Companies announce layer counts. The number that actually moves is span of control, and the conversion between the two is brutal.
Uber is cutting 3,300 roles, about 10% of its workforce, halving the number of “micro-teams” with one or two direct reports, and shrinking its manager population by 20%. Run the arithmetic and that combination lifts Uber’s average span of control by roughly 12%, which at 34,000 people is worth about half a management layer. Intel’s much louder claim, 12 layers down to six, quietly implies its average manager went from about 2.6 reports to about 6.6. Layers are a logarithm of span. Almost nobody announcing a flattening seems to be pricing that.
Here is the thing about org charts: everybody talks about them in the wrong unit.
Executives announce layers because layers are legible. Six sounds better than 12. Five below the CEO sounds disciplined. But layers are an output. The input is span of control, the average number of people each manager oversees, and the relationship between the two is logarithmic. That means small changes in span produce dramatic changes in layers, and dramatic-sounding manager cuts can produce almost nothing.
That gap between the announcement and the arithmetic is where this week’s news lives.
What Happened
On September 2, Uber told employees it would eliminate about 3,300 roles, roughly 10% of a workforce that stood near 34,000 at the end of 2025. It is the company’s largest reduction since the pandemic collapse of 2020.
The structural detail is more interesting than the headline number. Uber said it would cut the number of teams with only one or two direct reports by nearly half, reduce the population sitting seven or more reporting layers below the CEO by 20%, and shrink its manager count by 20% overall. Some of those managers are leaving. Others are staying and moving into individual-contributor roles.
Dara Khosrowshahi framed it as clearer ownership and faster decisions, with savings redirected into growth and the robotaxi push. Uber is also merging engineering, science, and delivery into combined organizations, consolidating its restaurant, retail, and direct delivery operations, and ending remote work for all but a sliver of staff.
Notably, Uber did not blame AI. The memo language is about complexity and coordination, not agents replacing coordinators. The Wall Street Journal reported the same day that the small-team boss has become the most exposed role in corporate America, and Uber is only the loudest recent example.
Markets liked it. The stock ticked up on the announcement, in a year where it has otherwise been down.
The Backstory
Uber is late to this. The micro-team purge has been running for at least two years.
Google’s people-analytics leadership told staff last year that the company had 35% fewer managers overseeing fewer than three people than it had a year earlier. Many of those managers stayed at Google as individual contributors. Axon cut hundreds of manager roles and moved small-team supervisors into IC work, with president Josh Isner arguing that an average of four employees per manager was indefensible for a company that wanted to move quickly.
Coinbase went furthest in public. In May it cut 14% of staff, roughly 700 people, and announced it would have no “pure managers.” Brian Armstrong’s replacement concept was the player-coach, a leader who ships alongside the team, with the hierarchy capped at five layers below him and an employee-to-manager ratio of 15 or more. Block’s Jack Dorsey has floated an even blunter version, describing an ideal in which the layer count is effectively zero.
Meanwhile the baseline moved on its own. Gallup put the average US manager’s span at 12.1 direct reports in 2025, up from 10.9 the year before. Meta’s applied engineering group reportedly runs closer to 50 to one. Gartner had predicted that a fifth of organizations would use technology to flatten by 2026 and eliminate more than half of middle management along the way.
The demand side moved too, which is the part most coverage skips. Robert Half found 40% of Gen Z professionals want a promotion that does not turn them into a manager, the highest share of any generation. Robert Walters put the figure at 57% preferring to advance as individual contributors. The people companies are declining to promote into management were increasingly declining the job anyway.
The Plan
Strip away the language and every one of these announcements is the same instruction: raise span, and let the layer count fall out of it.
That is where the arithmetic gets uncomfortable, because layers and span are the same variable wearing different clothes. If a company of N people organizes itself with an average span of s, the number of layers it needs is roughly the logarithm of N divided by the logarithm of s. Doubling span removes about two layers, and it removes about two layers whether the company has 5,000 people or 500,000.
Apply it to Intel and the claim gets much bigger than it sounds. Intel ended fiscal 2024 with 108,900 employees. Twelve layers across that headcount implies an average of about 2.6 direct reports per manager. Six layers across the 85,100 it reported at the end of fiscal 2025 implies about 6.6. Halving the layer count does not halve the span. It squares it.
Which reframes the whole micro-team story. At 12 layers, Intel did not have a micro-team problem at the margins of the org chart. The micro-team was the median.
Now apply the same math to Uber, and the claim gets much smaller than it sounds. Cutting 20% of managers while also removing 10% of the people those managers were managing raises average span by about 12.5%. A company that had roughly seven layers ends up with roughly six and a half. The 20% headline buys half a layer.
And for scale: at the US average span of 12.1, a 34,000-person company fits inside about 4.2 layers. Uber’s stated goal is not to get there. It is to reduce, by a fifth, the number of people sitting seven or more layers down.
