How Doomed are Large Organizations?

by Zvi Don't Worry About the Vase8 min read21st Jan 202041 comments


We now take the model from the previous post, and ask the questions over the next several posts. This first answer post asks these questions:

  1. Are these dynamics the inevitable results of large organizations?
  2. How can we forestall these dynamics within an organization?
  3. To what extent should we avoid creating large organizations?
  4. Has this dynamic ever been different in the past in other times and places?

These are the best answers I was able to come up with. Some of this is reiteration of previous observations and prescriptions. Some of it is new.

There are some bold claims in these answer posts, which I lack the space and time to defend in detail or provide citations for properly, with which I am confident many readers will disagree. I am fine with that. I do not intend to defend them further unless I see an opportunity in doing so.

I would love to be missing much better strategies for making organizations less doomed – if you have ideas please please please share them in the comments and/or elsewhere.

Are these dynamics the inevitable result of large organizations? 

These dynamics are the default result of large organizations. There is continuous pressure over time pushing towards such outcomes.

The larger the organization, the longer it exists, and the more such outcomes have already happened, both there and elsewhere, the greater the pressure towards such outcomes. 

Once such dynamics take hold, reversing them within an organization is extremely difficult.

Non-locally within a civilization, one can allow new organizations to periodically take the place of old ones to reset the damage.

Locally within a sufficiently large organization and over a sufficiently long time horizon, this makes these dynamics inevitable. The speed at which this occurs still varies greatly, and depends on choices made.

How can we forestall these dynamics within an organization?

These dynamics can be forestalled somewhat through a strong organizational culture that devotes substantial head space and resources to keeping the wrong people and behaviors out. This requires a leader who believes in this and in making it a top priority. Usually this person is a founder. Losing the founder is often the trigger for a rapid ramp up in maze level.

Keeping maze levels in check means continuously sacrificing substantial head space, resources, ability to scale and short-term effectiveness to this cause. This holds both for the organization overall and the leader personally. 

Head space is sacrificed three ways: You have less people, you devote some of those people to the maze-fighting process, and the process takes up space in everyone’s head.

Central to this is to ruthlessly enforce an organizational culture with zero tolerance for maze behaviors.

Doing anything with an intent to deceive, or an intent to game your metrics at the expense of object level results, needs to be an automatic “you’re fired.”

Some amount of politics is a human universal, but it needs to be strongly discouraged. Similarly, some amount of putting in extra effort at crucial times is necessary, but strong patterns of guarding people’s non-work lives from work, both in terms of time and other influences, are also strongly necessary.

Workers and managers need to have as much effective skin in the game as you can muster.

One must hire carefully, with a keen eye to the motivations and instincts of applicants, and a long period of teaching them the new cultural norms. This means at least growing slowly, so new people can be properly incorporated.

You also want a relatively flat hierarchy, to the extent possible.

There will always be bosses when crunch time comes. Someone is always in charge. Don’t let anyone tell you different. But the less this is felt in ordinary interactions, and thus the more technically direct reports each boss can have and still be effective, and thus the less levels of hierarchy you need for a given number of people, the better off you’ll be. 

You can run things in these ways. I have seen it. It helps. A lot.

Another approach is to lower the outside maze level. Doing so by changing society at large is exceedingly hard. Doing so by associating with outside organizations with lower maze levels, and going into industries and problems with lower maze levels, seems more realistic. If you want to ‘disrupt’ an area that is suffering from maze dysfunction, it makes sense to bypass the existing systems entirely. Thus, move fast, break things.

One can think of all these tactics as taking the questions one uses to identify or predict a maze, and trying to engineer the answers you want. That is a fine intuitive place to start.

However, if Goodhart’s Law alarm bells did not go off in your head when you read that last paragraph, you do not appreciate how dangerous Goodhart Traps are.

The Goodhart Trap 

The fatal flaw is that no matter what you target when distributing rewards and punishments and cultural approval, it has to be something. If you spell it out, and a sufficiently large organization has little choice but to spell it out, you inevitably replace one type of Goodharting with another. One type of deception becomes another.

