By /

Authority Follows the Value — and Finds It at the Edge

I often wonder where in an organisation the money is actually made. Not in the top team’s meeting room, though that’s where direction is decided. Not in the middle managers’ meeting, though that’s where the work is divided up. It’s made where the organisation touches the customer — at the till, on the phone, on the site, at the other end of the email. That’s where the moment arises in which the customer either stays or leaves, recommends or warns. Everything else is support for that moment.

And yet the person deciding at that very moment usually has the least power. They see the situation most clearly, but they have to ask permission. This is the core of the pyramid’s logic: information is gathered up and the decision is made there, because the whole picture is at the top. But the previous parts showed that this logic rested on a single assumption — that information is expensive to move — and that AI is breaking exactly that assumption. So it’s time to ask the question that follows: if information no longer has to climb, is the decentralisation of authority ahead?

Formal authority is in the chart — real authority is in the knowledge

The core of the answer is in a 1997 article by two economists, Philippe Aghion and Jean Tirole. They distinguished two things we usually conflate: formal authority and real authority. Formal authority is the right to decide — what the org chart says. Real authority is actual control over the decision. And their sharp observation was that these two by no means always land on the same person.

Imagine a manager with the formal authority to approve a subordinate’s proposal. But they’re busy, with ten other things on their desk, and they haven’t time to get to grips with this one properly. What do they do? They approve what the subordinate puts forward — because they’re afraid of choosing the worse option if they start meddling. Aghion and Tirole called this rubber-stamping. The manager has the formal authority, but the real authority lies with whoever knows the matter.

This is more familiar than we’d like to admit. The board approves the chief executive’s proposal, because it can’t know the business as well as they do. The leadership team nods to the specialist’s recommendation, because no one else has gone as deep into the matter. The formal decision is made at the top, but the actual choice has already been made lower down — where the matter is really known. The decisive insight is that real authority follows knowledge. It drains to where the matter is known, regardless of what the chart says. Tirole, who won the Nobel in economics in 2014, did not write this about AI — but it’s hard to find a more apt description of what is happening now. When AI gives the person at the edge the same whole picture as the central management — the same data, the same analysis, the same situational view in seconds — real authority drains to the edge, whether the chart likes it or not.

Give intent, not orders

This is not a new idea. It was developed furthest in a place where the combination of slow information and high stakes was a matter of fate: the Prussian army of the nineteenth century. The chief of the general staff, Helmuth von Moltke the Elder, built a way of leading that in German is called Auftragstaktik. Its core is simple: the commander tells the subordinate what the objective is and why — but not how to do it. The execution is decided by whoever is in the situation. Moltke’s famous reasoning was that no plan survives, with certainty, the first encounter with the enemy’s main force. When the situation changes, the person on the spot must be able to act without waiting for a new order from far away.

The freshest and most convincing evidence of this comes from General Stanley McChrystal, who commanded US special forces in Iraq in the early 2000s. He faced an enemy that was networked, decentralised, and fast — and found that his own rigid chain of command was losing to it, despite being overwhelmingly superior in resources. The solution was twofold. First, radical transparency: information was shared so openly that everyone saw the whole picture. McChrystal called this shared consciousness. Then, empowered execution: decision-making power was pushed down to those who were in the situation. The result was staggering. The tempo of operations rose from a few a month to around three hundred — and the most important thing is this: the quality of decisions did not fall with the speed but rose. They had expected to get a rougher solution quickly, but got a better solution quickly. The leader’s role changed from chess player to gardener — no longer the one making the moves, but the one creating the conditions.

The same principle has been translated into the language of business. Netflix’s founder Reed Hastings distilled it into two words: context, not control. The leader’s job is to give people the information and the purpose on which they make good decisions themselves — not to supervise every decision. Netflix’s internal principle is that an organisation should be both tightly aligned and loosely coupled: a strong shared direction, but independent execution. It’s the same formula as Moltke’s and McChrystal’s — shared intent and decentralised execution — just without the uniform.

The human’s role doesn’t shrink — it changes

Notice what this does to the human’s role. In the pyramid, the person in the middle was above all a relay: they gathered information from below, condensed it upward, and carried the orders back down. AI takes over exactly this relay work. But it doesn’t make the human unnecessary — it frees them for what the machine can’t do: judgement, carrying responsibility, the real encounter with the customer.

The best organisations have known this long before AI. In Ritz-Carlton hotels, every employee may spend up to two thousand dollars to solve a customer’s problem without a manager’s permission. The sum is rarely used in full — its value isn’t in the money but in the trust, which tells the employee: you get to decide. On Toyota’s factory floor, any line worker may stop the entire production if they spot a fault. In both cases, power has been taken to where the eyes are. AI makes this kind of thing possible far more widely, because it gives the person at the edge the whole picture that used to be only at the top.

For a leader this means three practical changes that sound easy but are hard. First: give intent, not orders — say what is being aimed at and why, and leave the how to whoever is in the situation. Second: share the whole picture, not a filtered version — the edge can make good decisions only if it sees what you see, and AI finally makes this sharing cheap. Third, and hardest: tolerate the edge making a decision differently from how you’d have made it. If every decision has to be made your way, you haven’t decentralised power — you’ve outsourced the typing. Empowerment without the right to disagree is no empowerment at all.

