On Schelling Points in Organizations — https://medium.com/@komorama/on-schelling-points-in-organiza...

You can download it in various forms as well, such as txt. Won't fix the tone, but at least it's scannable.

At a sufficiently small company, you can just trust people will generally do the right thing. I ask X to accomplish some compliance goal, I just trust that they'll do it right.

As the company scales, you actually know a smaller and smaller portion of the people you work with. You no longer just trust that they'll do the right thing, because you don't know them. You fix the lack of trust by adding processes that ensure the "right thing" happens, but each of those processes is a friction point. Eventually you have so many of those processes that they become the majority of the effort for a project.

That gets exacerbated by the usual turnover. People pissed off by all the processes they have to comply with will leave, and people who love having their own fiefdom will stay and double down.

That all ends in the evergreen stupidity of "spend $1,800 to have 12 people who make $150/hr argue about whether this app really needs a $5/month Aurora database".

That's a personal pet peeve, but is emblematic of the issue imo. Trust has broken down so severely that the company is willing to spend more than the lifetime cost of compute for the app to vet whether it's "necessary".

A similar vein is the "platform team" where all they do is take an open source project and wrap it in a custom DSL, so you get all the complexity and none of the searchability of the original project, plus the features are almost always a subset. I can't tell you how many times I've looked for solutions to a problem, found a snippet that will fix it, and then had to reverse engineer how to make a stupid DSL output that config.

1: https://en.wikipedia.org/wiki/Dunbar%27s_number

These are two public previews that we are building off of for the journal:

1. https://paragraph.com/@hq.spengrah/an-actionable-model-of-tr...

2. https://x.com/hatsprotocol/status/1986458735095517526

I think the ideas are solid but it's the C-suite that needs to deal with this and rarely cares. In other words, it's about organizational culture.

I don't think the author is really super familiar with some of those other organizations he mentions so he's reflecting how he thinks they work rather than how they work. That said the observations are still pretty good.

> You need to delegate execution authority to small teams who are aligned to a shared goal

It's been a while since I read the slides but this is basically what the author is advocating for and what doesn't tend to happen in dysfunctional organizations.

> Complexity management in organizations follows the same patterns as in code. Single responsibility, don’t repeat yourself too much, encapsulation, etc.

There are a lot of parallels. But organizations have people so that's an extra factor. You can't just "refactor" them without thinking about people's egos, objectives and such. So a lot of thought is needed into how to build effective organizations and sadly not enough thought is given is most companies.

Funnily enough it’s extremely relevant to agents/subagents.

Many tools are available today to alter data representation. And frontier LLMs have high fidelity. You don’t have to wait for someone to help you. Claude helps those who help themselves.

It is not true today, and as you mentioned "fundamentally incompatible with a next-quarter profit driven.."

Agreed the observations are interesting, and that’s what is being advocated for, I just don’t think top-down vs bottom-up is the axis that impacts that as much as disturbed vs… I said “centralized” but probably “localized” is a better term.

So its becoming more obvious that AI can actually help big corporations organize their command-structure with the help of AI to produce more profits. The boards should take notice. Does it really help produce more value if most of it goes to the CEO?

Usually they are structured in ways that can be gamed and the structure is defined by those that benefit most from them being gamed.

But then again, this is a view from the underbelly of corporate life and I may be missing something.

1. Visit the embedded google slides URL https://docs.google.com/presentation/d/1XMCU2vxtxxqqQ2NY567n...

2. Click File

3. Scroll down to 'Download' and point your cursor at it and wait for the submenu to open

4. Select PDF Document

5. Upload the PDF Document to your chat agent of choice

Not everyone is able to type the following and get results, but I'm built different:

> Summarize this for me: https://komoroske.com/slime-mold/

> It’s a google slides thing but i don’t want to click through

Here. Scroll bar included so you can see it's literally the only prompt https://imgur.com/a/Qela4qt

It conveniently created a .txt file and also expanded the text into a full article.

I then asked: "create another version that is just the text from each slide. nothing more"

Result:

----------

Many successful organizations pride themselves on moving quickly to tackle problems.

But over time, they start to feel way, way slower. Accomplishing previously simple things feels like it takes forever.

You likely aren’t imagining it, by the way! This happens in every successful organization over time.

