The Empire Trap
Second Level Thinking · Issue 49
Historically we measured executives by the size of the army they commanded. That proxy is breaking and what replaces it will separate those who compound from those who merely occupy.
Chamath Palihapitiya posted something on X last week that I haven’t been able to put down. The gist of it: enterprises still measure an executive’s importance by the size of the team they manage, but that’s an old proxy, built for a world where more people was the only way to do more work. That world is gone. The leaders who matter most over the next five years, he argues, will be the ones whose people spend all their time on the work that actually moves the business, because systems do more of the rest.
I’ve been chewing on this, and not just because it’s a sharp observation. It’s because I’ve lived inside the proxy he’s describing. Nearly two decades in heavy industry, in an organisation where headcount was currency. Where the honest, unspoken measure of your standing wasn’t the value you created, it was the size of the org chart beneath your name. Where restructures were a kind of quiet warfare over who ended up with more boxes. I’m not being cynical, it was rational behaviour in a system that rewarded exactly that. Which is precisely the point.
I want to trace where that proxy came from, why it worked for a hundred years, and what happens to the executive role now that the ground beneath it is genuinely shifting. Because here’s my thesis: leadership itself isn’t changing at all. The architecture around it is changing completely.
Where the proxy came from
The headcount-as-status equation isn’t a corporate pathology that appeared from nowhere. It was a rational response to a real constraint, and it has a specific origin.
Let’s go all the way back to General Motors in the early 1920s. Alfred Sloan inherits a shambolic, near-bankrupt confederation of car companies, far behind Ford and held together by little more than Billy Durant’s instincts. What Sloan builds in response is the blueprint for the modern corporation: a divisional structure where Chevrolet, Pontiac, Buick, Cadillac and others, each operate as autonomous profit centres, with wide operational freedom, all held to account by centralised financial control and a rigorous reporting system.
“Decentralised operations with centralised control.” It’s arguably the most consequential organisational idea of the twentieth century, and by the early 1930s it had taken GM past Ford to become the largest industrial enterprise the world had ever seen.
But notice what Sloan’s design implicitly assumes. Coordination is expensive. Information moves slowly, on paper, up and down a hierarchy. To do more, you genuinely need more people, and to know what those people are doing, you need layers of managers whose entire job is gathering, summarising and passing information upward. In that world, the size of your organisation is not a vanity metric. It is a reasonable proxy for your capacity to produce output. Headcount was throughput.
Andy Grove, running Intel sixty years later, saw the flaw in the proxy more clearly than anyone. His formula from High Output Management is still, to my mind, the single best definition of the executive job ever written:
A manager’s output equals the output of their organisation, plus the output of the neighbouring organisations under their influence.
Read that carefully. Grove is measuring output, not headcount. He is saying the manager’s contribution is the leverage they create, not the bodies they command. He was right in 1983 and almost nobody listened, because in an era where output and headcount were still tightly coupled, the distinction was academic. You could use one as a shorthand for the other and rarely be wrong.
Jack Welch spent the eighties attacking the same problem from the other side by stripping layers out of GE, widening spans of control, waging war on what he called the “boundaries” that slowed information down. Bezos did it again with the two-pizza team: if you can’t feed the group with two pizzas, it’s too big to move fast. Every great operator of the last century has been, in one way or another, fighting the coordination tax. But none of them could actually break the link between headcount and capacity. They could only manage it.
That link is now breaking. And when it breaks, the proxy doesn’t just weaken, it inverts.
The great flattening
This is no longer a theoretical argument. It is happening on balance sheets right now.
Gartner has projected that through 2026, roughly one in five organisations will use AI to flatten their structure, eliminating more than half of their middle-management positions. That’s not a fringe forecast; it’s already visible. Amazon cut around 14,000 middle-management roles. Oracle undertook a restructure tied explicitly to a massive AI infrastructure push, with leadership stating plainly that AI tools now let smaller engineering teams deliver more, faster while reducing the need for layers. Microsoft has reportedly pushed toward something like a ten-to-one engineer-to-manager ratio. Middle-management roles now account for a hugely disproportionate share of layoffs compared to a decade ago.
The most instructive case, though, is Bayer because they didn’t just cut, they redesigned. Under a model they call Dynamic Shared Ownership, Bayer collapsed their layers from roughly eleven or twelve down to six or seven. Average span of control widened to around fourteen, with some leaders carrying far more. They replaced the annual budget cycle with ninety-day resource cycles. They cut the internal rulebook by something like ninety-nine per cent. Thousands of roles went, the vast majority of them management.
And the results are worth pausing on. Their transformation chief reported that some customer teams doubled revenue inside a single ninety-day cycle. A major product launch arrived a year ahead of schedule. A production line that once required fifteen specialists now runs with five, resolving in days problems that used to take six months to grind through the approval chain.
