After the Triangle — Part 5/5, flattening the organisation Lenni Laukkanen

The Triangle Isn’t a Diagram — It’s How Power Is Divided

I’ve given this same talk to more leadership teams than I can remember. I go through roughly what this series has covered: that the pyramid was a clever solution in its time, that AI dismantles its foundation, that decision-making power would be best taken closer to the point where value is created. And almost every time the same thing happens. There’s nodding around the table. Someone says that the flattening the organisation is exactly what they ought to be talking about more. The mood is enthusiastic. And then — nothing happens. Six months later the org chart looks exactly the same as before.

For a long time I thought the fault was in my talk. That I wasn’t convincing or concrete enough. But eventually I realised it isn’t about understanding at all. Everyone understood. They simply couldn’t. And that is the question of this final part: if the logic is so clear, why does nothing change?

And I’m not alone with my observation. When Tieto asked over six hundred Nordic IT decision-makers in early 2026 where the use of AI gets stuck, the answer was blunt: the market is stuck in the pilot trap, and scaling is held back by the organisation, not the technology. Only just under a third said AI was in production use across the whole business. The technology would be ready. The organisation is not.

A chart isn’t dismantled by an insight

The answer is that the pyramid is not a chart. It is how power, pay, career paths, and esteem are divided. And that is not dismantled by an insight — because it isn’t a matter of insight, but of infrastructure.

Think about what dismantling the triangle would actually mean. Someone loses their subordinates, and with them the reason for the pay they’ve earned. Someone realises that the career they’ve climbed for fifteen years, rung by rung, leads nowhere, because the rungs were removed. Someone whose entire professional identity is built on being a “manager” has to ask who they are without the title. These are not resistance born of an aversion to change. They are entirely rational reactions to the fact that the triangle is not a mere thinking aid — it is a map of who gets what, who decides, and who is listened to. Redrawing the chart is easy. Dismantling what the chart represents is something else entirely.

Why no one wants to give up legibility

There’s a deeper reason still, and the one who named it best was the American social scientist James C. Scott, in his 1998 book Seeing Like a State. Scott’s central concept is legibility. His observation was that the exercise of power requires the governed object to be legible from above — measurable, countable, structurable. States invented surnames, street addresses, and censuses to make people legible: before them, the central administration couldn’t tax, conscript, or police a population it couldn’t even name. Legibility was not a service to the citizen. It was a precondition of power.

The organisational pyramid does exactly the same. It makes the organisation legible to leadership. One glance at the chart shows who is responsible for what, who reports to whom, and whose desk a problem belongs on when something goes wrong. This is an enormous convenience for the one looking from above. And that is exactly why the pyramid is held onto even when it’s inefficient: giving it up would mean giving up the view by which the organisation is governed. A flat, networked organisation may be more efficient, but from above it’s blurry. And you can’t get a grip on the blurry. Here is the very heart of the difficulty of change: we don’t hold onto the pyramid because it works best, but because it is legible.

Flattening is real — and just as real is that it reverses

Here there’s a temptation to declare that the pyramid is now breaking for good. There’s plenty of data. The average number of subordinates per supervisor has risen, according to Gallup, from just over eight in 2013 to just over twelve by 2025. Amazon has deliberately cut management levels, and Bayer has slashed its layers from well over a dozen to half that and moved power to thousands of self-directing teams. The headlines speak of a great flattening.

But here is the point the hype skips and intellectual honesty demands saying aloud: every previous wave of flattening has reversed. When offices were computerised in the 1980s and 1990s, middle management was cut on exactly the same grounds as now — the machine handles the relaying of information, the intermediate rung is redundant. And then middle management came back, bigger than before, as organisations grew more complex and coordination proved harder than expected. The long-run figures are unambiguous on this: managers’ share of the workforce has not fallen over the decades but grown — in the United States from around nine per cent in the 1980s to over thirteen per cent by the early 2000s. And there are already signs that some of the AI-driven cuts made in haste are being reversed: in recent surveys, a large share of employers regret them, and forecasts estimate that a great many of the roles removed in AI’s name will be quietly hired back — sometimes cheaper, sometimes under a different title.

So is this time different? Perhaps. There’s a good case for thinking it’s different now: AI doesn’t replace the middle person with some vague efficiency, but does the very coordination and information-relaying work the middle rung existed for in the first place. But it’s equally possible that the cycle repeats. On earlier occasions it went like this: when the layers were removed, the work they did didn’t go anywhere. Someone had still reconciled, resolved conflicts, coached the new people, carried responsibility. When that someone was removed, the tasks drained to those who remained, whose number of subordinates swelled so large that no one had time to lead anyone anymore. And then the layers came back — under a different name, but to do the same work. Flattening collapsed under its own weight, because coordination proved harder than imagined. It may be that AI now handles exactly that coordination, in which case there’s no return. Or we may again find that part of that work was a human quality the machine doesn’t do. I don’t know which it’ll be. And I don’t trust anyone who claims to know.

