Andrew Anagnost, the CEO of Autodesk, fired a thousand people in January and wanted you to know something. The cuts, he said, were “not driven by the external environment or an effort to replace people with AI.”
The press wrote it up as Autodesk’s AI pivot anyway. He’d spent the whole memo talking about “strategic shifts” toward AI, then bolted on a sentence swearing AI had nothing to do with the people he’d just put on the street. Both things, in one document, with a straight face. That is the entire genre in a single man: blame the AI for the strategy, absolve the AI for the bodies. Pick one, Andrew.
I have been collecting these. Forty-five CEOs and counting have stood at a podium in the last eighteen months and pointed at a machine. Almost none of them can keep their story straight for the length of a press release.
“A moment of strength”
ASML cut 1,700 jobs and the CFO, Roger Dassen, called it choosing “to make these changes at a moment of strength.” We are apparently so strong that we are removing 3.8% of the humans. Julie Sweet at Accenture went one better, announcing that the people Accenture couldn’t retrain for the AI era would, in her words, be “exited.” Exited. Like a building. Eleven thousand of them.
Klarna’s Sebastian Siemiatkowski spent a year bragging that his AI did the work of 700 customer service agents. Magnificent. Then satisfaction cratered and Klarna quietly started hiring humans back, and Siemiatkowski conceded that cost had been “too predominant a factor” and that people, it turns out, prefer to talk to people. The AI did the work of 700 agents right up until it didn’t.
Nobody in the press release says the obvious. The AI is a costume. Underneath it is the same tired story it has always been: the company overhired, the growth slowed, the board wants the margin back, and “AI” is the word that makes a cut sound like a vision instead of a confession. I have written before about what happens when the org chart fills up with AI agents, and almost none of it looks like a layoff.
The men who build the machine are calling bullshit
The two people on Earth with the most to gain from you believing AI eats jobs are the two people telling you to stop lying about it. I can’t get over it. Jensen Huang sells the chips. Every single one of these layoffs is supposedly funded by hardware his company makes, so when he went on Channel NewsAsia and torched the whole narrative, it landed hard. “Too lazy,” he called it. “Irresponsible.” Then he asked the question that should have emptied a hundred boardrooms: “AI has just arrived. How is it possible they’re already losing jobs?” “How is it possible that AI became productive and useful only six months ago, and they were somehow laying people off two years ago because of AI?” His verdict on the motive: it lets executives “sound smart and I really hate that.” His verdict on the cost: “we’re scaring people, and that’s irresponsible.”
Demis Hassabis runs Google DeepMind. He builds the coding models the layoffs are blamed on. He called the layoff logic “dumb,” “a lack of imagination,” then did the math the cost-cutters refuse to do out loud: “If engineers are becoming three or four times more productive, then we just [want to] do three or four times more stuff.” He said he’d take more engineers tomorrow, not fewer. The line that should be the headline: “Perhaps there is an ulterior motive for putting those messages out, raising money or whatever.” Raising money. The guy who makes the machine just told you the layoffs might be a pitch deck. That’s the whole game.
Sam Altman’s own valuation depends on you fearing his product, and even he gave it a name. “AI washing”: companies “blaming AI for layoffs that they would otherwise do.” Strauss Zelnick at Take-Two said it with no anesthetic at all. The big tech companies “who laid off thousands of people and said it was because of AI were not telling the truth. It was because they overhired in the pandemic and they were sloppy about it.”
The catch: the layoffs aren’t paying for the AI
Meta’s entire human cost, every salary, every benefit, every stock grant for every employee, runs about $27 billion a year. Meta’s 2026 AI capital budget is $125 to $145 billion. Fire every human being who works at Meta, on the planet, and you don’t cover a fifth of the chip bill. The 7,000 people cut this May? A rounding error. A decimal place.
The layoffs cannot be how Zuckerberg pays for the machines. The math is off by a factor of five. The firing pays for something else entirely.
It pays for the story. The layoff is the gesture of restraint that lets a CEO write the most reckless check in corporate history without his stock cratering. He commits $145 billion to a bet that barely works yet, and to keep Wall Street from gagging, he tosses a few thousand humans overboard and calls it discipline. The analysts nod. Ah, he’s funding the future by trimming the fat. The engineer’s salary never touched a GPU. It bought a sentence on an earnings call. She was sacrificed to make a number look brave.
Zuckerberg, give him this, said it on the record. He told staff the May cuts were a direct consequence of the AI infrastructure budget. Not “AI did your job.” He’d rather own the chips than employ you. At least that’s honest.
The catch under the catch: the money runs in a circle
What is the $725 billion of compute actually for? Serving AI demand. Where does the demand come from? The floor gives out under that question. Roughly half of the $2 trillion in cloud commitments holding this thing up comes from two companies, OpenAI and Anthropic, neither of which makes a profit. OpenAI burns north of $60 billion a year in cloud against maybe $25 billion in revenue.