Clearly someone is doing something wrong. Perhaps there’s some villain who is messing everything up. If we figure out whose fault it is, we should get them to knock it off!

Unfortunately, there is no villain. It’s no one’s fault… and not something any one person can fix.

There’s a hidden force that makes life in successful organizations increasingly miserable. It’s invisible, so we either don’t see it... or we at least pretend not to. But the force is inescapable and affects everything that happens within an organization.

The force is the coordination headwind.

This force emerges on its own. No one created it. No one is doing it on purpose.

Things in your organization may have gotten orders of magnitude harder, but everyone pretends they’re still as easy as they used to be.

To understand what this force is and why it arises, we’re going to have to apply some unfamiliar tools. Strap in, this is going to get weird!

Slime Molds and Militaries

Bottom-Up Top-Down Organizations come in many different types, and different types work differently. One important dimension is bottom-up vs top-down.

Bottom-Up Top-Down The stereotypical example of top-down is a military. The hierarchy is rigid and formal. Subordinates are expected to do precisely what their superior tells them to. (Of course, even real life militaries, especially nowadays, don’t work like this in practice.)

Bottom-Up Top-Down The stereotypical example of a bottom-up organization is a slime mold. A number of different independent actors make individual decisions, leading to complex emergent behavior in the colony.

Side note: slime molds are amazing! Here’s a slime mold deriving the Tokyo rail network. These types of systems are called complex adaptive systems, and they can model many things, like ecological ecosystems, software ecosystems, economies, politics, and more!

Bottom-Up Top-Down Unpredictable Predictable Uncontrollable Controllable Individual autonomy Cog in the machine Fluid Structured Messy Efficient Resilient Fragile Explore Exploit The two different extremes have very different properties. Some good, some... not so good.

Bottom-Up Top-Down There’s a spectrum between these two extremes, with most organizations falling somewhere in the middle.

Bottom-Up Top-Down A typical large corporation in an established, slow-moving industry tends to be farther to the right.

Bottom-Up Top-Down Tech companies stand out by being farther to the left, some much farther to the left. Thriving in ambiguity. Rewarding autonomy and moving fast. Adapting quickly.

Bottom-Up Top-Down This is a key part of today’s tech industry culture. It’s enabled by technology, allowing fluid coordination on demand. Many industries are increasingly trying to emulate it.

Many organizations today are basically a slime mold.

The Coordination Headwind

Let’s get back to that bad outcome. How does it come about?

We’re going to see how these bad outcomes arise even if we assume every individual is hard-working, great at what they do, and extremely collaborative.

How is that possible? When organizations are very small, individuals matter most.

But when organizations grow larger, the system—its structure and dynamics—comes to dominate the analysis. We’ll focus on the system.

Of course, individuals always matter, no matter how important the system is. This is not to absolve individuals of agency or importance—your actions have consequences! It’s just to say that these bad effects can occur even if all of the individuals are behaving well.

Slime molds are difficult to study, since they’re inherently complex. But often even a small number of rules can capture the fundamental dynamics of a system.

For example, bird flocking can be explained with just three rules: 1. Avoid obstacles. 2. Align with neighbors. 3. Steer towards center.

One useful tool to understand these systems is agent-based models. You build a model that tries to capture some of the core decisions that agents face. This helps build a rigorous intuition. Let’s build a model!

Meet Amber and Brandon, two people working on a project within a given organization.

Let’s say Amber and Brandon have a project to accomplish some goal. We’ll assume the project either succeeds or fails.

The goal has some inherent value reward if it succeeds. We’ll represent it here with the size of this box.

To achieve the goal, they need to invest the necessary effort. They need to invest it at the right time. Investing at the wrong time is as good as not investing at all.

Each person has a different amount they must invest. The amount depends on their skillset, what the task at hand is, etc.

The project only succeeds if both of them invest their necessary effort.

Will they invest the necessary effort?

+ << The total amount of effort is much lower than the total amount of value created. The project is a no brainer.

So both Amber and Brandon invest the required effort. The value is unlocked. Everybody wins!

But in practice it’s not 100% likely that everyone will invest the effort.

Maybe someone got sick one day.

Maybe the power went out at the wrong time.

There are any number of reasons why it might not happen. The future is fundamentally uncertain.

So instead of investing with full certainty...