The Business Model Angle
Layers are the vanity metric, span is the operating metric. Announcing layer counts is announcing a logarithm, which flatters small moves and understates large ones. Intel’s flattening was a genuine structural rebuild that the “12 to six” phrasing actually undersells. Uber’s is a trim that the “20% of managers” phrasing oversells. Any operator or investor reading these releases should convert to span before forming a view, because the same sentence can mean two completely different companies.
The micro-team is usually a fossil, not a design. A manager with one or two reports is rarely someone who was hired to manage one or two people. It is far more often a team that shrank in a previous layoff and never got backfilled, or an acquisition that was absorbed and never re-parented. Companies are cleaning up the sediment of their own prior cuts and presenting it as a redesign. That also explains why this cut is so popular: detaching a manager with one report orphans one person, while detaching a manager with 12 orphans 12. The flattening runs from the leaves inward because that is the cheapest place to start, not because that is where the value is.
You buy span with proximity, and Uber just paid for it. The same memo that widens manager scopes also ends remote work for all but a fraction of staff. Intel already requires four days on site. This is not two separate policies stapled together. Managing 12 people you rarely see is a materially harder job than managing four you never see, and the return-to-office mandate is the enabling condition for the span increase, not a culture war running alongside it. Anyone modeling the cost savings should note that some of the savings are being spent on real estate and attrition.
The manager title was a currency, and it is being demonetized. For decades the only way to give a strong contributor more money and more status was to give them somebody to supervise. Micro-teams are what that constraint looks like when it shows up on an org chart. Deleting them without building a credible individual-contributor ladder means deleting the promotion mechanism and keeping the promotion expectation. The good news is that the workforce got there first: the Robert Half and Robert Walters numbers say a majority of young professionals would rather have the money than the reports. The bill is that the raise now has to be paid in cash instead of headcount.
The Risk
Start with the honest limitation in our own math. The layers-equal-log-of-span formula assumes a roughly uniform tree, and no real company is one. A 300-person operations function with a span of 30 sits inside the same org chart as a six-person research team reporting to a director. Average span is a crude summary statistic, and the numbers above are order-of-magnitude reasoning, not an audit. They tell you the shape of the tradeoff, not the exact depth of any particular branch.
There is also a reasonable defense of the layer framing. David Zinsner’s account of Intel’s problem was about veto power and slow decisions, not about ratios. Layers are what an employee actually experiences: the number of approvals between an idea and a decision. If the goal is to shorten that chain, counting the chain is not obviously the wrong measure, even if span is the lever that moves it.
Not every micro-team is waste, either. Specialist functions, safety-critical engineering, regulated roles, newly acquired teams, and geographic coverage all legitimately run small. A four-to-one ratio in a sales operations group and a four-to-one ratio in a group certifying autonomous driving software are not the same fact, and a policy that treats them identically will remove the wrong ones. Companies applying a headcount threshold rather than a judgment are optimizing for a number that is easy to query in an HR system.
The pipeline question is real but already well covered, including on this site. When the base of the pyramid thins, the mechanism that produced senior people thins with it, an argument we made at length about the consulting pyramid and the entry-level collapse. The micro-team was, among other things, the cheap trial run where a company found out whether someone could manage before betting a real team on them. Removing the trial does not remove the need for the result.
Finally, purges like this have reversed before. The middle-management cuts of the early 1990s were followed by a rehiring cycle, and the coordination work does not disappear when the coordinator does. It reappears somewhere, often at a higher unit price, which is the pattern behind the chief of staff boom. Whether AI changes that this time is the open question, and Uber pointedly did not claim that it does.
Quick Questions
What exactly is a micro-team? A team with only one or two direct reports. Google’s internal definition used fewer than three. Uber is halving the number of them.
Is this an AI layoff? Uber did not say so. Its stated reasons are organizational complexity and reinvestment in ridesharing, delivery, and robotaxis. Coinbase, by contrast, explicitly tied its restructuring to AI-driven productivity. Treat them as different cases.
Do the cut managers lose their jobs? Not all of them. At Google, Axon, and Uber, a meaningful share moved into individual-contributor roles. Uber’s 3,300 figure combines managers who are leaving with non-managers who are leaving.
How flat is flat? Coinbase’s stated 15-to-1 ratio would organize its remaining workforce in about three layers, comfortably inside its own five-layer cap. The ratio is the binding constraint. The cap is decoration.
What should a smaller company take from this? Check your own span before copying anyone’s layer count. If your average manager has three reports, you have a compensation architecture problem wearing an org chart costume.
The Business Model Analyst Take
Corporate America has spent two years announcing flattening in a unit that hides what it is doing. Intel squared its span and described it as halving something. Uber cut a fifth of its managers and moved the org chart by half a layer. Both are defensible decisions. Neither announcement told you which one you were looking at.
The micro-team boss is being cut because that role is the easiest thing in the building to remove, and because it was never really a management job in the first place. It was a pay grade with a person attached. Companies are discovering they can delete the person. What they have not yet worked out is what replaces the pay grade.
