The Great Restructuring: How AI Workforce Intelligence Will Transform 28 Million American Businesses

For two thousand years, hierarchy has been the dominant organizational technology. Not because it was elegant. Not because it was efficient. Because humans have finite bandwidth for coordination, and hierarchy was the best available solution for routing information across groups of people. Every layer of management exists to multiply coordination capacity. Every vice president, director, and team lead is, at their core, a human router: receiving information from below, synthesizing it, and passing it upward. The entire management profession is built on this constraint.

That constraint no longer exists. Large language models can hold context across an entire company’s operations. Persistent memory systems can retain institutional knowledge that used to live exclusively in people’s heads. Agent architectures can compose capabilities across scheduling, CRM, email, analysis, and content without a human routing the request between departments. For the first time in organizational history, the bandwidth bottleneck that created hierarchy has a technological solution. The coordination layer can be replaced.

The businesses that restructure around this reality, around what this paper calls AI Workforce Intelligence, will compound advantages that are nearly impossible to replicate. Their world models will deepen every month. Their intelligence layers will compose increasingly sophisticated solutions. Their people will focus on judgment, creativity, and relationship, the work that actually creates value. The businesses that wait will not just fall behind. They will be coordinated to death by competitors who moved first.

The organizational chart is older than the corporation. The Roman military operated with a precision that would not be matched for over a millennium, and the reason was structural. A contubernium of eight soldiers reported to a decanus. Ten contubernia formed a century under a centurion. Six centuries made a cohort. Ten cohorts made a legion under a legatus.

This was not designed for aesthetics. It was designed for one purpose: moving information and orders across distance with acceptable fidelity. A legatus commanding 5,000 men could not personally coordinate with each soldier. The hierarchy existed so that a command issued at the top could reach the front line through a series of trusted intermediaries, each responsible for a manageable span of control. The centurion, the backbone of Roman military power, was fundamentally a coordination node. He translated strategic intent into tactical action for eighty men.

The insight that matters: every layer in this structure added coordination capacity. But every layer also added latency and lost context. An order from the legatus arrived at the front line slightly different than intended. A signal from the front line reached the legatus slightly late. The hierarchy was a tradeoff between scale and fidelity, and Rome accepted that tradeoff because there was no alternative.

In the early 1800s, after catastrophic defeats by Napoleon at Jena and Auerstedt, the Prussian military undertook the most significant organizational reform in modern history. Gerhard von Scharnhorst and August von Gneisenau created the General Staff system: a permanent body of trained officers responsible for planning, coordination, and institutional memory.

This was the first modern bureaucracy. The General Staff did not fight. They coordinated. They maintained maps, logistics tables, contingency plans, and communication protocols. They ensured that what the commander intended was what the field units executed. The system was so effective that it became the template for every large organization that followed, military and civilian alike.

The Prussian innovation was not hierarchy itself. It was the professionalization of coordination. They recognized that as operations scale, you need dedicated humans whose entire job is making sure the right information reaches the right people at the right time. This is the origin of middle management.

In 1855, Daniel McCallum, General Superintendent of the New York and Erie Railroad, produced something that had never existed before: a formal organizational chart for a business enterprise. The NY&E spanned over 460 miles with more than 5,000 employees across five operating divisions. McCallum recognized that the informal coordination that worked for a 50-mile railway was “entirely inadequate to the wants of a long one.”

His chart, illustrated by civil engineer George Holt Henshaw and rediscovered at the Library of Congress in 2005, established the principles we still use: functional departments, clear lines of authority, defined reporting relationships, and separation of line and staff roles. McCallum’s six management principles, published in 1856, read like a modern management textbook: proper division of responsibilities, sufficient authority for each role, means of verification, prompt failure reporting, and accountability through every grade.

The railroad forced the invention of the corporate hierarchy for the same reason Rome invented the military one: scale exceeded the capacity of informal human coordination. When you have 5,000 people spread across 460 miles, someone has to be in charge of making sure the 9:15 from Dunkirk does not collide with the 10:30 from Hornellsville. That someone is a manager.

Alfred Sloan’s reorganization of General Motors in the 1920s introduced the divisional structure: semi-autonomous business units (Chevrolet, Buick, Cadillac), each with its own management hierarchy, coordinated by a central executive office. Sloan proved that a company could operate at massive scale by layering hierarchies within hierarchies, each managing a piece of the whole. His 1963 memoir, My Years with General Motors, became the founding text of professional management.

Peter Drucker built the intellectual framework. His concept of “management by objectives,” introduced in The Practice of Management (1954), defined the manager as a coordinator of human effort toward defined goals. Drucker did not describe managers as leaders or visionaries. He described them as people who ensure that “the right things get done.” Coordination, allocation, alignment.

