AI changes almost everything in your company, except the power game

AI vs. the Power Game in Organizations - Image created with Gemini (Nano Banana 2)

Every promise of a leaner organization powered by artificial intelligence comes with the same built-in line: fewer layers, less bureaucracy and less politics. The idea is that if the machine takes care of coordination, the fight for power loses its reason to exist.

In practice, that is exactly the part of the promise that doesn’t come true. AI touches almost everything a hierarchy does, but there is one thing it doesn’t lay a finger on: power.

What a hierarchy distributes besides tasks

An organizational structure does several things at once. It routes information, absorbs the unexpected and distributes scarce things: positions, priority, access, salary, the right to decide…

Notice that the number of positions at the top is always smaller than the number of people who want them. That is not an execution flaw that a better-managed company would fix. It is a feature of organizational design.

Fred Hirsch gave it a name: the positional good, whose value depends on others not having it. The executive suite only means something because most people never get there. Growing the company doesn’t dissolve that scarcity, because scarcity is what defines the good.

Distributing something scarce by relative position means creating a game. Economists Edward Lazear and Sherwin Rosen described this arrangement as a tournament: the reward doesn’t depend on your absolute performance, it depends on where you stand relative to others. Under certain conditions, the tournament even works as an employment contract, because the competition for position motivates effort.

Where the game goes bad

I’ve spent a lot of time in organizational design consulting, helping companies adopt more self-managed styles, and I always run into the same wall. Leadership wants self-management, talks enthusiastically about autonomy and then freezes when it’s time to give up what is scarce: the right to decide, access, the final say. Distributing tasks is easy. Distributing power is where the conversation gets complicated.

Now those same leaders are hearing that artificial intelligence will flatten the hierarchy for them, at almost no political cost. The promise is convenient precisely because it outsources to the machine the decision that leaders have avoided making.

So the game stays on, but it turns sour when the criteria for who wins become opaque.

The murkier the criteria for moving up, the more the competition shifts from performance to politics. If nobody really knows what gets a person promoted, all that’s left is trying to please whoever decides, managing your image upward and optimizing appearances instead of results.

This is where an insight from Pierre Bourdieu’s sociology comes in. Every arena of competition has rules, and the subtlest power is the power to make those rules seem natural, as if the current distribution were simply the fair outcome of things.

That’s where the discomfort of anyone who tries to point out the game comes from: it sounds like a sore loser complaining, because the yardstick presents itself as pure merit, but that yardstick is usually designed by those who already won the previous rounds.

Why AI barely touches this

Artificial intelligence is very good at two of the three functions of a hierarchy. It routes information with frightening competence and helps coordinate the unexpected in real time, absorbing variability.

The third function is of a different nature. Distributing status is not an information-processing problem; it is a decision about who gets what is scarce.

Maybe that’s why this function is the one that most resists any structural change. You swap the org chart, automate the workflow, cut layers, but the positional game is still there, intact.

It doesn’t disappear because the technology changed. It changes currency.

The new currencies AI creates

And AI creates new currencies for the same game, almost always without anyone having decided it.

In a company that adopts AI tools, disputes appear that didn’t exist before. Who gets access to the best models. Who defines the prompts and workflows the rest of the team will use. Who owns the dashboard that became the official source of truth about everyone’s performance. Whose signature is on the agent that makes decisions on its own. Each of these is a new scarce position, and each scarce position is a new round of the tournament.

There’s a detail that makes it worse. When the distribution criteria start living inside an AI system, they become more opaque, not less. Before, you could at least argue with your boss about why you were passed over. Now the answer is that the model scored it that way, and the model doesn’t explain the yardstick.

Remember the law Marilyn Strathern popularized: “when a measure becomes a target, it ceases to be a good measure”. An indicator that starts deciding who wins the tournament stops measuring the thing and starts measuring who best optimizes the indicator. AI industrializes that distortion, because it scales the indicator to the entire organization at once.

What to do about it in practice

The conclusion is not to abandon AI tools. It is to stop expecting them to solve a problem that was never technical.

Before automating anything that touches evaluation, promotion or recognition, it’s worth asking three things:

1. Who defined the criteria the system applies, and how can the people being evaluated get to know them?
2. When the system ranks someone, will that person be able to challenge it, and with whom?
3. Was the new scarce resource the tool creates (whether access, ownership of a workflow or the signature on an agent) distributed by a conscious decision, or did it simply end up in the hands of whoever has the most power?

None of these questions is about technology. They are all about who decides what is scarce and on what basis, which is the question hierarchy has always answered quietly and that AI only makes more urgent.

Food for thought

• In your company, is what gets a person promoted written down somewhere they can read it, or is it a criterion only those who already made it know?
• Of the AI tools you adopted in the last year, how many created a new scarce resource, and who ended up with it?
• And when a system recommends who to promote, will you be able to explain the criteria to whoever was passed over, or will the answer be that the model decided?