The company had a few hundred people when it changed the culture. It had thousands by the time it couldn't find what made it special.
I was there from the early days. I watched the distance between having an idea and doing something about it get longer, one quarter at a time. A creative partnership that should have taken two weeks. The kind of thing that made the company electric, the kind of thing it used to do on instinct. Instead: four teams, a project manager, a stakeholder map, and an alignment session to figure out who the stakeholders were. By the time the deal closed, the cultural moment that made it matter had already passed.
Nobody in those rooms was slacking. That was the part I couldn't reconcile. Everyone was talented. Everyone was trying. The work had shifted, and the shift was invisible to the people inside it. What used to be about making the thing had become about coordinating the people who coordinated the people who made the thing.
Every new hire made it worse, and every new hire made sense. Revenue was growing, so the company hired. New teams needed managers. Managers needed process. Process created meetings, roadmaps, review cycles, alignment rituals. Each step was logical. Each step was defensible. The cumulative effect was an organization that spent more energy maintaining itself than making the product that earned the revenue.
I left before the correction came. Then I watched six CEOs write, almost word for word, the same letter to six different companies.
In 2023, Mark Zuckerberg declared a "Year of Efficiency" at Meta and cut 25,000 employees. Flatter is faster. He removed entire layers of management, restructured reporting so individual contributors could reach leadership without five levels of translation, and Meta's stock returned 194% that year. Best performance on record. The cuts didn't create the value. They made it visible.
The company I watched this happen to was Epic Games.
In 2017, a few hundred employees built Fortnite Battle Royale as a side project inside a side project. Within a year: a hundred million players. $5.8 billion in revenue from Fortnite alone. The most culturally relevant product in entertainment.
Then the hiring started. By 2019, roughly 900 employees. By 2022, around 5,600. Revenue over that period roughly doubled. Headcount grew three times faster.
Fortnite generated billions every year. The cultural peak was 2018, but the revenue kept flowing, hovering between four and six billion annually even as the initial frenzy cooled. Headcount kept climbing through all of it. The company was adding management layers while the product those layers were managing had already found its ceiling.
In September 2023, CEO Tim Sweeney sent a memo to the company. "For a while now, we've been spending way more money than we earn." He cut 830 people. Two-thirds of the cuts, in his words, were in teams "outside of core development."
Outside of core development. The CEO telling the world, in a public letter, that most of the organizational growth had nothing to do with the product. It was what grew around the product.
It didn't fix it.
March 2026. Another memo. "I'm sorry we're here again." A thousand more people. Fortnite's average monthly playtime cut in half since 2023. $500 million in additional cost savings identified from contracting, marketing, and open roles.
Two rounds. Thirty-six percent of the peak workforce. And still spending more than it earned.
He wasn't the only CEO writing this letter.
Stripe's Patrick Collison named it plainly: "allowing coordination costs to grow." He cut 14%. But Shopify's Tobi Lütke tried something different. He didn't fire anyone. He deleted meetings.
All of them. Every recurring meeting with more than two people: canceled. 322,000 hours of projected meeting time wiped from the calendar. Wednesdays became meeting-free. Large gatherings restricted to Thursday afternoons. He called it "chaos monkey," borrowed from the engineering practice of deliberately breaking systems to test what's actually load-bearing.
The result: 25% more projects moved toward completion in the following quarter. By year end, Shopify's free cash flow had swung from negative $186 million to positive $905 million. The company didn't launch a new product. It deleted the coordination tax and let the existing product breathe.
Lütke didn't cut the people. He cut the layer those people were trapped inside. "The best thing founders can do is subtraction," he said. The meetings weren't the work. They were the tax on the work. And 322,000 hours of them had been invisible until someone deleted them and measured what happened.
Same era. Same industry as Epic. Three hundred and fifty employees.
Valve Corporation generated $17 billion in revenue in 2025. Roughly $50 million per employee. More per person than Apple, Google, or Meta. Flat structure. Self-directed projects. No middle management to speak of.
Epic at its peak: 5,600 employees, roughly $4 billion in revenue. About $730,000 per employee.
The ratio overstates the case. Steam is a marketplace collecting commissions on work other companies' employees produce. Epic builds games, maintains an engine, and runs a store that operated at a loss for years. Halve the number. Quarter it. The gap is still staggering, and the organizational difference is real: three hundred and fifty people with no middle management built Steam, Half-Life, Portal, and Dota. No approval chains. No alignment meetings. No recurring syncs about the recurring syncs.
