Twenty founders who all agree aren’t making twenty bets. They’re making one bet twenty times.
Picture a room of startup founders. Each wants what the most successful founder in the room has, and that founder wants what the most successful founder outside the room has. So everyone is building similar companies, tracking similar metrics, hiring similar people and describing success in similar words. Each of them is being sensible. Copying the people who are winning is one of the fastest ways to learn what works, and the consensus in the room looks like accumulated wisdom. Yet when the paradigm shifts and the bust comes, rooms like this one tend to go down together, and hard. If every founder was acting reasonably, where did the fragility come from?
Desire runs downhill
René Girard noticed that desire flows like water: downhill toward wherever desire is already concentrated, along the gradient of whatever the most admired people want. You don’t choose what you want by looking at the object. You choose by looking at who already wants it, and then you call the output your own.
This isn’t a character flaw. It is an information-processing strategy, and social proof is often genuinely informative. In a stable environment, where the people at the top of the gradient are right about something real, following them is efficient.
The trouble is what efficient gradient-following does to a group. When enough people model enough other people’s desire, the crowd’s desires synchronize. Independent actors become correlated actors, and what looked like decentralized judgment is really one large, coordinated bet. Coordinated bets are non-ergodic: the average looks fine right up to the moment everyone hits ruin together.
How the crowd loses its brakes
A crowd corrects its errors through the people who don’t follow the gradient. Some founder ignores the fashionable market, builds the unglamorous company, holds the unpopular view. In stable periods this is costly, since she underperforms while the gradient keeps rising. But she is also how the system updates when the gradient reverses, because she is already standing somewhere else.
So the stable period quietly removes the brakes. In markets, the fundamental analysts underperform during a bubble and face redemptions, while the momentum followers outperform and attract capital. In ideological groups the dissenters are ostracized, and in status competitions they are simply exiled. By the top of the gradient, the crowd has eliminated most of the people who could have seen the reversal coming. The homogenization that produced the stability is the same thing that makes the collapse catastrophic, which is the Minsky cycle running on mimetic fuel. The crash isn’t proportional to whatever triggers it. It is proportional to the accumulated synchrony: how thoroughly the crowd has removed the independent actors who could have absorbed the shock.
The grandmother
Now put a grandmother in the room, one whose sense of what matters was formed in another country and another era. She has never heard of Y Combinator. She wants security, community and durable things. She is immune to the founders’ gradient, not because she thinks more independently than they do, but because she is following a different gradient entirely, set under different conditions decades ago. Her divergence looks like naivety. It is actually temporal diversification, and she may notice early warning signs the founders have trained themselves not to see.
That is what the multigenerational family was quietly doing. A grandmother whose anchors were formed in the 1930s, a father formed in the 1970s and a child formed now cannot fully synchronize with the current gradient, because the older calibrations act as a brake. The family was not a moral institution. It was a mimetic hedge. Atomization removes that ballast. The isolated individual, with no competing anchors from other eras, is exposed to whoever currently sits at the top of the visible hierarchy, and is maximally vulnerable when the gradient inverts. That isn’t because he is foolish. The institution that kept his desires diversified has dissolved.
Why you can’t just think for yourself
The obvious prescription is “think for yourself,” and it fails for a reason Girard pushed hard on: he argued that autonomous desire isn’t really available. Gradient-following operates below deliberate choice and feels like independent judgment from the inside. The desire arrives first and the conscious mind supplies the reasons. A founder who resolves to be independent usually ends up performing independence while still facing the same way as everyone else.
So the realistic aim is diversity of gradients rather than freedom from them: relationships with people whose anchors come from different times and places, serious engagement with thinkers formed in other eras, commitments that are anti-correlated with the current fashion. The deeper point is that mimetic desire is not a failure of individual rationality but a structural feature of how desire works — and its danger is not that individuals follow blindly but that efficient gradient-following eliminates the systemic diversity that makes error-correction possible. The founders aren’t irrational. They are individually rational actors whose rationality, in aggregate, has produced maximum fragility.
The counterweight is real: anti-correlation is expensive, and most of the time the grandmother’s caution simply costs money. She will be wrong about many booms. But her value lies in being out of step when everyone else is in step, not in being right on average. The grandmother’s calibrated anxiety about gradients that look like 1932 is not irrationality. It is anti-correlation. It is the only thing standing between a fully synchronized crowd and the absorbing state that crowds, moving together, reliably produce.
Which leaves an uncomfortable question for anyone in the founders’ room. Whose anchors, from what era, still disagree with you, and how long has it been since you let them?