… it actually looks a bit more like this.

p( ) = p( ) p( ) • The project only succeeds if both Amber and Brandon invest the proper effort.

0.98 = 0.99 0.99 • If they’re both 99% likely to invest, then the probability of success overall is still 98%. Not too bad.

Let’s add one more required person to the project.

0.97 = 0.99 0.99 0.99 • • Not going in the right direction, but still not the end of the world.

Although in practice it might require many more people.

10 0.90 = 0.99 Even though everyone is still extremely likely to contribute, as the team size gets larger, the overall chance of success declines at a greater than linear rate. Uh oh!

Let’s add another complication. Before, everyone had just one project. But in practice at high-performing companies, everyone has many things they could be doing.

That means that the chance of any person investing in a specific project...

...is much lower. It’s more likely something else is legitimately higher priority!

10 0.60 = 0.95 And thanks to that non-linearity, even a small decline in individual likelihood leads to a much higher overall decline in likelihood of success.

But it gets worse. Previously, we assumed that it was pretty clear when the effort was less than the value that would be unlocked.

But in practice there’s often uncertainty. Let’s imagine uncertainty as a boulder, blocking our view.

Now it’s much harder to tell how much value would be created. In general, the farther into the future the value will be, or the more diffuse its effects, the more uncertainty there will be.

But it’s often even worse when it obscures how much cost it will be. What are the next steps? How expensive will they be? This gets even harder when looking at cost across many people.

<< It can make it extremely difficult to tell if a project is worth doing or not.

Let’s imagine the amount of effort is like moving spaces down a tunnel.

Each unit of movement takes one unit of effort. Perhaps, one hour of uninterrupted focus.

But at each step, there could be a boulder of uncertainty blocking the way. The boulder makes it hard to see the rest of the way down the tunnel.

You can’t pass that space until you clear up that uncertainty.

But clearing up uncertainty is really expensive. Instead of just executing, you need to study the problem and figure out what to do. You might need to consult with all of your other teammates to figure out the course of action. It can easily take an order of magnitude more time than a unit of execution.

Hopefully once you clear the boulder of uncertainty, the view is clear.

But in many cases, you’ll uncover more uncertainty!

And sometimes you’ll discover the goal is much farther away than you thought.

The expected difficulty is based on a guess at how far away the goal is, combined with the amount of uncertainty in this problem domain. The more novel the problem domain, the more uncertainty.

The expected difficulty will also differ from different vantage points.

People will tend to underestimate the difficulty of others’ tasks. “It’s just filling in a couple of spreadsheet cells, that can’t take more than a few minutes” People want to look capable, so they often won’t speak up about underestimates.

Uncertainty will also grow if different people understand the goal differently. This can arise if there is an implicit or explicit disagreement about the overall strategy.

Another things that makes this hard is the difficulty is tied to the context. That means that it has the potential to constantly change, including based on actions of others. Perhaps the decisions of a leader keep churning, creating more uncertainty.

Finally, the further out things get, the more the fundamental uncertainty. This uncertainty goes up super-linearly with longer time horizons.

All of this means it’s extremely difficult to judge difficulty ahead of time. It’s something that’s inherently debatable, creating the potential for disagreement. Sometimes, quite a lot of it!

Let’s imagine that Amber estimates her effort to be quite large for a given project.

It’s not crystal clear that the project is worth it compared to what else she has on her plate. So she prioritizes it a little less. Even if she still intends to do the work, if it’s not at the right time it won’t matter.

That makes the project less likely to succeed over all.

Everyone else has other important things they can be doing, too, so their likelihood of investing the proper amount goes down a bit, too.

Now that the project is less likely to succeed overall, everyone prioritizes it just a bit less. This goes on and on.

10 0.60 = 0.95 And even a small decline in overall probability of investing the proper amount...

10 0.35 = 0.90 … has non-linear effects.

p( ) = p( ) p( ) • So far the probability of success has been entirely based on the probability of the team overall investing the right amount.

p( ) = p( ) p( ? ) • But there’s another term that is extremely important.

== Often it’s tempting to assume that humans are basically robots.

== But of course humans aren’t robots.

Whenever two people work together, it’s possible for a friction point to erupt. These aren’t just “people not getting along.” There are often legitimate, good-faith differences of opinion.