The question isn't what Valve's exact revenue per employee is. The question is how many of Epic's fifty-six hundred were doing something that three hundred and fifty people proved you didn't need.
Then there was Klarna.
In early 2024, CEO Sebastian Siemiatkowski announced that Klarna's AI assistant was handling two-thirds of customer service chats. Handling the equivalent volume of seven hundred human agents. Cases resolved in two minutes instead of eleven. $40 million a year saved. Five thousand employees became three thousand. He called it a win.
Eighteen months later, he reversed course.
"Lower quality." His words. The chatbot handled scripts. It could not handle a frustrated customer who needed to feel heard. Klarna began rehiring the exact people it had automated away.
Siemiatkowski had cut the moat thinking it was tax. The scripted responses were tax. AI handled those. The judgment, the instinct for when to deviate from the script, the pattern recognition built through thousands of difficult calls: that was moat. He couldn't tell the difference. The customers could.
Boeing made the same mistake. The consequences weren't lower satisfaction scores. They were planes falling apart.
In the early 2000s, Boeing outsourced 70% of the 787 Dreamliner's design, engineering, and manufacturing to fifty-plus partners. Shed the expensive capability. Keep the integration layer. Capture the margin.
An engineer warned that "the day would eventually come when there wouldn't be enough in-house capability to even write the specs." It came. The supplier management team, in one internal assessment, "didn't have diddly-squat in terms of engineering capability" once the work was gone. Manufacturing knowledge that lived in experienced workers' hands, the kind "not reflected on the planning documents," had to be relearned from scratch.
The 787 shipped three years late. Billions over budget. The 737 MAX crisis traces to the same hollowing: flight-critical software outsourced to engineers making $9 an hour. In 2024, Boeing bought back Spirit AeroSystems for $8.3 billion. The institutional capability it sent out the door a decade earlier cost eight billion to bring home.
Klarna lost customer service quality. Boeing lost the ability to build safe airplanes. Same diagnostic failure.
Here is the pattern underneath all of it.
Every organization carries two kinds of knowledge. The first kind can be documented, taught, systematized, repeated. It scales. It survives turnover. AI can increasingly do it. The second kind is what people know but cannot fully articulate. Judgment. Relationship memory. Sequencing instinct. The pattern recognition that comes from ten thousand reps and can't be reduced to a playbook.
This second kind splits into two varieties that look identical from the outside and produce opposite economic effects.
The creative instinct that produced Fortnite's cultural collisions. The veteran agent who reads a caller's frustration and adapts before they escalate. That's the moat.
The twelve-person meeting that exists because nobody documented the decision framework. The coordination costs Collison named. The 322,000 hours Lütke deleted. The layers Sweeney admitted were "outside of core development." That's the tax.
Call it the tribal tax.
The tax looks like work. That's why it takes a crisis to cut it.
Meetings look like alignment. Decks look like strategy. Approvals look like risk management. Every person performing these functions is talented and trying. They are contributing to the organization's self-maintenance. They are rarely contributing to the product, the customer, or the revenue.
The law underneath: organizational growth is self-funding and self-justifying until revenue stops growing. Then the tax appears all at once. Stock-based compensation finances the delay. Snowflake spent 34% of its revenue on SBC last fiscal year. When equity is cheap, the cost of organizational complexity disappears from the metrics everyone watches. Stock comp wasn't free labor. It was deferred organizational truth. When valuations compress, the mass that equity subsidized becomes suddenly, painfully visible.
Meta cut it and the stock returned 194%. Shopify subtracted it without cutting a single person. Klarna cut the wrong thing and had to reverse. Boeing outsourced it and spent $8.3 billion buying it back.
The skeptic's strongest card: Meta and Shopify had strong products that would have recovered regardless. The cuts were incidental. But Shopify's stock had dropped 85% before Lütke deleted the meetings. Boeing had decades of market dominance. It didn't save them from outsourcing the wrong capability. The product isn't the variable. The diagnosis is.
Thickness is not the disease. Opacity is. Not all thickness is tax. Some of it is safety, training, resilience, institutional memory. Amazon has more management layers than any company in this essay. But every layer is instrumented. Every team has a fitness function. Amazon is thick and legible. The tribal tax isn't thickness. It's thickness you can't see. And in regulated industries, some coordination layers ARE the product. Boeing's failure wasn't too many layers. It was removing layers that were safety infrastructure. The framework is hardest to apply exactly where the stakes are highest.