By the 1970s, management had become a profession with its own graduate schools, its own literature, and its own career track. The coordination layer was no longer a necessary evil. It was an industry.

Between 1960 and 1990, the number of managers and administrators in the American workforce roughly doubled relative to the number of production workers. The Bureau of Labor Statistics tracked this expansion as organizations grew more complex, more geographically distributed, and more specialized. Every new product line, every new office, every new regulation created demand for more people whose job was to coordinate other people.

This was rational. As companies grew, the coordination burden grew faster. A 10-person company needs almost no formal coordination. A 100-person company needs managers. A 1,000-person company needs managers of managers. A 10,000-person company needs an entire apparatus of VPs, directors, senior managers, program managers, project managers, and coordinators whose primary output is alignment.

The thread that runs through all of this, from Roman legions to Fortune 500 companies, is one insight: hierarchy exists because humans have limited bandwidth for coordination. Every organizational layer is a bandwidth multiplier. But it is also a latency adder, a context loser, and a cost center. For two thousand years, we accepted this tradeoff because there was no better option.

The modern American business pays an extraordinary tax for coordination, and most of its leaders do not realize the full cost because the tax is invisible. It does not appear on a balance sheet. It appears in the gap between what a company could accomplish and what it actually does.

Asana’s Anatomy of Work Index, surveying over 10,000 knowledge workers, found that the average knowledge worker spends 60% of their workday on “work about work”: chasing status updates, searching for information, switching between tools, sitting in alignment meetings, and routing requests to the right person. Managers spend even more, roughly 62%. That means for every dollar spent on a knowledge worker’s salary, only 40 cents goes to the skilled work they were hired to do. The other 60 cents goes to the coordination tax.

Microsoft’s Work Trend Index tells a similar story. Since February 2020, people are in 3x more Teams meetings and calls per week, a 192% increase. The heaviest meeting users, the top quartile, spend more than 7.5 hours per week in meetings alone. And these are just the formal meetings. They do not count the Slack threads, the email chains, the “quick syncs,” or the meetings to prepare for meetings.

The coordination tax manifests in four ways:

Every coordination failure is invisible until it becomes something else: a customer problem, a missed deal, a burned employee, a product that shipped late, a market that shifted before leadership noticed. Companies do not track “decisions we made too slowly” or “insights that died in middle management.” They track revenue, churn, and employee turnover, all of which are downstream effects of a coordination layer that cannot keep pace with the complexity it manages.

For a 50-person company with an average fully loaded cost of $95,000 per knowledge worker, the coordination tax represents roughly $2.85 million per year spent on work about work. That is not an abstraction. That is payroll spent on people doing work that a well-designed intelligence system could handle in milliseconds.

The conversation about artificial intelligence in business has been dominated by one question: what can AI do? Can it write emails? Can it generate reports? Can it analyze data? These are the wrong questions. They reduce AI to a tool, a faster version of something a human already does.

The right question is: can AI coordinate?

Until recently, the answer was no. Early AI systems were narrow. A chatbot could answer questions. An analytics tool could surface patterns. A scheduling assistant could book meetings. But none of them could hold the full context of a company’s operations and orchestrate actions across systems, people, and processes. You still needed humans to route the information between the chatbot, the analytics tool, the scheduling system, the CRM, and the email platform.

That limitation dissolved in the last eighteen months. Three capabilities converged simultaneously.

The cost of AI inference has dropped 90% in two years. A coordination task that cost $0.10 in 2024 costs $0.01 today. That $95,000 human coordinator processing 50 tasks a day at roughly $7.60 per task is competing against an intelligence layer processing the same task for pennies. The curves have crossed, and they are diverging further every quarter.

But cost is only half the equation. The other half is capability. A human coordinator processes 40-50 coordination tasks per day. An intelligence layer processes thousands. It does not get tired, does not forget context between conversations, does not need three days to get up to speed after a vacation. It runs at the speed of the network, not the speed of human cognition.

On March 31, 2026, Jack Dorsey published “From Hierarchy to Intelligence,” a document that should be required reading for every business owner in America. Co-authored with Roelof Botha, it describes Block’s plan to fundamentally restructure how the company operates. Not a reorganization. Not a round of layoffs dressed up as transformation. A genuine rethinking of how a $20 billion company coordinates 12,000 employees.

The World Model is your company’s living memory. Not a database. Not a dashboard. A semantic understanding of how your business actually operates.

A database stores records. A World Model understands relationships. It knows that your biggest client always pays late in Q1 because their fiscal year starts in February. It knows that your lead pipeline dries up in August because your target market goes on vacation. It knows that when your operations manager is overloaded, response times to prospects slip, and close rates drop two weeks later. It connects the dots that no single employee can see because no single employee has access to all the dots simultaneously.

