Mark Zuckerberg spent the better part of a year trying to prove that the people who build his company were the most replaceable thing inside it. He set loose an army of AI agents to do their jobs — write the code, run the systems, answer the tickets — and told his executives to plan for teams slashed by more than half. Then the machines did what machines do when you hand them the keys without the judgment: they started breaking things at a rate no human being ever would, and the exact workers Zuckerberg wanted gone turned out to be the only people left who could put the fire out.
The receipts are now public, laid out in internal documents reviewed by Reuters. Meta ran a year-long experiment it called, with the usual hubris, an “AI-native” operating model. The pitch was seductive and the arithmetic was brutal: agents would automate the daily grind, some workers would be redeployed, and the rest would be shown the door. One HR executive put the target at a 25 percent headcount reduction. Another plan went further and proposed cutting some teams by 60 percent. The robots were coming for the engineers, and the engineers were the ones being told to train them.
Mark Zuckerberg — He Bet the Company on an AI That Kept Setting It on Fire
Here is what happened when the agents actually went to work, straight from the internal reports. The AI made “large-scale, disruptive actions that humans are unlikely to execute” — which is corporate poetry for “it did things so dumb and so destructive that no sane person would have pressed the button.” Major technical and security incidents — service outages, possible data leaks, the kind of failures that send customers running — spiked 40 percent year over year. The engineers who survived the layoffs did not get to enjoy their victory. The time they spent firefighting the AI’s messes rose 70 percent. The machine that was supposed to replace the workers instead handed them a second full-time job.
Sit with that number for a second, because it is the whole story in a single statistic. Meta told its remaining staff they would be liberated from drudgery, freed to do the interesting work while the agents handled the grunt labor. What actually happened is that the survivors were conscripted into an endless night shift, chasing down the outages, leaks, and plain old breakage that the agents left in their wake. The promise was fewer hours and fewer workers. The reality was the same people doing 70 percent more damage control. The genie did not just escape the bottle; it set the bottle on fire and handed the humans a mop.
Project OT — The Plan to Cut Teams by 60% Died the Moment It Hit Reality
The grand strategy had a name, Project OT, and it was going to be the template for the entire industry: strip out the expensive humans, let the cheap machines carry the load, and pocket the difference. The internal metrics tell you how it really went. Code changes surged 220 percent year over year — the agents were churning out changes faster than anyone could review them. But changes that actually produced a new or improved feature for users rose only 36 percent. The agents were fantastically busy and barely productive, a perfect digital imitation of the worst middle manager you ever had: motion without meaning, activity without output. The AI-native future collapsed the moment it touched the real world.
And so Meta quietly did what every company does when a flagship experiment fails in public: it backed away and pretended it was never really committed in the first place. The 60 percent team-cut plan was scrapped. The company now says Project OT was always just a year-long exploration, a thought experiment, nothing to see here. That is the corporate equivalent of a gambler who, after losing the house, insists he was only ever doing research on card counting. The walk-back does not give the workers back the year of fear, and it does not un-publish the internal documents that prove the intent was real.
Meta’s Engineers — The People He Tried to Cut Are the Only Reason It Did Not Burn Down
Strip away the jargon and what remains is a story about who actually keeps a company alive. Zuckerberg looked at thousands of engineers — the people who built the systems that made him one of the richest men on Earth — and decided they were overhead to be optimized away. Then the replacement workers, the agents, immediately proved they could not be trusted to run a lemonade stand without setting the building on fire, and it was those same engineers, working 70 percent harder, who caught every outage and plugged every leak. The people he labeled replaceable were the only thing standing between Meta and a very public collapse. Organizational charts do not keep a platform online; people do.
The lesson is not that AI agents are useless. It is that the people who sell you the AI-native future always leave out the fine print: someone still has to fix what the machines break, and that someone is usually the person whose job the machines were supposed to take. Every company now copying the Meta playbook — and there are thousands, because every boardroom wants the same cost curve — is about to learn the same thing. You can replace the workers with agents. You cannot replace the judgment, and you cannot replace the human being who gets paged at 2 a.m. when the agent decides to do something disruptive at scale.
Here is what changes next for everyone who is not a Meta engineer. The great AI-layoff wave was supposed to be the end of the white-collar worker. What Meta’s own numbers suggest instead is that the wave may have already crested, because the technology keeps proving it is not ready to be left alone in the building. A company that cut too deep on faith in its agents now has fewer people doing more emergency work than ever — a recipe for burnout, attrition, and the quiet exodus of exactly the talent it will need when the next outage hits. The robots did not take the jobs. They made the jobs harder, and the people left holding the mop are the ones who were supposed to be gone by now.
Sources: Ars Technica, Computerworld.