Follow a single dollar. Nvidia commits up to $100 billion to OpenAI; OpenAI commits to buy Nvidia chips. Microsoft puts $13 billion into OpenAI as Azure credits; OpenAI spends the credits on Azure; Microsoft books it as revenue. Alphabet posts a record $62.6 billion quarterly profit, and $28.7 billion of it isn’t from selling a thing, it’s from marking up the value of its own stake in Anthropic. The supplier funds the customer who funds the supplier. It’s a hall of mirrors with a payroll department.
The whole machine is finally visible. A CEO bets a sovereign-scale fortune on compute. The demand justifying that bet is partly the same dollars lapping the track. To make it look disciplined, he fires people. To make the firing look like genius instead of fear, he blames AI. And the worker, who had no seat at that table, eats the risk of the entire wager, all to feed a machine that, by Scale AI’s own benchmark, can finish 2.5% of a real freelancer’s multi-day work at human quality. Two and a half percent. That’s the terminator you got fired for.
It doesn’t even work
If the cuts at least bought the productivity miracle, you could mount a cold defense. They don’t.
Gartner studied the companies deploying autonomous AI and found about 80% had cut staff, and the cuts “did not appear to translate into a stronger return on investment.” Helen Poitevin, who ran it, in one sentence: “Workforce reductions may create budget room, but they do not create return.” Budget room. Not return. They emptied the building and the ROI didn’t show up.
The wider data refuses to play along with the panic. Andreessen Horowitz, not a firm that talks AI down, called the jobs apocalypse “a complete fantasy,” citing an NBER paper finding “no meaningful changes in total employment,” a Federal Reserve Bank of Atlanta survey where over 90% of firms reported no employment impact from AI in three years, and the Yale Budget Lab finding the labor picture “largely reflects stability.” Ninety percent of firms, with no shareholders in the room, admitted the machine hadn’t touched their headcount. Then a good chunk of them walked to the podium and blamed the machine.
There is one real wound, and we did it to ourselves. Stanford’s Digital Economy Lab found a 16% drop in employment for workers aged 22 to 25 in the most AI-exposed jobs, while their older colleagues held steady. That isn’t the machine replacing the young. That’s companies cutting the cheapest cohort first to make the quarter pretty, and paving over the on-ramp while they’re at it. MIT’s Andrew McAfee said it plainly: automate the juniors and you snap your own talent pipeline. You’ll find out in four years, when nobody left on the team ever learned the job the hard way.
More with more
The honest door is Hassabis’s math, run forward. Your people get three or four times faster. Door one: keep the same output, fire most of them, tell everyone AI did it. Door two: do three or four times more. More products. More science. More markets you couldn’t afford to touch when humans were the bottleneck. Door one is what a frightened company picks. Door two is what an ambitious one picks, and the evidence says door two is where the money actually is. This is the same argument I keep making about where the AI spending spree is actually heading: the tool is neutral, the choice is not.
The augmentation numbers aren’t shy. Brynjolfsson, Li, and Raymond studied 5,179 support agents: AI lifted productivity 14% on average, and 34% for the novices. The European Investment Bank, across 12,000-plus firms, found AI raised labor productivity 4% with no short-run job losses. PwC found the most AI-exposed industries grew revenue per worker three times faster and paid a 56% premium for AI skills.
The gains go to the people, not the firings. Some companies already walked through door two while everyone else was firing. IBM said it would triple its US entry-level hiring, specifically for the software, security, and AI-engineering roles everyone else swears are being erased. Marc Benioff hired 1,000 new graduates to build the AI itself. People get faster, so do more, so hire more people to do it. More with more. It isn’t complicated. It’s just braver than a layoff.
What a board should actually do
Stop asking how many heads AI lets you cut. That question has already been answered dishonestly by everyone from Meta to a bank CEO who called his own staff “lower-value human capital” and then apologized when the room went quiet. He knew exactly how it sounded. He said it anyway.
Ask the three questions instead. Which tasks can AI actually automate today (the honest answer is fewer than the slide claims). Which skills do we now need more of. Which ambitions just became possible because our people got faster.
One rule above the rest. When you cut, name the real reason. If you overhired, say overhired. If growth stalled, say it stalled. If you’re firing people to make a $145 billion bet look responsible to analysts, then have the spine to say that, instead of telling a 26-year-old that a chatbot which can do 2.5% of a freelancer’s job somehow did all of hers. Calling overcapacity “AI” isn’t strategy. It’s gaslighting the exact people you need to adopt the tool, and nothing kills a rollout faster than convincing your own staff the thing is a guillotine with their name on the blade.
Jensen Huang handed every CEO a mirror this spring. The builders look in and reach for door two. The cost-cutters look in and see a man who ran out of ideas and found a word to hide behind.
The future belongs to the boards doing more with more. The rest are firing the future to pay for a bet they can’t explain, praying the music holds out longer than the depreciation. It won’t.
