Capital believes that "speed" is the best indicator of a startup's success . Most companies still view AI as a tool to improve productivity, with only a few beginning to focus on how AI can change the way people collaborate. Block is demonstrating a completely new path: fundamentally redesigning organizational structures and using AI as a compounding competitive advantage that continuously amplifies "speed."

The starting point of hierarchical organization: from the Roman legions to the modern corporation

Two thousand years before the emergence of organizational charts, the Roman army had already solved a problem that still plagues large organizations today: how to coordinate thousands of people in situations where communication is limited and distances are vast.

Their solution was to establish a nested command structure and maintain a relatively stable "span of control" at each level.

The smallest unit was the "tent group" (contubernium), consisting of 8 soldiers who shared a tent, equipment, and a mule, and were led by a decanus.
Ten tent groups formed a century (actually about 80 people), commanded by a centurion;
Six teams of 100 men each constitute a cohort;
Ten cohorts together make up a legion of about 5,000 men.

At each level, there is a clearly defined commander responsible for summarizing information upwards and relaying instructions downwards. This structure, progressing from 8 to 80 to 480 to 5000, is essentially an efficient information transmission mechanism built on a simple yet crucial premise: one person can typically only effectively and directly manage 3 to 8 people.

The Romans gradually discovered this pattern through long-term warfare. Even today, the hierarchical system of the U.S. military largely follows a similar logic. We call this constraint the "span of control," and it remains a fundamental limitation that all large organizations cannot circumvent.

The next major change came from Prussia.

Following the disastrous defeat to Napoleon at the Battle of Jena in 1806, Scharnhorst and Gneisenau spearheaded military reforms, presenting a somewhat uncomfortable reality: reliance on systems was no longer sufficient to depend on individual genius. They established the General Staff, training a class of specialized officers whose role was not combat, but rather planning operations, processing information, and coordinating across units. Scharnhorst's initial intention was to "compensate for incompetent generals and provide them with the necessary capabilities." This was essentially the prototype of "middle management": a group of professionals responsible for information dissemination, pre-calculated decision-making, and maintaining the coordination of complex organizations. Simultaneously, the military clearly distinguished between "line" and "staff" functions: the former driving core missions, and the latter providing specialized support. This division is still widely adopted in businesses today.

In the 1840s and 1850s, American railroad companies introduced the military hierarchy into the business world.

The U.S. Army supplied railroad companies with a large number of West Point-trained engineers, bringing with them military organizational thinking. Line-and-staff structures, divisional divisions, and bureaucratic reporting and control systems all originated in the military. In the mid-1850s, Daniel McCallum of the Erie Railroad in New York drew the world's first organizational chart to manage a railroad system spanning 500 miles and thousands of employees. The informal management methods previously used for smaller railroads had become ineffective, leading to frequent train collisions. McCallum institutionalized Romanesque hierarchical logic: clearly defined levels of authority, clear reporting relationships, and structured information flow. This became the prototype of the modern company.

Subsequently, Frederick Taylor (known as the "father of scientific management") optimized the internal structure of this system. He broke down work into specialized tasks, assigned them to trained experts, and managed them using quantitative indicators rather than intuition, thus forming the "functional pyramid" structure—an organizational form that maximizes efficiency within an existing information routing system.

The first major stress test of this functional structure occurred during the Manhattan Project in World War II. This project required interdisciplinary collaboration among physicists, chemists, engineers, metallurgists, and military personnel to achieve a single objective under extreme secrecy and time pressure. Robert Oppenheimer employed functional divisions at Los Alamos National Laboratory but insisted on open cross-departmental collaboration, resisting the military's tendency towards "isolation." In 1944, when the "implosion problem" became a critical bottleneck, he reorganized the team, establishing cross-functional groups—a practice virtually unprecedented in the corporate world at the time. This model worked, but it was a wartime exception, driven by a few exceptional individuals. The question facing the postwar business world was: could this cross-functional collaboration be normalized?

Post-war corporate expansion and globalization have made the limitations of functional structures increasingly apparent.

In 1959, Gilbert Clee and Alfred di Scipio of McKinsey published "Creating a World Enterprise" in the Harvard Business Review, proposing the "matrix organization" framework, which combined functional specialization with a divisional structure. Driven by Marvin Bower, McKinsey helped companies like Shell and General Electric implement this model, achieving a balance between "centralized standards" and "local flexibility." This system became the paradigm of the "modern enterprise" in the post-war global economy.

Subsequently, new management frameworks emerged to address the complexity and bureaucratization issues of matrix structures.

McKinsey proposed the "7-S model" in the 1970s, which distinguishes between "hard elements" (strategy, structure, systems) and "soft elements" (shared values, skills, people, style), emphasizing that structure alone cannot guarantee organizational effectiveness, and that synergy at the cultural and human level is also necessary.

In recent decades, technology companies have experimented more radically with organizational structures.

Spotify introduced cross-functional teams and short iteration cycles; Zappos experimented with Holacracy, eliminating management titles; Valve adopted a flat structure with no formal hierarchy. These attempts all revealed the limitations of traditional hierarchical structures, but none completely solved the problem: Spotify reverted to traditional management after scaling up, Zappos experienced significant staff turnover, and the Valve model struggled to scale to hundreds of people. When organizations reach thousands of employees, they still have to revert to hierarchical coordination because there is no more effective information routing mechanism.