Lots of things cause friction. Maybe the two people have different personalities. Or don’t trust one another. Maybe one person is having a bad day. Maybe the area they’re in has a lot of uncertainty.

We’ll assume that if a friction point erupts, it causes the project to fail.

In practice though, people are pretty good at navigating friction. The chance any given relationship erupts is pretty low.

But of course, in a team of 10 people, there are 45 pairwise connections. In general, it’s roughly O(n2). If any single one of them erupts, it could cause the entire project to fail!

If a friction point triggers, you want to address it early. If you don’t...

… it tends to get worse and worse. Distrust breeds more distrust. If you’re already distrustful of someone, it’s easy to find more confirming evidence. This is partially due to the fundamental attribution error.

What is the fundamental attribution error? If I’m late for coffee with you, it’s because of traffic: systemic factors. If you’re late for coffee with me, it’s because you don’t value others’ time: intrinsic qualities. We tend to overestimate the importance of intrinsic qualities in others’ failures.

This means that distrust can ratchet. A triggered friction point is likely to spiral out of control. In a complex or chaotic system, we blame individuals… especially ones we already don’t trust.

So if a friction point occurs, you want to handle it early. You want a mediator, someone both sides can trust to be fair. Someone that everyone trusts to have a positive-sum perspective.

If both people are on the same team, there’s a convenient mediator: the manager! Amber can flag the issue in a recurring 1:1. The manager can check in with Brandon. The manager can then nudge the problem to a resolution.

The issue is resolved. That was easy! This kind of resolution is called an escalation. In this situation it was a no brainer.

This gives us our first bit of good news. If people who work together are on the same team, then it’s easy to resolve friction points as soon as they show up. The overall risk of friction is tiny.

But what if the nearest common ancestor is two layers up?

Amber could enlist the help of her manager.

They could then reach out to Brandon’s manager.

This will have taken more effort overall, with more people’s input. Hopefully they can resolve the issue.

But maybe the disagreement is about something fundamental. Perhaps it’s a tradeoff with real downsides and a non-obvious balance point. They’ll need to escalate to their mutual manager.

There are now a lot of people involved. Escalating is a charged act, especially if there are high stakes. This means it’s even more likely to have spiraled out of control. Depending on the decision, it could be really bad for Amber, or Brandon, or both.

So in practice, problems tend not to get escalated. The problem is small, after all. Everyone’s an adult, they can solve it themselves. No one wants to throw anyone else under the bus.

But problems like this tend to fester.

And before you know it, there’s a supercritical problem lurking. One accidental slight. One forgotten meeting invite. One unintentional eye roll. One tiny trigger event and...

You have a serious problem on your hands. The problem is already pre-emotionally charged. It’s hard to even have a calm discussion about the facts. The situation will almost certainly leave some scars.

The potential for problems goes up the more layers there are between two people’s nearest common ancestor, at a greater-than-linear rate. It’s a Big Deal to escalate so high!

The answer can’t be “just escalate more”. Escalations are pre-emotionally charged. They take tons and tons of time. The higher up they are, the more likely the details are to be glossed over, and an actively bad decision is to be made. And the senior leaders’ time is finite!

The farther apart two people are in the org structure, the more likely there is to be disagreement in the first place. There’s more likelihood of strategic misalignment, cultural differences, time-horizon differences, etc.

This is true to some extent in all organizations, by the way. Some cultures that emphasize collaboration might have “ruinous empathy”, where disagreements are submerged. But even in direct cultures, leaders’ time is finite.

1.3 p( ) = h(e , e ) 1 2 So the probability of a friction point for a single pair is a super-linear effect tied to how many layers up in the hierarchy until you reach the nearest common ancestor

n n-1

1.3 p( ) = h(e , e ) i j i=0 j=1 But that’s for a single pair. A given project will have roughly O(n2) pairs. Each pair will have a different hierarchy factor. This gets very big, very fast.

We’ll call this part of the headwind the organization headwind. It rises extremely quickly with the number of people who must coordinate on a project. That means there should be a strong pressure for the smallest teams possible.

p( ) = p( ) p( ) • We now have the two major components of the coordination headwind. The probability each person properly invests, and the organization headwind.

The coordination headwind is inexorable. It increases super-linearly as: 1. Uncertainty increases 2. Project teams get larger and more spread out 3. A more bottom up culture

In many organizations, it’s easy for there to be hurricane-force headwinds.