Six months to days. Sit with that. That is not an efficiency gain. That is a different category of organism.
The honest counter-argument
I want to be careful here, because there’s another version of this story doing the rounds: Middle managers were never just human routers passing information up and down a pipe. They did something that doesn’t appear in any job description: they interpreted change for people. They noticed patterns, they gave feedback in context, they absorbed anxiety, they knew who was struggling before it showed up in a metric. Strip that layer out carelessly and you get exactly what the Korn Ferry data is starting to show; a large numbers of employees reporting that their leaders aren’t aligned, and that the absence of managers has left them feeling directionless.
There’s a second-order risk too, and it’s the one I’d worry about most if I were sitting on a board right now. The companies eliminating their middle managers in 2026 are simultaneously dismantling the apparatus that produces their senior leaders in 2032. The bench thins in two directions at once: the people who would have grown into the role are gone, and the ones who might have stepped up have looked at what the job now involves and decided it isn’t worth wanting.
So this isn’t a simple story of “layers bad, flat good.” It’s a story about what the layers were for. Where a management layer existed to move information, technology can now do that better. Where it existed to develop humans, exercise judgement, and hold a culture together; nothing has replaced it, and pretending otherwise is how you gut an organisation while congratulating yourself on the savings.
What actually changes, and what never does
Here’s where I’ve landed, and it’s the throughline I want you to take from this issue.
Leadership is constant. Everything I’ve ever seen that separates a genuine leader from a competent administrator remains untouched by any of this. Setting a mission worth following. Making the hard call with incomplete information. Telling the truth when it costs you. Developing the person in front of you. Building the trust that means people bring you the bad news early rather than late. Knowing which problems are worth solving; None of that has been automated, and none of it is close to being automated. If anything, it becomes more valuable as it becomes rarer.
The architecture is what changes. What’s shifting is the machinery through which a leader converts intent into outcome. For a century, that machinery was people. You wanted more throughput, you hired. You wanted more coordination, you added a layer. The org chart was the operating system.
Now the machinery is increasingly systems. And this is exactly what Chamath is pointing at when he talks about leaders whose people spend all their time on work that actually moves the business, “because they use systems and factories to do more of the rest.” The executive’s job is no longer to command the largest possible army — it is to design the largest possible leverage, so that the humans they do have are pointed at the small number of things only humans can do.
Which brings us back to something we covered a few issues ago in Musk’s algorithm and it lands differently now. Question every requirement. Delete what isn’t needed. Simplify. Accelerate. Only then automate. The ordering matters enormously here. Because the lazy version of this transformation is to automate the org chart you already have. The disciplined version is to first ask what work should exist at all, delete the rest, and then build the systems. One of those creates a faster version of the wrong company. The other creates a genuinely different one.
Three things to do about it
If you’re sitting in an executive seat right now, here is where I’d start.
Audit your own status metric. Ask yourself the uncomfortable question: if your headcount halved tomorrow but your output doubled, would you feel like you’d won or lost? Your gut reaction to that is diagnostic, and most of us don’t like the answer. Then take it to your organisation. How does your business actually measure executive seniority, by team size, budget, and boxes on a chart, or by outcomes owned? If it’s the former, you are still running Sloan’s operating system in a world that has moved on, and every rational person beneath you is being incentivised to empire-build. Fix the metric before you touch the structure.
Separate coordination work from judgement work. Take your leadership team’s calendar for the last month and split it into two columns. Column one: work that exists purely to move information around, such as status reports, updates, handoffs, meetings whose only output is a summary of another meeting. Column two: work that requires human judgement, relationship, taste, or courage. Column one is where systems should be doing the lifting, and every hour left in it is an hour stolen from column two.
Protect the human layer you’re about to lose. As you flatten, ask explicitly: what was this layer doing that wasn’t in the job description? If the answer includes developing your future leaders, interpreting change for anxious people, or catching problems before they hit a dashboard, you must consciously rebuild that function somewhere else. Wider spans mean each remaining leader carries more human load, not less. Formalise mentoring. Protect one-on-ones ruthlessly. Build the leadership pipeline on purpose, because the accidental one you used to have is being deleted. The companies that get this right will own the next decade. The ones that don’t will have wonderful margins in 2027 and no one fit to run the place by 2032.
So, the question I’d leave you with: if you stripped your role back to only the work that genuinely requires a human leader - the judgement, the mission, the people, the hard calls - how many hours a week would that actually be? And what on earth are you doing with the rest?
Parting Shot
A good reminder from Charlie Munger:
“Spend each day trying to be a little wiser than you were when you woke up. Discharge your duties faithfully and well. Step by step you get ahead, but not necessarily in fast spurts. But you build discipline by preparing for fast spurts ... In the end, most people get what they deserve.”