What picture replaces the triangle?

Let’s go back to where this series began. In the first part I observed that the triangle is not the natural structure of the world but a built picture — assembled from three different traditions, and not even especially old. And that is exactly where the hope is. What is built can be rebuilt.

But here it’s worth being careful not to simply swap one picture for another just as rigid. A network sounds liberating, but it blurs responsibility: when everyone is connected to everyone, who finally decides? A circle sounds egalitarian, but the best-known circular model, holacracy, literally draws the pyramid as a ring and still keeps a strict hierarchy of rules inside it. Every picture encodes its own assumptions. That’s why I don’t believe the triangle should be replaced with one new correct picture — but that the whole idea of a single correct picture is a mistake. The researcher Dave Snowden has argued that different situations require a different way of making decisions: a clear problem is solved differently from a complex one. If that’s so, there is no one correct structure, only the one that fits the situation.

And yet — if some picture had to be chosen, its most important property would be this. Remember the observation of the first part: the father of the management pyramid, Robert Anthony, did not originally speak of layers but of kinds of work. He was right that there are different kinds of decisions — direction is a different matter from prioritising, and prioritising is a different matter from today’s doing. The problem isn’t in the three levels. The problem is that they were stacked on top of one another so that the top looks the most important.

What if the levels were set side by side rather than on top of each other? Keep the idea of different decisions, but give up the hint that the one on top is the most valuable? The operational level placed at the base — the one where the customer is actually met — is not the least but the one where the money is made. In a stacked picture it looks the lowest. In a side-by-side picture it would be one of three equally important tasks, not the bottom rung. This is not mere drawing: how the picture arranges the levels steers whose voice is listened to and whose work is valued. The picture teaches us to see the top as the most important, even though value is created at the edge. A different picture would teach otherwise.

All the levels are needed — but not in different people, and not stacked

This is where the series ends. Strategy, tactics, and operational work are still needed — why, how, and what. Not one of them disappears. But two things around them are changing. They no longer need different people: as we saw in the third part, one person with AI can stand on all three levels at once. And they don’t need a stack: no law of nature says that direction must be above doing. The pyramid was a brilliant answer to a world in which information moved at the speed of a horse and cart and had to be gathered into a few hands at the top. That world is ending, and the new form has not been handed to us ready-made.

That is the whole point of this series. We drew the triangle once, for a particular world. We can draw it differently, when the world is different. But what the new picture looks like is not decided by technology. It is decided by people — those who dare to give up the comfort of legibility, to share the intent, and to trust that the person at the edge knows how to decide. Technology changes what is possible. People decide what is done with the possibility. And so, after all of this, I dare hold onto what I started with: competitive advantage isn’t in the machines. It’s in people — in their judgement, their trust, and in what they choose to build, once the old picture is finally let go.

This was the final part of the “After the Triangle” series. Thank you for reading. Lenni Laukkanen helps leadership teams turn AI into a competitive advantage through people. Invite me to speak at your event or to spar with your leadership team — I reply within a working day.

Sources

Durable theory

Scott, James C. Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed. Yale University Press, 1998. (Legibility.)
Snowden, David J. & Boone, Mary E. “A Leader’s Framework for Decision Making.” Harvard Business Review, November 2007 (the Cynefin framework).
Robertson, Brian J. Holacracy. Henry Holt, 2015 (the circular model, critically).
Anthony, Robert N. Planning and Control Systems. Harvard, 1965 (link to part 1 of the series).

Recent evidence

Zhang, Letian. “The Changing Role of Managers.” American Journal of Sociology, 2023 (the long-run growth in managers’ share, ~9% in the 1980s → ~13% by the early 2000s).
Handel, Michael J. “Theories of lean management: an empirical evaluation.” Social Science Research 44, 2014 (rehiring and relayering).
Littler, Wiesner & Dunford. “The Dynamics of Delayering.” Journal of Management Studies 40(2), 2003.
Amazon (Andy Jassy’s memo, 16 Sept 2024; corporate cuts 2025); Bayer “Dynamic Shared Ownership” (reform in progress, 2024–).
Gallup (2025, subordinates per manager); Korn Ferry, Workforce 2025.
Gartner and Forrester (2025–2026): forecasts and surveys on the reversal of AI-driven cuts — presented as forecasts, not as realised facts.
Tieto (Tietoevry): Nordic AI Survey 2026 (623 respondents, FI/SE/NO, IT decision-makers; fieldwork Feb 2026). “Scaling is held back by the organization, not the technology”; a commercial actor’s survey.
Lee, Saerom. “The Myth of the Flat Start-up.” Strategic Management Journal 43(1), 2022 (flat structures break down as size grows).