The World Model answers a question that every business owner asks and no existing system can answer reliably: “What is actually happening in my business right now?” Not what the CRM says. Not what the last status meeting covered. What is actually happening, synthesized from every signal across every system.

Building a World Model is less daunting than it sounds. Most businesses already generate the raw material: CRM records, calendar events, email threads, project tasks, financial transactions, customer communications. The World Model is built by connecting these existing sources to a persistent memory layer that learns the patterns, relationships, and rhythms of your specific operations. It starts knowing nothing. Within weeks, it understands your business better than a new hire could in six months. Within a year, it holds institutional knowledge that is immune to employee turnover.

The Intelligence Layer is not a chatbot. It is not a rules engine. It is an orchestration system that composes capabilities into solutions.

Consider what happens today when a qualified lead fills out a contact form at 9 PM on a Tuesday. The form submission sits in an inbox. The next morning, a coordinator reviews it, checks the CRM to see if it is a returning contact, looks at the sales team’s calendar to find availability, drafts a personalized response, sends it, and creates a follow-up task. Total elapsed time: 14 to 18 hours, assuming no one is out sick or in back-to-back meetings.

An Intelligence Layer composes the same solution in seconds. It reads the form submission, checks the CRM for history, analyzes the lead against qualification criteria, checks team availability, drafts a contextually appropriate response drawing from the World Model’s understanding of what messages convert best for this type of prospect, sends the response, schedules the follow-up, and updates the CRM. Not because it was programmed to follow those exact steps, but because the Intelligence Layer understands the goal (qualify and engage the lead) and composes available capabilities (email, CRM, calendar, analytics) to achieve it.

Traditional roadmaps are guesses: someone in a planning meeting hypothesizes what the company should build next. An Intelligence Layer’s roadmap writes itself from reality. Every failed composition is a data point about what the business needs that it does not yet have. The roadmap is not a guess. It is a gap analysis generated by actual operations.

AI Workforce Intelligence does not eliminate people. It transforms what people do.

The edge is where intelligence meets reality: the handshake with a client, the judgment call on a deal, the creative insight that no model can replicate, the relationship with a regulator, the read on a room during a negotiation. Humans own the edge. That does not change. What changes is everything between the edge and the edge.

Today, most of a knowledge worker’s day is spent in the middle: routing information, attending alignment meetings, updating systems, chasing follow-ups. AI Workforce Intelligence collapses the middle. What remains are the high-judgment, high-relationship, high-creativity roles that humans do better than any system.

Block calls this the “player-coach” model: people who both do the work and train the intelligence. Every human interaction with the system teaches it. Every correction improves its model. The people are not just workers. They are trainers of a system that gets better every day.

Jack Dorsey’s architecture solves Block’s problem: one company, one World Model, one Intelligence Layer. But Block is a $20 billion company with 12,000 engineers. The 34.8 million American small and mid-sized businesses face a different constraint. They cannot build world-class industry intelligence alone.

This is where AI Workforce Intelligence diverges from Block’s framework and introduces a fourth architectural pillar: Collective Intelligence.

Collective Intelligence sits between the World Model and the Intelligence Layer. It is the shared substrate of industry-specific knowledge, skills, and pattern recognition that no single business could build alone but that every business in a vertical benefits from.

Consider what this means in practice. A single accounting firm’s World Model understands that firm’s clients, seasonal patterns, and operational rhythms. Useful, but limited. Now connect fifty accounting firms to a Collective Intelligence layer. That layer understands engagement letter patterns across thousands of clients, seasonal staffing curves for the entire profession, common audit findings by industry, client churn signals that individual firms miss because their sample size is too small, and regulatory changes before they hit the mainstream press.

This is the architectural difference between a tool that helps one company and a network that elevates an entire industry. It is also the structural moat. Technology can be copied. An intelligence layer trained on the collective operations of fifty businesses in a specific vertical, refined over two years of continuous learning, cannot be replicated with a product launch. It can only be built over time, client by client, signal by signal, pattern by pattern.

The most underappreciated feature of AI Workforce Intelligence is that it compounds. A traditional business does not get meaningfully better at coordination as it ages. Employees leave, taking knowledge with them. New hires start from zero. Processes get documented in manuals that no one reads. Institutional memory degrades continuously.

A World Model does the opposite. Every customer interaction teaches it. Every operational decision refines its understanding. Every failure expands its capability. After twelve months of continuous operation, a company’s World Model understands its business with a depth that no single employee, and arguably no team of employees, can match. That understanding is durable. It does not resign. It does not need to be retrained from scratch when the company hires a replacement.

After twenty-four months, the compounding effect becomes a structural advantage. The Intelligence Layer is composing solutions that draw on two years of learned patterns, customer preferences, seasonal dynamics, and operational rhythms. A competitor starting from scratch is two years behind, not in technology, but in accumulated intelligence about how the business actually works.

