Who not to let in

2026-10-28

There's a line from Nietzsche I keep coming back to, aphorism 146 in Beyond Good and Evil:

"And if thou gaze long into an abyss, the abyss will also gaze into thee."

For months I stared into a problem of my own. The longer I looked, the more it looked back, and honestly it still weighs on me. For most of that time I blamed the wrong thing.

Where I've landed: who you surround yourself with is maybe the most important thing, in business and in friendship. With the wrong mix around you, you don't build much.

What growth does

I think every growing organization hits the same wall. With ten people you can hire only driven, really good people. Past a few hundred you can't. You need people faster than you can find them, so you compromise, and every compromise in hiring is a small compromise in culture.

After enough of them the culture becomes a patchwork and the original one fades. Driven people end up next to people who just put in the hours. A job being just a job is fine. The friction starts when some carry the load and others coast on it.

The clearest sign is what happens to honest feedback. Someone says "we are slow, let's speed up", and it lands as hostility, not as information about the work. The person who said it becomes the problem, so people stop saying it.

I think most of the value comes from a small, close group of people who push each other. At ten people that group is the whole company. At a few hundred it doesn't happen by default.

To be fair, "culture dilutes as you grow" is an old and contested idea. The usual backing is Dunbar's number, about 150 stable relationships, and a 2021 re-analysis put the range anywhere from 4 to 520 people. Plenty of big organizations keep a sharp culture. This is one person's observation, not a law.

Not the place

This is where I got it wrong. When something feels off, the easy suspect is the environment, and I went through all of it: the place, the setup, the routine. It wasn't the place. It was who was around.

By "the wrong person" I don't mean a type of person or anyone's worth. I mean behavior: cynicism as the default, coasting while others carry the load, punishing honest feedback, treating effort as naive. Most of us do some of this on a bad week, and people change. The question is only whether a pattern of behavior fits what a given team is trying to do.

One black speck

I saw a short video from some factory where a machine was pressing a big batch of white plasticine. They added one small black speck, and by the end the whole batch was grey.

I think teams and communities work the same way. One person with the wrong behavior can turn the whole group grey.

There's research pointing that way. In a 2006 review, Felps, Mitchell and Byington define the bad apple by behavior: withholding effort, persistent negativity and breaking interpersonal norms. Felps also had a hired actor join student teams as a jerk, a slacker or a pessimist. A 2008 This American Life episode says those teams did 30 to 40 percent worse, and the others started acting like him.

Honest caveats: I couldn't find that number in a published paper, so I treat it as a story, not a measurement. These were 45-minute lab tasks with students, so the effect in a real company is much less clear. And one team did fine, because one member kept asking questions and pulling everyone in. The speck doesn't always win.

The reverse question

People ask what to do all the time: what to build, how to work, where to live. Hardly anyone asks what not to do, where not to be, or who not to give their time to.

The idea isn't new. Charlie Munger liked to quote the mathematician Carl Jacobi, "Invert, always invert". Nassim Taleb calls it via negativa: improve things by removing what hurts, not by adding more.

I think this matters more now. Open spaces online are noisy, and it's getting hard to tell what's real. My guess, and it's only a guess, is that the places where good work actually happens will be smaller rooms you chose on purpose, online or in person. In a small room one speck changes everything, so who you let in is most of the design.

With people, I think the reverse question is the harder one. Who to let in sounds easy: anyone smart and nice. Who not to let in, said early, is where standards show. That isn't a closed door. It's clear expectations up front and honesty about who you spend your limited time on.

It applies to building things too, maybe more than anywhere else right now. Ask an agent for a feature and it will happily build you a castle: settings, options, edge cases, an admin page. Most of it is something almost nobody needs. It costs the agent nothing to build, so the scarce thing is deciding what not to build.

A founder wrote about this in an article on running his company with a fleet of agents, and I really liked it. His point is that when agents do most of the building, the job left for humans is taste. That's not only choosing what to build, it's also telling the agent no when the thing works but doesn't feel right. It includes deleting too. Every new model makes some of last year's workarounds pointless, so someone has to notice and remove them, even when a lot of work went into them. Building is cheap now, so deciding what's worth building is most of the job.

- another mind dump from a walk in a park in San Francisco