One universal is that in order to maintain a unique culture, you must filter for those that happily embrace that culture. That means you are now testing everyone constantly, no matter how explicit you avoid making this, on whether they happily embrace the company and its culture. People therefore pretend to embrace the culture and pretend to be constantly happy. Even if they do embrace the culture and are happy, they still additionally will put on a show of doing so.

If you punish deception you get people pretending not to deceive. If you punish pretending, you get people who pretend to not be the type of people who would pretend. People Goodhart on not appearing to Goodhart.

Which is a much more interesting level to play on, and usually far less destructive. If you do a good enough job picking your Goodhart targets, this beats the alternatives by a lot.

Still, you eventually end up in a version of the same place. Deception is deception. Pretending is pretending. Fraud is fraud. The soul still dies. Simulacrum levels still slowly rise. 

Either you strongly enforce a culture, and slowly get that result, or you don’t. If you don’t and are big enough, you quickly get a maze. If you do and/or are smaller, depending on your skill level and dedication to the task, you slowly get a maze.

Hiring well is better than enforcing or training later, since once people are in they can then be themselves. Also because enforcement of culture is, as pointed out above, toxic even if you mean to enforce a non-toxic ideal. But relying on employee selection puts a huge premium on not making hiring mistakes. Even one bad hire in the wrong place can be fatal. Especially if they then are in a position to bring others with them. You need to defend your hiring process especially strongly from these same corruptions.

My guess is that once an organization grows beyond about Dunbar’s number, defending your culture becomes a losing battle even under the best of circumstances. Enforcing the culture will fail outright in the medium term, unless the culture outside the organization is supporting you. 

If you are too big, every known strategy is only a holding action. There is no permanent solution.

To what extent should we avoid creating large organizations? 

Quite a lot. These effects are a really big deal. Organizations get less effective, more toxic and corrupt as places to work and interact with, and add more toxicity and corruption to society. 

Every level of hierarchy enhances this effect. The first five, dramatically so. Think hard before being or having a boss. Think harder before letting someone’s boss report to a boss. Think even harder than that before adding a fourth or fifth level of hierarchy. 

That does not mean such things can be fully avoided. The advantages of large organizations with many degrees of hierarchy are also a really big deal. We cannot avoid them entirely. 

We must treat creating additional managerial levels as having very high costs. This is not an action to be taken lightly. Wherever possible, create distinct organizations and allow them to interact. Even better, allow people to interact as individuals. 

This adds friction and transaction costs. It makes many forms of coordination harder. Sometimes it simply cannot be done if you want to do the thing you’d like to do. 

This is increasingly the case, largely as a result of enemy action. Some of this is technology and our problems being legitimately more complex. Most of it is regulatory frameworks and maze-supporting social norms that require massive costs, including massive fixed costs, be paid as part of doing anything at all. This is a key way mazes expropriate resources and reward other mazes while punishing non-mazes.

I often observe people who are stuck working in mazes who would much prefer to be self-employed or to exit their current job or location, but who are unable to do so because the legal deck is increasingly stacked against that.

Even if the work itself is permitted, health insurance issues alone force many into working for the man. 

When one has a successful small organization, the natural instinct is to scale it up and become a larger organization.

Resist this urge whenever possible. There is nothing wrong with being good at what you do at the scale you are good at doing it. Set an example others can emulate. Let others do other things, be other places. Any profits from that enterprise can be returned to investors and/or paid to employees, and used to live life or create or invest in other projects, or to help others.

One need not point to explicit quantified dangers to do this. Arguments that one cannot legitimately choose to  ‘leave money on the table’ or otherwise not maximize, are maximalist arguments for some utility function that does not properly capture human value and is subject to Goodhart’s Law, and against the legitimacy of slack

The fear that if you don’t grow, you’ll get ‘beaten’ by those that do, as in Raymond’s kingdoms? Overblown. Also asking the wrong question. So what if someone else is bigger or more superficially successful? So what if you do not build a giant thing that lasts? Everything ends. That is not, by default, what matters. A larger company is often not better than several smaller companies. A larger club is often not better than several smaller clubs. A larger state is often not better or longer lasting than several smaller ones. Have something good and positive, for as long as it is viable and makes sense, rather than transforming into something likely to be bad. 

