Will AI Bring Back Central Planning? The 90-Year-Old Debate Getting a Reboot
In the 1930s, economists fought a war of ideas over a simple question: can a central authority ever know enough to run an economy better than a free market can? Friedrich Hayek's answer was no — not because central planners weren't smart enough, but because the knowledge needed to make good decisions doesn't exist in one place. It's scattered across millions of people, tied to specific moments and circumstances, and often impossible to write down. He called it the "knowledge problem," and for decades it was considered settled. AI is reopening the case.
The Argument for Centralization
Modern AI systems can do two things that used to be impossible. First, they can capture tacit, on-the-ground knowledge — the kind of expertise a store manager or machine operator carries in their head — by learning from transcripts, sensor data, and behavioral telemetry. Second, they can process volumes of information that would overwhelm any human team, relaxing what economist Herbert Simon called "bounded rationality." Put together, the case is straightforward: if a central hub can absorb local knowledge and process it faster than any individual node, the traditional advantage of decentralized decision-making starts to erode. We're already seeing early signs of this. Retail concentration has roughly tripled since the 1970s. The ten largest U.S. companies now make up close to 40% of total market value — double their share from just over a decade ago. Centralized category management, not the local store manager, increasingly decides what's on the shelf.
The Argument Against It
The counter-case is not about AI's processing power — it's about what data actually is. Critics argue that markets don't just record information; they generate it. A customer doesn't know their own preference until they're standing in front of a real choice with a real price attached. That moment of discovery can't be backfilled from historical data, no matter how much of it a model has ingested. AI is exceptional at prediction. It has no proven way to discover what doesn't yet exist — new tastes, new products, new ways of living that markets surface precisely because no one, including the people who'll eventually want them, knew to ask for them in advance. There's a second, more human wrinkle: incentives. Even a flawless AI planner is fed by people, and people shade the truth when honesty costs them something. A regional manager who reports softening demand accurately may simply be handed a harder target next quarter. That dynamic doesn't go away because the recipient of the report is an algorithm instead of a bureaucrat.
The Part That Should Concern Every Business Leader
Whichever side of the economic debate proves right, the centralizing pull of AI carries real-world stakes already in motion, well beyond the question of who runs supply chains better. The same data-aggregation capability driving corporate consolidation is the engine behind a sharper rise in state surveillance capacity. That's not an abstract geopolitical concern. It's a preview of the tradeoffs every organization adopting AI at scale will face: more efficient central coordination, paired with a much thinner margin for local autonomy and discretion.
A Structural Irony
Historian Yuval Noah Harari has pointed out a structural irony that AI now threatens to invert: distributed decision-making has long helped democracies out-process centralized bureaucracies. AI may be the first technology in a century to flip that equation — rewarding centralized control of data rather than punishing it.
“20th-century democracies generally outperformed dictatorships because distributed decision-making handled information better than centralized bureaucracies could.”
— Yuval Noah Harari, Historian
What This Means in Practice
For most organizations, the honest answer isn't "centralize everything" or "stay fully decentralized" — it's recognizing that AI changes the cost-benefit math of that decision, asset by asset, team by team.
Questions Worth Asking
Two questions are worth asking of every asset and every team:
- —Where does our advantage come from genuinely local, fast-moving judgment that doesn't compress well into data?
- —Where have we been protecting local discretion mostly out of habit, in places where a central system with better data could now coordinate more effectively?
Getting that distinction right — rather than defaulting to either extreme — is likely to be one of the more consequential strategic calls of the next decade.