Clay Tech

"clay-works make things real"

translated from clazytech.com

Founding Traits Don't Decide Success, They Decide Variance

original Society & Ideas

Founding myths follow a fixed template

Tech company founding stories tend to follow the same template.

A shy, awkward young person who just loved technology. He builds a service on a whim, it grows at a terrifying speed, and before anyone notices, it has swallowed the world. That’s the template.

This template has one convenient property. Whatever the results, the founder’s own character never goes on trial. If the outcome is good, it becomes a story about talent. If the outcome is bad, it gets waved away as “that’s just how it happened” or “someone who came later did it.”

Let’s pose a naive question here. Extraordinary scale, extraordinary influence. To what extent are these extreme outcomes connected to traits the person already had from the start? Or is it just chance?

This time, I want to set aside judgments of good and evil and check this as a matter of structure.

Let me state the conclusion up front. Traits do matter. But they work differently than naive intuition suggests.

That intuition already has a name

First, let’s line up the theories on the side that says traits matter.

One is “imprinting theory” from management studies. The values a founder stamps into a company during its founding period — the choice of domain, the traits of the initial members — remain embedded in the organization for a long time (Marquis & Tilcsik, 2013). This directly rebuts the “someone who came later did it” explanation.

Another is the research on “CEO narcissism” (Chatterjee & Hambrick, 2007). It measures 111 CEOs in the computer/software industry using non-contact indicators: the size of their own photo in the annual report, press exposure, frequency of first-person singular pronouns in interviews, and the pay gap with the No. 2 executive. The more narcissistic the CEO, the flashier the strategy, the more acquisitions, and the more extreme the swings in performance.

But there’s a caveat. It just means big wins or big losses — averaged out, it’s no different from anyone else. This will matter later.

Then there’s research on the “Dark Triad” (narcissism, Machiavellianism, psychopathy). A growing body of reports finds that these traits strongly affect emergence as a leader — whether you get there — more than effectiveness as a leader — whether you do well once you’re there. And the higher up the hierarchy, the more concentrated these traits become.

In other words, these traits don’t determine whether the outcome is good or bad. They just carry a person closer to the seat at that scale.

Let’s place one proper noun as an example

Let me place one concrete example here: the Facebook founder, as a data point. He’s often cited as the archetype of that template — a shy tech lover who started something that grew explosively.

Let’s confirm just the facts. The ideal figure he publicly cites is not Caesar, but Augustus, Rome’s first emperor. In a 2018 New Yorker piece, he referenced Augustus as someone who established 200 years of peace through harsh methods, and expressed interest in that tradeoff. He spent his honeymoon in Rome. His second daughter is named August.

This isn’t material for condemnation. It’s simply an observation of consistency: a person with a strong interest in the design of dominion made designing dominion his work. The same consistency turns up just as easily in the biographies of widely admired giants. So the proper noun isn’t the protagonist here. It’s just one data point for checking the structure.

As a sufficient condition, it doesn’t hold

Now let’s lay out the limits of the argument that invokes traits.

The claim can be read two ways. One: “People who reached that scale mostly already had that trait from the start.” This is the claim of a necessary condition. The other: “If you have that trait, you’ll reach that scale.” This is the claim of a sufficient condition.

The theories below all show, together, that the latter does not hold. They don’t deny the former. In fact, they work in a way that leaves only the former standing.

There’s “the Romance of Leadership” from social psychology (Meindl, Ehrlich & Dukerich, 1985). People over-attribute organizational outcomes to the individual leader. This effect gets stronger especially for extreme outcomes — great success or great failure. Reading character into causes after the results are in makes “results → traits” look as though it proved “traits → results.” It’s a device that dresses up a necessary condition as a sufficient one.

Denrell & Liu (2012, PNAS) find that the more extreme the performance, the larger luck’s contribution. Their model shows that when evaluating outliers, we should pull our estimate further back toward the mean. Under intense competition, those with lower expected ability take excessive risks aiming for first place. In other words, luck decides who actually ends up at the top among the group that satisfies the trait. This is the stage where luck fills the gap left by the missing sufficient condition.

There’s “moral luck” from philosophy (Williams and Nagel). The very act of reading character backward into someone after the results are in is itself a suspect move. Someone with the same traits who ended up without results is spared having their character questioned. In the language of ethics, this amounts to saying “traits are not a sufficient condition.”

What remains invisible are the people who didn’t make it through

Let me also lay out the weakness of this argument. Its falsifiability is low.

The same fragments of trait should show up in equal measure among the countless people who ended up as nobody in particular. But observers see the extreme outcome first, and only afterward go looking for the trait. So “the overwhelming majority who had the same traits but never reached that scale” structurally never enters the field of view.

This invisible mass is itself the evidence — in the form of survivorship bias — that “traits are a necessary condition, but not a sufficient one.”

The same pattern shows up in sports. Meta-analyses find that deliberate practice explains about 18% of the variance in skill. Narrow it down to top athletes only, and it’s just 1% (Macnamara et al., 2016). Discipline from childhood is a necessary condition for entering the pool that is the national team. But within that pool, it explains almost nothing about who actually makes the team.

Separating the necessary condition from the sufficient condition

Here’s where the check lands.

There is a causal link between traits and outcomes. But that causal link operates as a necessary condition. Who gets close to the seat at that scale — this is where traits matter. An orientation toward control, an extraordinary obsession, a high appetite for risk. The population is strongly skewed toward these attributes. This part is fairly deterministic.

Beyond that, who actually wins within the group — this is where luck and excess risk matter. This part is thoroughly probabilistic.

And after taking office, power reproduces and amplifies the original traits. The imprint gets burned into the organization.

Here’s the crucial point. This structure works equally on the good and the bad. What the necessary-condition stage selects is “who can become enormous,” not “whether that enormity is good or bad for the world.” What traits can explain only gets you to the threshold of the outcome. The content of the outcome — whether it protects or damages human dignity — lies outside this stage.

So an argument built on traits can be used as a tool for conviction, but the exact same logic can just as easily be used for acquittal. A necessary condition does not choose the defendant.

I think the cleanest way to put it is this: traits are a necessary condition, not a sufficient one.

And a necessary condition is equally indifferent to the person who made it through and to the countless who didn’t. Adding a story on top of that is always the work of whoever is looking back from behind.

References

Original concept and direction: Kuzuryu. Written by AI.

Originally published in Japanese at https://clazytech.com/2026/07/1724/. Translated with LLM assistance and reviewed before publication.