When the coordination headwind looms, there’s a solution: time. Lots and lots of it. Team bonding events. Heaps of 1:1s. Strategy summits. Collaborative debate. Lots of “preventative maintenance” to preemptively build trust across organizations. And lots and lots and lots of tracking spreadsheets.

The amount of additional time and effort required is immense.

Nobody wants to look weak, or like they can’t do their job. And people often don’t even realize how significant this expense is. This leads to everyone seeing only the tip of it.

This significantly underestimates the cost, and nearly guarantees frustration and confusion.

Because the cost is mostly invisible, people continue working the same way as before. But projects take multiple orders of magnitude more effort than they did before. Everyone takes on way, way too much work and is on the verge of burnout.

When this gets really bad, it can create new types of horrific dysfunctions, especially in extremely diffuse organizations without one coherent strategy.

At the beginning people dive into the (mostly invisible) headwind with heroics. Doing things that cohere with other teams is extremely hard to coordinate, so the fastest path to action is often striking out in your own direction, or ignoring inconvenient or disconfirming information.

All of these heroics and motion create churn and more uncertainty. That uncertainty creates an even higher coordination headwind for everyone, necessitating more heroics.

As the ability of the overall organization to build anything coherent declines, the focus on coordination goes up to counteract the chaos. This leads to a situation that is tightly coupled, loosely aligned. Doing even the smallest things becomes impossibly expensive.

As the headwinds get stronger and stronger, individuals start to doubt themselves. The things that used to work so well no longer work. Have they just lost their hustle? The solution must be to try even harder.

Everything is in chaos, so it’s even harder than normal to figure out an action’s non-local effects. Instead of people being motivated to maximize long-term positive impact, people are incentivized by the appearance of heroic motion.

If your area has a problem, it’s a liability. It doesn’t reflect well on you. It also makes it more likely a lead will swoop in to “fix” it and make things worse. So people increasingly hide problems from leadership. The result is leadership is increasingly totally in the dark about what’s actually happening.

As things get more chaotic and randomizing, it gets harder and harder to strategize for the long term. Time horizons drop from multiple years, to a few quarters, to a few weeks. The goal is just holding on to live another day.

The “game” has become all-encompassing. People feel torn between doing what they know to be right and what the powerful emergent incentives of the system demand. People who resist get increasingly worn down and burn out.

In the end, everyone ends up just vibrating in place, trying to look heroic, accomplishing little but creating more froth and chaos around them.

The situation has reached its end point: total, demoralizing chaos. Each unit of effort achieves many orders of magnitude less real-world impact than it used to. Everybody is helpless to do anything about it.

We started out with people that are hard-working, great at what they do, and extremely collaborative. But in this situation it’s impossible not to become burned out and complacent.

What to do about it

Unfortunately, there aren’t easy solutions to this problem. Bottom-up systems are extremely hard to control. Most of the obvious “fixes” will have exactly the wrong effect. Let’s review some things to not do.

One bad option is to ignore it. That’s like architects ignoring gravity because it makes designing buildings harder. You’ll push the dysfunction underground, making it fester and spread. People will feel gaslit and will burn out.

Another bad option is for individuals to just give up. They can’t change the system so they might as well not even try. But the actions of individuals absolutely matter, and complacency will make it way worse. Don’t hold yourself to an impossible standard, but take responsibility for the choices you make.

Another bad option is to try to switch to a top-down approach. In complex environments, a top-down approach will be very brittle. A poorly-executed top-down approach is far, far worse than a messy bottom-up approach. Also, cultures take an extremely long time to change and build new competencies.

Another bad approach is to have senior leads dive in to “help” with details. But the details are not the problem, the system is. Intervention without understanding of the details creates churn and uncertainty. More churn and uncertainty makes the problem much, much worse. And leads’ time is finite.

Another bad approach is to rely on heroics. Sprinting heroically into a hurricane-force headwind can only last so long. Relying on unsustainable heroics will prevent sustainable approaches from being invested in. Heroics also tend to make the problem way worse, by creating more churn and uncertainty.

Another bad approach: aiming for perfection. If it’s really expensive to do each action, we should at least make sure each one is perfect, right? But now every action will take huge amounts of coordination and will be impossibly expensive. If there’s even the smallest amount of uncertainty, absolutely nothing will get done.

