The middle manager's decade: delayering from the 1990s to 2026

In 1958, two professors predicted in the Harvard Business Review that information technology would eliminate the middle manager. Sixty-eight years later, Amazon has cut about 30,000 corporate jobs since October 2025 while “reducing layers”. Google says it has 35% fewer managers of small teams. Meta has set up an AI unit with 50 engineers per manager. The middle manager has been declared dead at least twice before. The evidence on what happens when one manager runs 15, 25 or 50 people is thinner than the confidence behind these moves.

Contents · 10 sections

TL;DR #

• The idea that computers would remove middle managers dates from 1958. The 1980s and the 1990s re-engineering wave both flattened firms, and re-engineering’s own founders later admitted they had underrated the human side.
• Since 2023, Meta, Amazon and Google have documented flattening: fewer layers, more direct reports, many managers turned back into individual contributors.
• Gallup’s data show the average US manager now has about 12 direct reports, but the median is still 5 to 6. Engagement falls as teams grow unless the manager is strong and feedback is frequent.
• Spans of 15 or more can work for similar, experienced work. I know of no evidence that they work where managers are also expected to coach and train.

The prediction that keeps coming back #

In my library, Thomas W. Malone’s Inventing the Organizations of the 21st Century opens its chapter on control with that old forecast: “In 1958, Leavitt and Whisler predicted that IT would lead to the elimination of middle managers and to greater centralization of decision making.” Malone’s own reading is that changes “have occurred in both directions” since then. Computers have both centralised and decentralised work, depending on the firm.

The first real wave came in the 1980s. As cheap computers took over the middle manager’s job of collecting and passing on numbers, firms flattened. Average spans of control moved from about one manager to four people toward about one to ten, as the standard reference summaries put it.

The 1990s: re-engineering and its regrets #

Then came the manifesto. Michael Hammer’s 1990 Harvard Business Review article, “Reengineering Work: Don’t Automate, Obliterate”, and his 1993 book with James Champy, Reengineering the Corporation, told managers to redesign work from a clean sheet. By 1993, as many as 60% of the Fortune 500 said they were re-engineering or planned to, according to a US General Accounting Office guide. Layers of middle management were among the first things to go.

At the time, the mood in management books was cheerful. In my library, Bob Nelson and Peter Economy’s The Management Bible sums up the era: it “was not a happy time for the people who lost their jobs”, but it “was a happy time for their newly trimmed down and responsive organizations.”

The people inside saw it differently. Resizing the Organization, edited by Kenneth De Meuse and Mitchell Marks, also on my shelf, studies the people who stayed. One chapter puts their experience in a single line: “Middle managers were paralyzed by this lack of direction, waiting to see which way they should lead their work teams.” In the same book, Wayne Cascio and Clifford Young followed S&P 500 firms from 1982 to 2000 and found that firms that cut staff never beat the profitability of stable employers.

The founders of the movement came to regret how it was used. Thomas Davenport, an early proponent, wrote that “once out of the bottle, the reengineering genie quickly turned ugly.” Hammer himself said later: “I wasn’t smart enough about that. I was reflecting my engineering background and was insufficiently appreciative of the human dimension.” By the late 1990s, re-engineering had become a byword for layoffs, and the fashion faded.

2023–2026: the second flattening #

The current wave has its own vocabulary: “efficiency”, “layers”, “ownership”. Here is what is documented, as of 13 November.

Meta. On its February 2023 earnings call, Mark Zuckerberg declared a “year of efficiency” and said Meta was “flattening our org structure and removing some layers of middle management to make decisions faster.” In March 2023 it announced 10,000 more job cuts. In March 2026, according to the Wall Street Journal as reported by Fortune, Meta’s new applied AI engineering unit was designed with one manager for every 50 individual contributors. Fortune calls that double the 25-to-1 ratio usually seen as the outer limit.

Amazon. In September 2024, Andy Jassy asked every senior team to raise the ratio of individual contributors to managers by at least 15% by the end of March 2025. Amazon said it met the goal. A spokesperson told HR Grapevine this did not mean a 15% cut in management roles; teams were merged and some managers moved to non-manager roles. Then came the cuts: about 14,000 corporate roles in October 2025 and about 16,000 in January 2026. The January memo, from Beth Galetti, spoke of “reducing layers, increasing ownership, and removing bureaucracy.”

Google. In August 2025, a Google executive told staff the company had 35% fewer managers of small teams than a year earlier, according to CNBC. That figure covers managers with fewer than three reports, and many of them stayed on as individual contributors.