This constraint is exactly the same problem faced by the Romans and the Marines in World War II: reducing the span of control means increasing layers, and increased layers lead to slower information flow. For two millennia, organizational innovation has been trying to circumvent this trade-off, but has never truly broken it.

So, what's different now?

At Block, we began to question a fundamental assumption: that organizations must rely on humans as coordinating mechanisms and employ hierarchical structures . Our goal is to replace the functions of hierarchy with systems. Currently, most companies simply equip employees with AI co-pilots to make the existing structure function a little better, but the essence remains unchanged.

We are building a different model: a company that is itself an "intelligent agent" (or even a small AGI).

We are not the first organization to attempt to transcend hierarchical structures. Haier's "Rendanheyi" (employee-customer integration), platform-based organization, and "data-driven management" are all similar explorations. However, they lack a crucial element: technology capable of truly fulfilling coordination functions. AI is precisely this technology. For the first time, a system has emerged that can continuously maintain a model of the overall operation of the enterprise and coordinate based on it, without requiring humans to pass information through hierarchical levels.

To achieve this, a company needs two things: a "world model" of its own operations and a sufficiently rich set of customer signals.

Block primarily operates remotely, and all work leaves a recordable "trace": decisions, discussions, code, designs, plans, problems, and progress. These constitute the raw materials of the company's world model.

In traditional companies , managers are responsible for understanding the team's status and communicating information up and down the hierarchy.
In a machine-readable organization , AI can continuously build this global view: what is being done, where is it stuck, how resources are allocated, what is effective, and what is ineffective. This information, which used to be carried by hierarchical structures, is now carried by models.

However, the system's capabilities depend on the quality of the input signals, and "money" is the most authentic signal. People may lie in questionnaires, ignore ads, or abandon their shopping carts, but when they spend, save, transfer money, borrow, or repay, these behaviors are real. Block sees both ends of the transaction daily: the buyer through Cash App, the seller through Square, and has access to merchant operational data. This allows it to build a rare customer-world model—a customer- and merchant-specific understanding of financial behavior based on real transaction signals, and these signals continuously accumulate and strengthen.

The company-world model and the customer-world model together form the foundation of a new type of company. In this model, the company no longer operates around a predetermined roadmap with product teams, but is instead built around four core elements :

First, capabilities : basic financial capabilities such as payment, lending, card issuance, banking, buy-now-pay-later, and salary disbursement. These are not products, but underlying modules. They have no user interface, but they meet the requirements of reliability, compliance, and performance.
Second, the world model : including the company model (understanding its own operation) and the customer model (customer and market representations built based on transaction data), and gradually evolving into a system with causal and predictive capabilities.
Third, the intelligence layer : This layer proactively provides solutions by combining capabilities for specific customers at specific times. For example, when the system predicts a restaurant is about to experience cash flow difficulties, it automatically combines loan and repayment plans and pushes them to the customer in advance; or when changes in user behavior suggest a move, it automatically configures a new combination of financial services. None of this requires pre-design by product managers.
Fourth, interfaces such as Square, Cash App, Afterpay, and TIDAL are merely delivery interfaces; the real value lies in the model and intelligence layer.

When a system attempts to combine solutions but discovers a lack of a certain capability, this "failure signal" becomes the roadmap for future products. The traditional approach of product managers envisioning requirements is directly replaced by real customer behavior.

Under this structure, the organization also changes accordingly.

In traditional companies, intelligence is distributed among people and routed through hierarchical levels;
Here, intelligence exists within the system, while humans are relegated to the "edge."

The edge is where intelligence meets reality. Humans can perceive intuition, culture, trust, and complex situations that models cannot capture, and play a role in ethical and high-risk decision-making. But they do not need to collaborate through hierarchical structures because the world model already provides the necessary context.

In practice, organizations will simplify into three types of roles:

IC (Individual Contributor): An expert in building capabilities, models, and interfaces;
DRI (Direct Responsible Person): Mobilizes resources around specific problems or customer outcomes;
Player-coach: Participates in frontline work and cultivates talent, replacing traditional managers.

There is no longer a need for a fixed middle management team; the rest of the coordination work is handled by the system.

Block is still in the early stages of this transformation, and it will be a difficult process, with some attempts likely to fail. But we're sharing this direction because we believe every company will eventually face the same question: Are you deepening your understanding of a complex problem?

If the answer is no, AI is merely a cost-reduction tool; if the answer is yes, AI will reveal the true nature of a company.

Block's answer is the "economic graph": connecting millions of merchants and consumers, understanding the behavior of both ends of a transaction in real time, and continuously accumulating data. We believe that this model of "organizing companies with intelligence rather than hierarchy" will reshape how various types of businesses operate in the coming years.

The speed of a company's growth essentially depends on the speed of information flow. Hierarchical structures and middle management slow this flow down. For two millennia, from the Roman army to the modern enterprise, we have had no better alternative. But now, this premise is changing. Block is building the next form.