What makes an organization fit for the AI era? While reading an MIT Sloan article about Andrew McAfee’s argument, one answer stayed with me: Successful companies, he argues, work iteratively, have flat structures, and are science-driven. That sounds plausible. But is a flat hierarchy really the decisive variable—or merely a visible feature of organizations that are particularly good at learning quickly?
Is “flat” the cause—or just a proxy?
McAfee points to SpaceX and Netflix, among others. Both transformed established industries through rapid experimentation and unusual organizational models. The MIT IDE Annual Conference report turns this into a clear message: Above all, these companies were willing to learn and adapt rapidly.
The examples illustrate a possible mechanism, but they come from the usual lineup of “showcase companies.” They tell us neither how many similarly organized companies failed nor how successfully more hierarchical organizations can innovate.
I am not a big fan of hierarchies—Agile is dead, long live Agile! ;) But perhaps “flatness” is only a proxy: for short information paths, local decision authority, rapid access to feedback, and low political costs when an earlier decision has to be revised. The more interesting question is then not how many layers an organization has, but:
How quickly can new evidence change what the organization actually does?
Running many experiments does not mean learning much
An organization can launch experiments every week and still learn slowly. It may measure the wrong thing. Negative results may be explained away. A team may be allowed to test, but nobody may be authorized to change the process, product, or policy afterwards. Or an insight remains local and never reaches the rest of the organization. Anyone who has never experienced anything like this, please send me an email.
Experimentation rate and learning rate are therefore not the same. For me, organizational learning rate covers the whole path from noticing a possible change through hypothesis, test, and interpretation to a revised decision and a new way of working in practice.
The slowest or least reliable step constrains the entire loop. Ten rapid experiments do not help much if their results then wait three months for a decision committee. Or if it was never clear what the organization was actually trying to measure. Or achieve. Or, or, or.
Which friction slows learning—and which even protects it?
Speed alone is not a useful goal either. An experiment on a website or an e-commerce platform can often be deployed, measured, and reversed quickly. In medicine, banking, aviation, or critical infrastructure, a failed experiment can endanger people, trigger regulatory consequences, or be difficult to reverse. As we have always said in the agile world: You should not move faster than your market allows—there is no point.
Reviews, verification, compliance, and independent approvals are therefore not automatically evil bureaucracy. They can be productive friction: friction that slows the process but improves the quality of the evidence or the decision derived from it.
A fast organization that scales bad decisions even faster is not learning quickly. It is simply producing errors at a higher throughput.
Isn’t this exactly what organizational agility is about?
Much of what McAfee calls for is, to me, the core of organizational agility: sensing change, learning through short feedback loops, and adapting how the organization acts. I do not mean adherence to a particular framework (you are not agile just because you have “implemented” SAFe), but responsiveness across the whole system. AI did not invent this requirement; it makes it even more urgent.
My experience as an Agile Coach also showed me the problem: The more traditionally an organization is designed, the harder it is to develop this kind of responsiveness. A team can be as agile as it likes: If budgets, decision rights, departmental boundaries, and approvals remain optimized for long-term stability and top-down control, organizational agility will never, ever emerge from that.
That, I fear, is also part of Europe’s disadvantage: In more traditional structures, the path towards rapid learning cycles is longer. It does not explain Europe’s position on its own, however. The Draghi report on European competitiveness describes a broader combination of productivity, demographics, energy costs, investment needs, and global competition. To me, slow organizational learning is certainly part of the explanation—but not the whole explanation.
I still find McAfee’s basic intuition compelling: When technological capabilities change faster, organizational assumptions also expire faster. It simply does not follow that the flattest or fastest organization will win.
The advantage may belong to the organization that can learn fastest without destroying the controls that make its learning reliable!