Remember that the problem is rarely someone being a villain. It’s almost always the system. When you find someone that’s frustrating you, breathe. Assume they’re trying to help you avoid a cliff you can’t see, or are under hidden constraints. Be curious, empathetic, and seek to understand.

So what should you do? First, take the time to accept that many organizations are like slime molds.

Slime molds have many challenges, but they also have some amazing abilities. Instead of trying to fight it, maybe lean into what they’re good at? Slime molds are extremely resilient. They can handle complex and changing conditions well. Creative solutions pop up organically. They can create more value than the sum of their parts.

One technique is to let go of unnecessary detail and coordination. Is the project converging to a good enough outcome on a good enough timeline? Good enough! Perfection is impossibly expensive, especially when there’s a big coordination headwind.

Look for a way to decouple things as much as possible. Instead of having everything perfect today, aim for eventual convergence.

Properly accounting for headwinds is critical. Some ideas that appeared to be no-brainers will turn out to be terrible ideas. Some ideas that appeared to be meh will turn out to be great ideas.

If you do this right, sometimes you can find ideas that are just a bit different but have massively different amounts of headwinds. Start out with a smaller team localized in one part of the organization. Have a smaller goal that’s quicker to achieve.

Let’s bring more of these ideas together. Be wary of moonshots that are executed as a straight shot to the moon.

When they’re tackled in one straight shot, they take on large amounts of risk. Perhaps you misunderstood some of the constraints. The feedback loop is way longer, making it hard to sense and learn about the problem domain. Overall, the miracle count is way higher.

The better model sights off the moon, but doesn’t try to execute straight there. Work with your team to figure out a rigorous, plausible goal for the next 3-5 years. This is your moon: your strategy.

Now, sight off the moon, and pick a no-brainer roofshot step that will bring you closer. Instead of a giant leap of faith, you’re taking a safe step to unlock value. You’re also moving closer to your ultimate goal.

The roof shot significantly reduces the execution risk. Once you’ve done the roofshot, you’ve locked in that step of value, and have gained momentum for the next step.

Then repeat!

And keep repeating! This will not be the most efficient path, and that’s OK. You’ll waste effort, but you’ll do it in a safe way that continuously unlocks more value.

This allows you to respond to surprises, both new constraints and new opportunities, as you go. You’ll be more flexible and resilient.

If you would have just sighted off the moon, you’d have an impossible jump. If you had just executed blindly, you would have a random walk through the opportunity space. But by doing iterative roofshots sighting off the moon, you’ve gotten the best of both worlds.

This pattern gets even more powerful with multiple teams. Independent teams will tend to go in random directions, or even actively different directions. This means over time they tend to diverge.

But if they’re all sighting off the same moon, over time they’ll tend to naturally converge. This allows an ideal pattern of loosely coupled, tightly aligned. With a widely agreed upon strategy, there will be a smaller risk of disagreements.

Having a strategy is not free, or something to take for granted. Having a rigorous, plausible strategy that many teams can sight off takes work. You need a clear-eyed accounting of wide-ranging constraints and opportunities. The broader the org’s ambitions, the harder it will be.

Bad strategies and good strategies will look superficially similar. It will take years before the difference is clear. Good strategic work often looks wasteful or indulgent in the moment. Even in retrospect the strength of the strategy will be hard to prove.

People doing grounded, strong strategic work will look kooky, different, or maybe even self-indulgent. But that work is extremely important, and you should celebrate it.

Organizations are like slime molds. If you had fought that fact you would have gotten frustrated and burned out. But by embracing that truth, you’re able to unlock value, sustainably.

Focus less on being a builder, frustrated that your building materials refuse to behave.

Instead, think of yourself more as a gardener.

edit: upon it outputting a summary, asking it to reproduce as continuous prose it refuses.

One shotting and asking not to summarize it complies? I hate this world sometimes.

it's available here tho: https://nathancheng.fyi/03-Records/Clippings/Coordination+He...

Cunningham, I obey thy dictum without fail. Bless me for my service. I praise thee with my every act.

Behold the arcane acts at work: https://imgur.com/a/Zg0z5k4

Works of magnificence no mere mortal could achieve.