Mark Zuckerberg spent the spring of 2026 trying to prove his own workforce was obsolete — and it cost real people their jobs before he quietly admitted the machines were not ready to take over. The experiment had a name, a surveillance apparatus pointed at his own employees’ fingertips, and a target list. Project OT was a blueprint to shrink entire teams by as much as 60 percent, replacing thousands of roles with small pods of people propped up by AI agents. Then the software got its hands on real work and made the kind of mess no human manager would survive.
This is not a story about a product launch. It is the story of what happens when a man becomes so convinced of his own thesis that he discards the people who built his company before that thesis has passed a single real test. The first wave of layoffs landed in May. A second wave was scrapped only after the workforce revolted and the numbers turned ugly. The machines were supposed to be the future. Instead, they became the reason Zuckerberg had to walk the whole thing back. And every worker in the country watching this unfold just got a preview of the logic their own boss is quietly running through.
Mark Zuckerberg — You Don’t Bet the Company on Workers You Haven’t Finished Building
To build the agents meant to replace his staff, Meta first needed to learn how humans actually did the work. So it pointed its own surveillance inward. The company’s tracking software was set to record U.S. employees’ mouse movements, clicks, and keystrokes — even periodic screenshots — to feed the very AI that would one day do their jobs. The pitch to the workforce had the generosity of a redundancy notice: keep doing your daily work, and you’ll help train the thing that renders you unnecessary. Meta insists the data would not be used to evaluate employees, only to teach machines to click buttons and navigate menus. The people being recorded were not reassured.
European employees were largely spared that particular indignity, because logging a worker’s every keystroke runs headlong into privacy law there. The fact that the company had to draw a border around its surveillance tells you how invasive the program actually was. This was never about productivity. It was about harvesting the motions of human labor so the machines could copy them — and then make the humans redundant.
The numbers that leaked out of Meta did not look like progress. Code output soared, but the improvements users could actually feel stayed flat. Technical and security incidents jumped 40 percent, and the time workers spent cleaning up those fires rose 70 percent. Zuckerberg’s army of agents was producing more code and more breakage in equal measure. The genie didn’t just escape the bottle. It redesigned the bottle and then dropped it.
The Workers — Ten Percent of You Were Laid Off So a Chatbot Could Learn to Click a Mouse
Hours before the first wave of cuts, Zuckerberg paused the plan. Then the next morning, roughly ten percent of the workforce was still shown the door, and the second wave scheduled for November was shelved. By then the damage to the people who remained was already done. Internal sentiment collapsed from 74 percent favorable to 55 percent, employees described an open revolt, and labor organizing picked up steam. One worker slated to become a pod lead told reporters he was given no manager training, no access to ratings, and no tools for a job that suddenly carried authority over other human beings. Organizational charts don’t build the next breakthrough. People do — and these people had been asked to hand over every keystroke to a machine being trained to erase them.
Meta even built an HR tool to hunt for what it called “Irreplaceable Talent,” a phrase with a chilling subtext: everyone else, by definition, was replaceable. The ideal was a 10X performer. The ordinary employee was inventory. The message landed exactly as cruelly as it reads.
The Machines — A Sweeping, Disruptive Mess No Human Would Have Made
The agents Meta built to replace its workers turned out to be the most expensive proof that the work matters. According to Ars Technica, the software took sweeping actions no careful human would attempt — the polite way of saying the machines broke things a cautious person never would. A human engineer hesitates before mass-editing production systems. An agent, optimized to act, does not. That gap is the entire difference between a tool and a liability.
At a company town hall, Zuckerberg finally said the quiet part out loud: the bet on AI agents had not paid off, and the transformation still had not delivered. It was the closest thing to a public confession a chief executive can make without using the words “I was wrong.” The man who promised a future of abundant automation had overestimated the machines and underestimated the people he was trying to automate away. You can lay people off on a belief. You cannot run a company on one.
The Reckoning — “Betting on People” Is a Nicer Way to Say the Robots Weren’t Ready
Now Meta is running ads declaring it is “betting on people,” while still pouring an estimated $130 billion into AI infrastructure and keeping some of the AI-driven pod teams in place. The pivot reads less like a change of heart than a change of slogan. The workers who survived May are the same workers whose keystrokes were logged, whose morale cratered, and who now watch the company spend fortunes on the very agents that were supposed to replace them. Those thousands of laid-off employees are still updating their LinkedIn profiles, and a rebranded ad campaign does not put groceries on their tables this week.
The lesson here is not that AI cannot do the work. It is that a leader can become so sure of his own story that he surveils and discards his people before the story has earned a shred of evidence. For everyone else, the warning is simpler and colder: your boss is watching the same demo reel, and waiting for the model that finally gets it right. The containment systems worked right up until they didn’t.
Sources: Ars Technica, Reuters via Newser, Economic Times.