So the honest summary: flattening is documented at the biggest tech firms. Much of it is demotion and team merging, not only layoffs.

Does a span of 15 or more work? #

This is where the evidence gets thin.

The theory is old and inconclusive. V. A. Graicunas tried to work out the ideal span from mental capacity in 1933; Lyndall Urwick tried again in 1956. The standard summary today is that “no convincing theories have been presented,” because the right span depends on the work, the tools and the people.

The best recent data is from Gallup, published in January 2026:

• The average US manager now has 12.1 direct reports, up from 10.9 in 2024 and nearly 50% more than in 2013.
• The median is still about 5 to 6. A minority of very large teams pushes the average up: 13% of managers have 25 or more reports.
• Engagement falls as teams grow, especially for remote and hybrid managers. For on-site managers, team size mattered less.
• Talent matters more than span. Gallup’s own line is “Manager talent trumps span of control.” Highly talented managers kept large teams engaged; average ones did not.
• Feedback is the hinge. About 7 in 10 employees who got meaningful feedback in the past week were engaged, whatever their team size. Without it, about 1 in 4 were.

Gallup’s conclusion is careful: “Larger teams succeed only when organizations invest in the conditions that allow managers to lead them well.”

Gartner, meanwhile, reports that 75% of HR leaders say managers are already overwhelmed by their growing responsibilities.

Putting this together, here is what I think the evidence supports:

• Spans of 15 and more can work where the work is similar, the staff are experienced, the manager is strong, and someone else handles the paperwork.
• They work badly where the manager is also meant to coach, review and train. The standard management texts note that a manager who adds value by developing people needs a narrower span, not a wider one.
• 50 to 1 is not a management structure in the old sense. It is a bet that peers, tools and clear ownership will do what managers used to. That may work in an elite engineering team. I know of no evidence that it works across a large organisation.

Who did the middle managers’ real work? #

Org charts show middle managers as a layer of approval. In practice they did three things that rarely appear in a job description:

1. They translated strategy into this week’s priorities.
2. They carried knowledge sideways, knowing who in another department could fix a problem.
3. They trained people, especially juniors.

AI tools can help with the first: summarising, drafting and tracking. They do little for the second and almost nothing for the third. When firms flatten, those tasks don’t vanish. They fall on whoever is left, usually without being counted. That is the same knowledge-loss problem the 1990s ran into, in a new form.

What to watch #

These are inferences from the 1990s pattern, not facts.

• Whether flattening overshoots, then partly reverses. As in the late 1990s, firms may rediscover coordination problems. Titles such as “team lead” and “tech lead” would then quietly take on management work without the management title or pay.
• Whether the cost shows up in juniors first. With fewer managers, there is less time to train new people. Combined with weaker graduate hiring, this thins the pipeline for future leaders.
• Whether the language changes faster than the structure. “Year of efficiency” will be replaced by a new phrase. The question underneath, how many people one person can really lead, will stay.

What I don’t know #

The Gallup data covers US managers in general, not specifically the tech firms flattening fastest. Company figures on manager ratios come from memos and executive remarks, not audited data. The Davenport and Hammer quotes are as reported in the standard accounts of the period. And LSN News, my news engine, only has data since 7 November, so it can show this week’s coverage but not a decade-long trend.

Frequently asked questions #

Are companies really cutting middle managers? #

Yes, at several of the largest tech firms, and it is documented. Meta announced it was removing layers in 2023, Amazon set a target to raise the ratio of staff to managers in 2024 and cut about 30,000 corporate roles from October 2025, and Google reported 35% fewer managers of small teams in 2025. Much of it is demotion and team merging, not only layoffs.

How many direct reports can one manager handle? #

There is no proven number. Gallup finds the US average is about 12 and the median 5 to 6, and engagement falls as teams grow unless the manager is talented and gives frequent feedback. Wider spans suit similar, experienced work; narrower spans suit teams that need coaching.

Is this the same as 1990s re-engineering? #

The pattern is similar: a new technology, a promise of faster decisions, and cuts in the middle layers. In the 1990s the founders of re-engineering later admitted they had underrated the human side, and the fashion faded. Whether this wave ends the same way is not yet known.

Can AI replace what middle managers do? #

Partly. AI helps with summarising, tracking and drafting. It does far less for carrying knowledge between teams and almost nothing for training people. Those tasks fall on whoever is left.

The lesson #

Each time technology made information cheaper, someone predicted the end of the middle manager. Each time, the layer got thinner, the remaining managers got busier, and the uncounted work they did, translating, connecting and teaching, turned out to matter. Flatten if the work allows it, but count that work before you remove the people who do it.