Cities Tolerate Crazy People, Companies Don't
West's most quoted observation captures a structural distinction with precise mathematical grounding. Cities tolerate idiosyncratic behavior — the street preacher, the avant-garde gallery, the speculative startup — because their open, non-hierarchical network topology can absorb deviance without disruption to the whole. Companies cannot tolerate analogous deviance because their hierarchical topology requires every node to occupy a defined position in a delivery chain. Deviation from expected output at any node disrupts the chain. The tolerance is not a cultural virtue; it is a structural property of the network. Cities' indefinite persistence and companies' mortality both flow from this single fact. The AI transition offers companies, for the first time, the theoretical possibility of acquiring city-like tolerance for deviance — but only if they accept what cities accept: the mess, the unpredictability, the loss of the clean hierarchical control that makes organizations feel manageable.
In the AI Story
The aphorism appears in West's public lectures and in Scale, where it functions as a compressed expression of the deeper structural claim about network topology. It is characteristically direct — West does not hedge or qualify — and it exposes something that the business literature typically obscures.
The structural claim is precise. An open, non-hierarchical network has many redundant pathways. A node that behaves strangely — produces unexpected outputs, interacts with other nodes in unauthorized ways — can be absorbed by the network because no single pathway is critical. The street artist's deviance may be disturbing to neighbors, but it does not break the city. The city's functioning does not depend on the artist conforming to expected outputs.
A hierarchical network has designated pathways. Each node has a defined role, defined outputs, defined relationships to the nodes above and below. Deviance at any node disrupts the delivery chain. A middle manager who behaves unexpectedly — approves projects she was supposed to reject, channels information through unauthorized pathways, refuses to aggregate her team's output in the expected format — breaks the system she is part of.
The implication is that tolerance for deviance is not a moral choice organizations can make; it is a capacity their network topology either possesses or does not. A hierarchy cannot decide to 'tolerate crazy people'; its structural logic cannot absorb the disruption that crazy people generate. A city doesn't decide to tolerate; its topology absorbs whether it chooses to or not.
This is why cultural programs about 'embracing innovation' or 'tolerating failure' so often fail in mature companies. The programs attempt to change a topological property (tolerance) without changing the topology that determines it (hierarchy). The message is 'be more like a startup' delivered through structures that enforce branching-tree behavior.
AI's potential contribution is to enable genuine topological change. By dissolving the translation costs that make cross-domain collaboration expensive, AI allows organizations to actually function with meshed rather than branched topology — and therefore actually to absorb deviance rather than merely rhetorically tolerating it.
Origin
The observation appears in West's public lectures beginning around 2016 and is developed in Scale (2017). It draws on his collaborative work with Bettencourt on the structural differences between urban and corporate networks.
Key Ideas
Tolerance is structural, not cultural. Whether a network can absorb deviance depends on its topology, not on values or norms.
Open networks absorb deviance. Multiple redundant pathways mean no single deviation breaks the system.
Hierarchies cannot absorb deviance. Designated pathways mean every deviation disrupts the delivery chain.
Cultural programs fail without topology change. Telling hierarchies to 'embrace innovation' asks them to behave contrary to their structural logic.
AI enables genuine change. By dissolving translation costs, AI allows meshed topology that actually tolerates deviance rather than merely rhetorically endorsing it.
Appears in the Orange Pill Cycle
Further reading
- Geoffrey West, Scale (2017)
- Jane Jacobs, The Death and Life of Great American Cities (1961)
- Frederic Laloux, Reinventing Organizations (2014)