Inside Meta Platforms Inc., the initiative had a sweeping ambition: use artificial intelligence to replace thousands of employees. The plan, pushed by Chief Executive Mark Zuckerberg, was to shrink the company’s workforce dramatically as AI tools matured, letting models handle work that people had done. On Wednesday, Reuters published a special report describing how that plan collapsed, and the picture it paints is of an organization that overestimated its machines and underestimated its own mess.
The plan’s failure was not a single event but a slow unwinding, according to the report. AI tools deployed in key roles, including content review, customer support, and parts of engineering, underperformed the human teams they were meant to replace. Errors that humans had caught were missed, response times that had been predictable became variable, and the quality bar that Meta’s own systems were supposed to enforce slipped. The tools worked well enough in demonstrations, current and former employees told Reuters, and poorly enough in production that managers began quietly reinstating the humans.
The organizational damage compounded the technical failure. Meta ran layoffs and hiring in parallel, cutting roles in some departments while scrambling to fill others, and the chaos that resulted made the workforce harder to manage than before the experiment began. New York Magazine, in a parallel report, called the episode “Zuckerberg’s botched AI overhaul,” a verdict that captures how the company’s most aggressive bet on AI labor played out in public: loudly announced, incompletely executed, and eventually walked back.
The episode has its roots in the efficiency drive that reshaped Meta starting in 2023. Zuckerberg declared a “year of efficiency,” cut more than 20,000 jobs, and promised investors a leaner company that would do more with less. The AI-replacement plan was the logical extension of that promise, an attempt to show that the efficiency gains could compound as models improved. What the report describes is the gap between that thesis and the reality of running a company where much of the work is judgment, context, and institutional memory.
The lesson extends beyond Meta. For two years, the technology industry has debated how quickly AI would displace white-collar work, with forecasts ranging from gradual to cataclysmic. Meta’s experience offers a concrete data point at the high end of ambition: a company with essentially unlimited AI resources, a CEO willing to force the change, and a workforce already accustomed to restructuring still could not make wholesale replacement work. The bottleneck was not the models’ capability on isolated tasks, employees said, but their inability to absorb the tacit knowledge that organizations carry: how decisions actually get made, which exceptions matter, and what the company has learned from its own mistakes.
The collapse does not mean Meta has abandoned AI. The company continues to invest heavily in models, in its recommendation systems, and in the AI features that Zuckerberg has described as central to its future. What has changed is the target: instead of replacing employees wholesale, the company is now deploying AI alongside them, in tools that assist rather than substitute. The distinction may sound small, but it is the difference between a workforce strategy built on displacement and one built on augmentation, and Meta’s own experience has made the company’s executives careful about which they promise.
The episode will be studied as a case study in organizational change. Meta’s mistake, the report suggests, was treating AI adoption as a headcount problem rather than a capability problem: measuring success by how many roles could be eliminated instead of by whether the work actually improved. The teams that succeeded, current employees said, were those that gave people and models overlapping responsibilities, with humans reviewing AI output and AI handling the volume. The teams that failed were those that removed the humans first and discovered the models’ limits later.
The collapse has not changed Meta’s public commitment to AI. The company’s core businesses depend on recommendation systems that are among the largest in the world, and its advertising engine has been re-architected around AI for years. What has shifted is the framing: Zuckerberg now talks about AI as a tool for the people who use Meta’s apps, not as a substitute for the people who run the company. The difference matters for investors, who had been told the efficiency story would compound, and for employees, who spent two years wondering whether they were being trained out of their own jobs. The report suggests the workforce can now exhale, and that the next round of Meta’s AI story will be measured in products, not headcount.
For the industry, the report arrives at a moment when the AI-replacement narrative is being tested everywhere. Companies across sectors have announced productivity gains from AI, and some have tied workforce cuts to those gains. Meta’s experience suggests the claims should be examined closely, and that the companies most confident about replacing workers may be the ones least familiar with what their workers actually do. Zuckerberg’s plan is over, but the questions it raised, about what AI can replace, when, and at what cost, are only beginning to be answered.