But this doesn’t happen by itself

Here I have to stop, so the essay doesn’t slip into wishful thinking. Taking decision-making power to the edge can go badly wrong — and it does, if it’s done wrong.

Jo Freeman wrote, back in the 1970s, an essay every leader who dreams of flattening should read. Her observation was that there is no such thing as a structureless group. When the formal structure is dismantled, power doesn’t disappear — it becomes invisible and shifts to informal, unaccountable cliques that no one chose and no one can replace. A visible hierarchy is often better than a hidden one. When the online retailer Zappos switched a few years ago to a radically hierarchy-free model, almost a fifth of the staff left, and the organisation never quite recovered. Decentralising decision-making power without shared intent and clear responsibility is not freedom but chaos.

It’s also worth remembering that the transfer of power isn’t a mere structural switch you can flip. When American carmakers copied Toyota’s famous cord for stopping the line, the result was often empty: workers didn’t dare pull it, because they feared the consequences. The same tool, a different culture, the opposite outcome. Empowerment works only if there is trust and shared purpose behind it — otherwise it’s just a new button nobody presses.

And there is a third pitfall, one that has only recently been given a name. TEK, the Finnish union of academic engineers, named an “AI elite” in its recent AI report: the group that plans and steers the adoption of AI in their workplaces. The group is narrow and one-sided: 28% of respondents, and of those nearly four in five men, typically middle-aged and in technical roles. The effects spread into everyone’s work, but the power over adoption concentrates in the hands of a few. Nobody decided it should be so. An elite like this is not created by a decision but forms by itself: out of who got excited first, who already held the technical turf, and who was given working time to learn. And because it forms by itself, it also inherits the past — who held the power back when information was expensive.

I see the same phenomenon in my everyday work, without the statistics. In many organisations AI is turning into internal politics: IT feels that AI belongs on its turf and that it alone should decide where and on what terms the machine may be used. And IT is not wrong that the risks must be managed — but gatekeeper is a different role from owner. When AI becomes a question of budgets and prestige, the most important question goes unasked: what does the customer get out of this? If the power to decide on adoption stays far from where the work is done, the workflows are never redesigned — a tool is simply glued on top of the old ones, and the benefit never arrives. The best expert on a process is almost always the person who runs it: they know where the work actually snags and which step is mere tradition. That is why I’d argue the decisive thing is which game the organisation plays. The inside game divides power and budgets; the outside game creates value for the customer. Only one of them pays the salaries. A power structure that formed by itself will not, however, dismantle itself — challenging it demands conscious work from leadership, and keeping the eyes on the customer at the very moment when power’s natural tendency is to concentrate just as customer value would demand it to spread.

And there is one more, bigger caveat. The same AI that can give the edge power can just as easily take it away. If the machine gives leadership a real-time view of everything, it can tempt them to supervise and centralise more than ever. And the infrastructure the whole of AI runs on is held by a few: three large companies already control about two-thirds of the enterprise cloud-infrastructure market. So the edge can just as well strengthen as shrink — what decides is who controls the models and the data, and what leadership chooses to do.

Technology makes it possible — people make it real

This is exactly why this series is called what it is. AI makes it possible for decision-making power to drain to where value is created. But it doesn’t do it automatically. Whether power drains to the edge or piles up ever more tightly at the top is not a technological question but a choice of leadership. And that choice depends entirely on people — on whether leadership dares to give intent and to trust, and on whether the person at the edge carries the responsibility that comes with the power. Technology gets you to this point. People decide which way you go from here.

At this point many a leader nods. The logic feels clear: take power to where the knowledge and the value are. But if it’s this clear, why do so few organisations actually do it? Why does the pyramid stay standing even as its rationale crumbles? The answer is that the pyramid is not a mere chart. It is the way power, pay, career, and esteem are divided — and that is not dismantled by a single insight. That’s the subject of the series’ final part.

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.

Next part: The Triangle Isn’t a Diagram — It’s How Power Is Divided: why the change is hard, and what replaces the triangle.

Sources

Durable theory

  • Aghion, Philippe & Tirole, Jean. “Formal and Real Authority in Organizations.” Journal of Political Economy 105(1), 1997, pp. 1–29. (Tirole, Nobel in economics 2014.)
  • Hayek, Friedrich A. “The Use of Knowledge in Society.” The American Economic Review 35(4), 1945, pp. 519–530.
  • Auftragstaktik / mission command: Helmuth von Moltke the Elder, Über Strategie (1871) and others; modern doctrine US Army ADP 6-0, Mission Command (2019).
  • Freeman, Jo. “The Tyranny of Structurelessness” (1970–73; Berkeley Journal of Sociology 17).

Recent evidence

1 thought on “After the Triangle — Part 4/5, the decentralisation of authority”

  1. Pingback: After the Triangle — Part 3/5, AI and organisational structure - Lenni Laukkanen

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top