This essay is built on survivor data. Every company here lived to tell the story. The ones that couldn't recover aren't writing memos about what they lost. The framework isn't "cut." The framework is "see what you're carrying before you decide what to set down."
Two questions separate the companies that diagnosed correctly from the ones still writing memos.
If the three people with the most institutional knowledge left tomorrow, what breaks first: the customer experience or the internal navigation? If the answer is internal navigation, the knowledge those people hold serves the maze, not the product. You're running on tax.
Can a new hire become productive in weeks, or does it take months of learning unwritten rules, building relationships, and decoding how things actually work? Long onboarding is a tribal tax audit. It means the organization charges every new employee the cost of deciphering its own opacity.
If you cannot explain how work gets done, you are renting tribal tax and calling it experience.
The hardest version of this argument is that the tax is invisible by definition. If you could see it clearly, it wouldn't be tribal. The two questions above are proxies, not guarantees. Boeing's engineers presumably knew what they were outsourcing. They still lost capability they didn't know they had. Lütke's meeting purge worked because meetings are visible and deletable. The deeper tribal tax, the kind that lives in one person's judgment or one team's unwritten sequencing instinct, doesn't show up on a calendar. It shows up in the quality of decisions six months after the person who held it is gone.
Which is exactly why the framework matters. The diagnosis will never be perfect. But companies that ask these questions before cutting will get closer than companies that reach for headcount reduction as a default. Shopify asked what was load-bearing and deleted what wasn't. Boeing assumed it knew and spent $8.3 billion learning it was wrong.
AI accelerates the reckoning without resolving it. It compresses the codified layer: drafting, summarizing, diffusing best practices. Dorsey is right that most companies will reach the same conclusion within a year. What he left unsaid is that some of them will make Klarna's mistake. They'll automate the judgment layer because it looks expensive and keep the coordination layer because it looks like work.
Within a few years, at least one major company will have a Klarna moment in a core function. Not customer service. Legal, engineering, product strategy. They'll compress the codified layer with AI, hollow out the judgment layer through attrition, and discover what they lost only when something breaks that the model couldn't anticipate. Boeing learned this lesson over a decade of outsourcing. AI will compress it into months. The tribal tax framework predicts exactly where: the closer work sits to judgment, the more dangerous it is to automate without first asking what's moat and what's tax.
In a thinner organization, information is cheap. Conversion is scarce. The ability to turn agreement into action. Translators who move between functions. Codifiers who turn memory into systems. Judgment operators who decide what to build, what to skip, and in what sequence.
The next org chart will not eliminate hierarchy. It will justify it differently. Fewer layers whose purpose is translation by default. More explicit decision rights. More systems that make coordination costs visible before they compound. What the org chart rewards is what the org chart gets. Right now most of them reward navigating complexity. The ones that survive the next decade will reward reducing it.
I've spent my career oscillating between scales and languages.
Investment banking. Venture capital. Business school. Gaming partnerships at a company that went from a few hundred people to thousands. Independent studio. Healthcare acquisitions. Every transition crossed a boundary no org chart anticipated. Every role I was good at required fluency in things that didn't share a department.
At Epic, I helped close partnerships that put Fortnite in conversation with the NFL, with Balenciaga, with the estate of Martin Luther King Jr. Each one meant holding four conversations simultaneously in four different languages: what the partner needed to protect (legacy, brand equity, competitive position), what Epic needed to build (something culturally authentic, not exploitative), what the community would accept (earned presence, not corporate intrusion), and what the press would write (either a meaningful use of the medium or a tech company commercializing something sacred). Four conversations. None in the same vocabulary. All of them invisible to the org chart.
In a thin company, that work is the most visible thing in the building. In a thick company, it becomes illegible. It crosses too many lines. It doesn't map to a single metric. The person doing it looks like a misfit because their contribution doesn't sit in one place.
Thick organizations structurally select for skills that maintain their complexity and deselect for skills that would simplify them. The people who thrive are the ones who navigate layers well: process management, stakeholder alignment, political fluency. Real skills. They matter inside the system. They only matter because the system is thick.
The people whose value lies in making layers unnecessary, in reducing the distance, in turning tribal memory into systems and judgment into action: those people look awkward on the org chart. Their work threatens the structure because their existence implies the structure is overbuilt.
A clean fit was often just someone whose skills matched the organizational tax.
I didn't know that when I was inside it. I thought the problem was me. It took leaving, and years, and watching six CEOs write the same letter to different companies, to see what I'd actually been doing all along.
The distance was always the job.