Uber is cutting 3,300 people this week, about 10% of the company. Dara Khosrowshahi’s memo is already in the genre: more layers, more coordination, more fragmented ownership. A simpler org chart “geared toward building versus managing.” Managers down 20%. Half of the one- and two-person “micro-teams” gone. People sitting seven-plus layers from the CEO, down 20%. Remote work, down to about 1%. The savings go into growth, and into what Dara calls the autonomous future. Uber has pledged more than $10 billion to robotaxi partnerships.
Gergely Orosz talked to old Uber colleagues. For months, HR had been calling long-tenured engineers and managers into rooms, offering packages, making it clear the company would rather they leave. Some negotiated. They left. So the 10% isn’t a surprise. It’s the public version of a decision that already had a dress rehearsal.
I don’t think this is really an Uber story.
After Twitter
It’s impressive how much one person moved an entire industry. Elon didn’t invent layoffs. He gave everyone in tech license to do them in public, at a scale that used to get you a crisis comms team and a week of CNBC. Twitter in 2022 didn’t die. The product kept running. “Hardcore” became a compliment. “Managers who don’t code” became a punchline.
After that, flattening stopped looking like a failure of leadership and started looking like proof of it. Meta had a Year of Efficiency. Amazon and Google and Microsoft all found ways to say the same thing in their own dialect. The memos converged on one moral: coordination is waste, layers are the disease, and the highest-status engineer is a hardcore coding machine.
Uber’s letter could have been written in 2023. The fact that it’s landing in 2026 is the point. The laggards are still internalizing the mandate.
What actually got made uncool
Command never went out of fashion. Tesla and SpaceX still run on extremely strong leadership archetypes. Founder-mode is high status. The IC who ships is high status. What got recoded as bloat is a specific job: the people manager. The skip-level. The career ladder. The person whose work is holding context, resolving conflict, and keeping a messy org pointed at one thing without being the founder.
I did that job. It is not the same as writing the code. It is also not “alignment meetings.” When it works, a team of people who would otherwise drift into local optima keep making the same product. When it doesn’t, you get the cartoon: calendar as identity, process as product, a manager who hasn’t opened the codebase in two years.
The industry looked at the cartoon and deleted the job. Some of those people became ICs. Some got packages. The ones still doing it learned to describe themselves as tech leads, player-coaches, “still in the code.” You can watch the language change in real time. Engineering manager is something you apologize for.
The other timeline
AI made the budget move easy. Headcount is a cost line. GPUs and robotaxis are an investment line. Same dollars, better story on an earnings call. Uber is unusually honest about the substitution: fewer people, more autonomous vehicles. Most software companies are doing a softer version of the same trade. Humans to data centres.
You can imagine the other timeline. These companies use AI to press on the gas. More products, more cities, more experiments. The org gets weirder and bigger, not smaller. Output per person goes up, so you attempt more, not less. You eat the world.
That timeline lost for boring reasons. After cheap money ended, “efficiency” was the story that moved a multiple. A layoff is a way to tell the market you’re serious. “AI will cover it” is cheaper to say than “we’re going to get bigger and the org will get stranger.” For Uber the substitution is partly real. Drivers and ops versus cars that don’t need either. For most of the industry it is fashion with a capex receipt.
I wrote about this last year as theater: the language of transformation, the practice of shrinking. The ladder got pulled up first, at the junior end. Now it’s the middle.
A few more years
This has some time left to run. Finance, healthcare, government, the second wave of software companies that still have skip-levels and 1:4 manager ratios. They will all produce a version of Dara’s letter. The words are already written. Building versus managing. Clearer ownership. Faster decisions.
Some of that is true. Plenty of orgs did get fat. Plenty of managers were coordinating because the org was too complicated for anyone to own anything. Flattening a 1-2 person “team” is often just admitting it wasn’t a team. I’m not arguing for the 2019 org chart.
I’m arguing that we are running an industry-wide experiment in deleting a skill, and we are doing it on a fashion cycle.
The bottleneck
Models keep getting better at producing. Code, docs, tickets, designs, analysis. The thing they do not do is decide what deserves attention. They do not hold a P&L. They do not notice that two teams are building the same thing, or that a quiet person has the actual answer, or that the agent swarm is confidently executing last week’s goal.
The more capable the models get, the more expensive attention and alignment become. A swarm of agents plus eight humans plus a marketplace plus regulators is a management problem. Org design starts requiring the thing we just made uncool.
That job will not look like 2019 EM. It will not be skip-levels and promo packets. It will look more like operating a factory of cheap cognition: what to point it at, what to check, when to stop, who is accountable when it is wrong. That is still management. Calling it “orchestration” or “context engineering” does not make it an IC skill.
Then the shortage
By the time the fashion reverses, the pipeline is gone. You spent years converting managers into ICs, or out. You taught ambitious people that the job was a career death. You cannot mint judgment in a bootcamp. The people who could hold a messy org together will be founding, consulting, or happily writing code with a model in their ear, and they will not come back for a title they watched get mocked.
I would bet on this sequence. A few more years of laggards flattening. An inflection when some company actually hits the attention wall: too much output, not enough people who can aim it. Then a scramble. Comp for “operators” spikes. Everyone pretends they always valued management. The same memos will get rewritten with the opposite moral.
We’ve done this before. We over-hired in the cheap-money years, then treated the correction as a philosophy. We will over-delete management, then treat the correction as a discovery.
How humans work
Status is contagious. A high-status person fires a lot of people and the product doesn’t fall over, so firing people becomes high-status. A high-status person treats people-managers as overhead, so the archetype becomes embarrassing. Then a high-status company wins by being well-managed, and we will all hire managers again, and we will act like this was obvious.
The expensive part is the gap. You can rebuild a toolchain in a quarter. You cannot rebuild a generation of people who know how to run a team through a messy year.
Uber will be fine. The robotaxis might even work. That’s not the interesting question. The interesting question is what happens when output is cheap, attention is scarce, and the people who used to hold the middle have been trained to be ashamed of the job.
Incredible how humans work.