The compounding effect accelerates further when World Models learn across businesses in the same vertical. A World Model trained on fifty law firms understands the rhythms of legal operations, client intake patterns, court filing deadlines, associate utilization curves, and seasonal demand, better than any single firm’s staff could. Each new firm that joins the network contributes data that refines the base model for all.

This is the moat. Technology can be copied. Features can be replicated. A World Model that has been learning for two years, trained on the operational patterns of dozens of businesses in a specific vertical, cannot be copied. It can only be built over time. The first mover in each industry vertical accumulates an intelligence advantage that widens with every month of operation.

Automating tasks instead of coordination. The point is not to make individual tasks faster. It is to eliminate the routing, synthesizing, and aligning that connects tasks. Automating a report is useful. Eliminating the three meetings that exist because the report was not available in real time is transformative.

Underestimating the World Model. The World Model is not a nice-to-have. It is the foundation. Without it, the Intelligence Layer is composing solutions without context, like a new employee on their first day. Invest in the World Model first.

Treating AI as a tool instead of a team member. A tool does what you tell it. A team member understands the mission, anticipates needs, and acts proactively. AI Workforce Intelligence is designed for the latter. If you are still manually triggering every action, you have a tool, not an intelligence.

The compounding nature of AI Workforce Intelligence creates a first-mover advantage that is structurally different from typical technology adoption. When a competitor adopts a new CRM, you can adopt the same CRM and be at parity within months. When a competitor has been building a World Model for two years, you cannot close that gap with a purchase order.

The businesses that begin building AI Workforce Intelligence in 2026 will enter 2028 with a coordination advantage that late adopters will spend years trying to close. Their teams will be smaller, faster, and focused on judgment and relationships. Their World Models will understand their markets with a depth that new entrants cannot match. Their Intelligence Layers will be composing solutions that competitors are still routing through three layers of management.

The question is not whether your industry will restructure around AI Workforce Intelligence. Jack Dorsey is restructuring Block. Jensen Huang is telling every company to build an agent strategy. The cost curves have crossed. The technology exists. The organizational design pattern is proven.

The question is whether you lead the restructuring or react to it.

This is not a technology trend. It is a structural shift in how businesses operate, the most significant since Alfred Sloan invented the divisional corporation a century ago.

For two thousand years, hierarchy was the only solution to the coordination problem. It scaled human bandwidth at the cost of latency, context loss, and an ever-growing class of workers whose job was to route information between other workers. That tradeoff was rational when it was the only option.

It is no longer the only option.

The Great Restructuring is already underway. Block is rebuilding a $20 billion company around World Models and an Intelligence Layer. The infrastructure to bring this architecture to small and mid-sized businesses exists today. The 34.8 million American businesses that adopt AI Workforce Intelligence will define the next era of commerce, not because they adopted a new tool, but because they eliminated a two-thousand-year-old constraint.

1. McCallum, Daniel. “Superintendent’s Report.” Annual Report to the Stockholders of the New York and Erie Railroad Company, March 25, 1856. Reprinted in American Railroad Journal, 1857.

2. Chandler, Alfred D., Jr. The Visible Hand: The Managerial Revolution in American Business. Cambridge: Harvard University Press, 1977.

3. Sloan, Alfred P., Jr. My Years with General Motors. New York: Doubleday, 1963.

4. Drucker, Peter F. The Practice of Management. New York: Harper & Brothers, 1954.

5. Asana. “Anatomy of Work Index 2022.” Asana, Inc., 2022. Survey of 10,624 knowledge workers across the U.S., UK, Australia, France, Germany, and Japan.

6. Asana. “Anatomy of Work Global Index 2023.” Asana, Inc., 2023.

7. Microsoft. “Work Trend Index: Will AI Fix Work?” Microsoft, May 2023. Based on Microsoft 365 usage data and survey of 31,000 people in 31 countries.

8. Microsoft. “Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part.” Microsoft, May 2024.

9. Dorsey, Jack, and Roelof Botha. “From Hierarchy to Intelligence.” Block, Inc., March 31, 2026.

10. Huang, Jensen. Keynote Address, NVIDIA GTC 2026. San Jose, CA. March 17, 2026.

11. U.S. Small Business Administration, Office of Advocacy. “2024 Small Business Profile.” November 2024.

12. U.S. Small Business Administration, Office of Advocacy. “Frequently Asked Questions About Small Business, 2024.” July 2024.

13. U.S. Bureau of Labor Statistics. “Occupational Employment and Wage Statistics.” Various years. Data on management occupations as share of total employment, 1960-2020.

14. Craig, Gordon A. The Politics of the Prussian Army, 1640-1945. Oxford: Oxford University Press, 1955.

15. Goldsworthy, Adrian. The Complete Roman Army. London: Thames & Hudson, 2003.