People like to build empires. Those with power usually want more power. That does not make more power a good idea. It is only a good idea where it is instrumentally useful.   

In some places, competition really is winner-take-all and/or regulations and conditions too heavily favor the large over the small. One must grow to survive. Once again, we should be suspicious that this dynamic has been engineered rather than being inherent in the underlying problem space. 

Especially in those cases, this leads back to the question of how we can grow larger and keep these dynamics in check. 

Has this dynamic ever been different in the past in other places and times?

These dynamics seem to me to be getting increasingly worse, which implies they have been better in the past.

Recent developments indicate an increasing simulacrum level, an increasing reluctance to allow older institutions to be replaced by newer ones, and an increasing reliance on cronyism and corruption that props up failure, allowing mazes to survive past when they are no longer able to fulfill their original functions.

Those in the political and academic systems, on all sides, increasingly openly advocate against the very concept of objective truth, or that people should tell it, or are blameworthy for not doing so. Our president’s supporters admit and admire that he is a corrupt liar, claiming that his honesty about his corruption and lying, and his admiration for others who are corrupt, who lie and who bully, is refreshing, because they are distinct from the corrupt, the liars and the bullies who are more locally relevant to their lives. Discourse is increasingly fraught and difficult. When someone wants to engage in discourse, I frequently now observe them spending much of their time pointing out how difficult it is to engage in discourse (and I am not claiming myself as an exception here), as opposed to what such people used to do instead, which was engage in discourse.

We are increasingly paralyzed and unable to do things across a wide variety of potential human activities. 

Expropriation by existing mazes and systems eats increasing shares of everything, especially in education, health care and housing.

I don’t have time for a full takedown here, but: Claims to the contrary, such as those recently made by Alex Tabbrok in Why Are The Prices So Damn High?, are statistical artifacts that defy the evidence of one’s eyes. They are the product of Moloch’s Army. When I have insurance and am asked with no warning to pay $850 for literally five minutes of a doctor’s time, after being kept waiting for an hour (and everyone I ask about this says just refuse to pay it)? When sending my child to a kindergarten costs the majority of a skilled educator’s salary? When you look at rents?

Don’t tell me the problem is labor costs due to increasing demand for real services.

Just. Don’t.

Some technological innovations remain permitted for now, and many of the organizations exploiting this are relatively new and reliant on object-level work, and thus less maze-like for now, but this is sufficiently narrow that we call the result “the tech industry.” We see rapid progress in the few places where innovation and actual work is permitted to those without mazes and connections, and where there is sufficient motivation for work, either intrinsic or monetary. 

The tech industry also exhibits some very maze-like behaviors of its own, but it takes a different form. I am unlikely to be the best person to tackle those details, as others have better direct experience, and I will not attempt to tackle them here and now.

We see very little everywhere else. Increasingly we live in an amalgamated giant maze, and the maze is paralyzing us and taking away our ability to think or talk while robbing us blind. Mazes are increasingly in direct position to censor, deplatform or punish us, even if we do not work for them.

The idea of positive-sum, object-level interactions being someone’s primary source of income is increasingly seen as illegitimate, and risky and irresponsible, in contrast to working for a maze. People instinctively think there’s something shady or rebellious about that entire enterprise of having an actual enterprise. A proper person seeks rent, plays the game starting in childhood, sends the right signals and finds ways to game the system. They increase their appeal to mazes by making themselves as dependent on them and their income and legitimacy streams, and as vulnerable to their blackmail, as possible. 

The best way to see that positive-sum games are a thing is to notice that the sum changes. If everything is zero-sum, the sum would always be zero.

The best way to see that these dynamics used to be much less severe, at least in many times and places, is that those times and places looked and felt different, and got us here without collapsing. Moral Mazes was written before I was born, but the spread of these dynamics is clear as day within my lifetime, and yours as well.

Did some times and places, including our recent past, have it less bad than us in these ways? I see this as almost certainly true, but I am uncertain of the magnitude of this effect due to not having good enough models of the past.

Did some times and places have it worse than we do now? Very possible. But they’re not around anymore. Which is how it works.

The next section will ask why it was different in the past, what the causes are in general, and whether we can duplicate past conditions in good ways.