Approximately one-third (34%) of American managers today supervise more than ten people. The median manager leads 5 to 6 — but the average has now reached 12.1 in 2025, pulled upward by a minority of very large teams, according to data compiled by Fortune from Bureau of Labor Statistics and Gallup sources. This progression has occurred without management training tools evolving at the same pace, without formal responsibilities being redefined, and without most of those affected being consulted.

Artificial intelligence was supposed to free managers from repetitive tasks. Instead, it has primarily made them responsible for larger teams. This isn’t quite what software vendors’ promises announced.

The Essentials

  • The average number of subordinates per American manager has now reached 12.1 in 2025, compared to 8.2 in 2013, according to BLS/Gallup data compiled by Fortune. The median, meanwhile, has remained stable at around 5 to 6 people.
  • Gartner forecasts that 20% of organizations will eliminate more than half of their middle management positions by the end of 2026.
  • Pressure to cut comes primarily from the promise of immediate gains on payroll, not from an analysis of the actual coordination functions performed by these hierarchical levels.
  • The documented risk is not the loss of positions, but the loss of knowledge transfer and informal coordination that middle management ensured invisibly.
  • Companies like Shopify and Klarna, which have cut their managerial workforce most aggressively, now serve as models — but their medium-term results remain uncertain.

The Average of 12.1: What the Number Reveals

The figure of 12.1 doesn’t tell the whole story by itself. What matters is what it has replaced. In 2013, the Gallup average was 8.2 subordinates per manager. The median, meanwhile, has remained stable at 5-6 people throughout the period — this figure best reflects the experience of the typical manager, with the average pulled upward by the 13% of managers supervising 25 people or more. Organizational science research has established since the 1990s that human capacity to maintain individualized working relationships plateaus between six and nine people, depending on task complexity. Beyond that, management ceases to be a relational act and becomes an administrative one.

At twelve, a manager who devotes thirty minutes weekly to each team member already spends six hours on this. In a forty-five-hour week, this leaves thirty-nine hours to manage their own operational responsibilities, cross-functional meetings, reporting, crisis situations, and non-routine decisions. In practice, one-on-one meetings are the first to be cut.

This shift has a clear accounting logic. Each managerial layer eliminated represents, in a large American organization, between $80,000 and $150,000 in annual payroll per position, figures to which are added charges, benefits, and office costs. For a company of 10,000 people that traditionally has 800 middle managers, reducing this number by a third represents an immediately visible saving in quarterly accounts. The value generated by these eliminated positions appears nowhere on the balance sheet.

What Gartner Measures and What It Does Not

Gartner’s forecast deserves to be read with precision. Twenty percent of organizations will eliminate more than half of their middle management positions by the end of 2026 — not all organizations, not all sectors, and not a complete disappearance. This is a signal of concentrated trend, not a prophecy of hierarchical collapse.

But the signal is serious because it often precedes broader diffusion. American tech companies that led the first wave of reductions between 2022 and 2024 — Meta, Alphabet, Amazon, Salesforce — served as references for entire sectors. The finance departments of industrial companies, financial services firms, and healthcare organizations looked at their savings and asked the same question: why not us?

This sectoral mimicry is documented. McKinsey, in its annual surveys on organizational transformations, notes that hierarchical restructuring decisions spread by sector after an initial wave in tech companies, with a lag of several quarters to several years depending on the industry. CFOs look at the same indices, read the same reports, and attend the same conferences.

What Gartner does not measure are deferred costs. The firm identifies the trend; it does not calculate what is lost in the transition. This is precisely where the discussion becomes more difficult — and more useful.

Invisible Coordination: What Nobody Put in a Job Description

The official functions of a middle manager are readable in any HR reference framework: set objectives, evaluate performance, conduct annual reviews, relay strategic decisions downward. These are visible functions, those you can put in a PowerPoint presentation to justify eliminating a position.

Invisible functions exist elsewhere. A frontline manager who has been at a company for five years knows why Project X failed in 2021, what tensions persist between sales and technical teams, how to circumvent the budget approval process when an important client calls on a Friday evening. He knows whom to call in which team to solve a problem in two hours instead of two weeks. He knows which colleague is going through a difficult period, who needs to be challenged, who needs to be protected.

This organizational capital appears in no database. It does not transfer through an internal wiki or an online training module. It transfers through proximity, over time, person to person. This is what labor economists call tacit knowledge transfer, and this is precisely what intermediate managerial layers ensured in an informal manner.

When these layers disappear, this capital does not disappear instantaneously. It degrades progressively, over twelve to twenty-four months, as the remaining knowledge bearers leave the organization or new situations reveal the absence of institutional memory. Mistakes repeat themselves. Frictions increase. Projects derail for reasons nobody quite knows how to explain. The connection with the frontline, already weakened by other recent transformations, stretches further.

AI Amplifies Teams but Does Not Replace Judgment

The initial promise of AI in organizations was precisely this: automate coordination and information relay tasks so managers can focus on high-value functions. In theory, a manager assisted by AI tools can track twelve simultaneous projects, identify delay risks, consolidate reporting automatically.

In practice, tools available in 2025 and 2026 do indeed do part of this work, under specific conditions of use. They read dashboards, aggregate project data, flag anomalies. What they do not do is assess whether a collaborator delivering a deliverable late is going through a personal crisis or simply lacks specific skills. They do not conduct difficult conversations. They do not arbitrate between two team members in territorial conflict. They do not transmit the unwritten rules that make an organization function.

The confusion between amplification and substitution is at the heart of the problem. AI agents can execute complex tasks, but they operate in controlled environments with defined objectives. Human coordination in an organization occurs in ambiguous environments, with contradictory objectives and actors whose motivations are partially opaque. This is structurally different.

Companies that have reduced intermediate management most aggressively — Shopify is often cited, as is Klarna — have communicated about their short-term efficiency gains. Their results at a three- to five-year horizon, particularly regarding the ability to conduct complex transformations, retain experienced talent, and maintain cohesion in distributed teams, will be the true indicators. They are not yet available.

Some Organizations Resist the Pressure and Redefine the Role

Not all organizations follow the trend. Some have made the opposite choice: maintain or strengthen intermediate management, but by redefining its content.

The logic is as follows. If AI tools take on reporting, data consolidation, and progress tracking, the manager freed from these tasks can indeed devote time to what machines cannot do: skills development, transmission of institutional knowledge, arbitration of complex situations, maintenance of team culture in increasingly hybrid and distributed work environments.

This is the so-called “manager-mentor” model, which companies like Microsoft, in some of its divisions, and several large European banks have begun to formalize. The manager is no longer evaluated on the volume of information relayed or the number of projects tracked. He is evaluated on the development of collaborators, on internal promotions generated within his team, on documented skills transfers.

This model assumes one thing that finance departments do not like to hear: that the return on investment of a good manager is real but deferred, difficult to attribute directly, and rarely visible in the next quarter. This is where pressure to cut reasserts itself in most organizations.

The question of survivor manager productivity is part of a broader tension that data on American productivity illustrate without resolving: aggregate gains exist, but their distribution across hierarchical levels remains opaque.

Training to Supervise Twelve People, Not Six

If the megamanager trend is here to stay, the open question is that of training. Management development programs in large American companies have been built around a ratio of six to eight subordinates. The taught competencies — weekly one-on-one meetings, personalized development plans, continuous feedback — assume a relational bandwidth that twelve direct reports no longer allow.

There are concrete approaches to adapting practice. Managers supervising large teams can formally delegate certain mentoring functions to senior collaborators without a managerial title, what’s called a “tech lead” in technology organizations. They can replace formal weekly one-on-one meetings with shorter and more frequent formats, less ritualized but more reactive. They can use AI tools not to track performance but to free up time for presence.

These adaptations exist. They are not systematically taught in management training programs at large American companies, which remain largely modeled on 1990s frameworks. The Society for Human Resource Management, the leading HR professional association in the United States, published in 2025 an update to its management competency frameworks that explicitly integrates managing large AI-assisted teams. It remains to be seen how many companies have adopted it.

The challenge for the next five years is not to choose between cutting managers and maintaining them. It is to decide what you expect of them. Information relays in a vertical hierarchy? AI does it better and cheaper. Mentors, coordinators, and culture bearers in distributed and complex organizations? Not yet, and not anytime soon.

Sources

  1. Fortune / BLS / Gallup — “The Megamanager Era: How Many Direct Reports Is Too Many?” (April 2026)
  2. Gartner — forecasts on middle management 2026 (cited in Fortune)
  3. McKinsey — annual surveys on organizational transformations (McKinsey Global Institute)
  4. Society for Human Resource Management (SHRM) — management competency frameworks 2025
  5. Gallup – Span of Control: What’s the Optimal Team Size for Managers? (January 2026)
  6. Gartner – Top Predictions IT 2025 and Beyond (October 2024)
  7. Meta SEC Filing – Announcement of layoffs November 2022
  8. BLS – Occupational Outlook Handbook, Management Occupations
  9. APQC – How Many Middle Management Positions Do You Need?