Meta employees clash with leadership over AI restructuring

Coverage of Meta's AI reorganization has centered on layoffs, but reporting shows two distinct disruptions happening in parallel. In May 2026, Meta cut about 10% of its workforce, or roughly 8,000 jobs; Meta spokesperson Erica Sackin confirmed to NPR that affected employees had been notified. Separately, and on top of those cuts, NPR and The Guardian both report that Meta moved roughly 7,000 more employees into new AI-focused teams, including two new units, one for AI cloud infrastructure and one for an internal AI agent called Hatch, where transfers were mandatory ("Transfers aren't optional," per an internal post viewed by The Guardian). Many reassigned staff landed in Meta's Applied AI data-labeling group, which had grown to around 6,500 engineers and product managers by June, up from roughly 1,000 in an earlier wave, according to The Guardian and reporting from WIRED relayed by TechCrunch and ARY News. Employees describe the work, largely coding challenges and test cases used to train AI agents rather than product features, as repetitive; some call themselves "draftees." Zuckerberg has defended the approach internally by arguing that Meta employees are better suited to this data-labeling work than outside contractors, according to TechCrunch, even as Meta separately paid $14.3 billion for a stake in the contract data-labeling firm Scale AI.

Two episodes pushed the story into view. An employee hijacked a livestreamed internal presentation to deliver a profanity-laced demand that leadership be told what the workforce thought of them, reporting that WIRED broke and TechCrunch and ARY News both relayed. Separately, more than 1,600 employees signed a petition against software that logged mouse movements, clicks and keystrokes on company devices to generate AI training data, a program some staff nicknamed an "Employee Data Extraction Factory," per Reuters reporting carried by The Daily Star and The Hindu. Meta scaled the program back within weeks: an internal memo from Superintelligence Labs vice president Stephane Kasriel added the ability to pause tracking for up to 30 minutes or request a full exemption, plus battery and bandwidth changes after complaints the software was inflating employees' home internet bills. A Meta spokesperson declined to comment on the reversal.

Zuckerberg told employees internally that the company had "made mistakes" during the shift and does not expect further company-wide layoffs this year, an internal memo first reported by Reuters and picked up by Business Standard and other outlets. Chief Product Officer Chris Cox separately described the environment to staff as "difficult" and "brutal," comparing it to running a marathon through a hailstorm. Taken together, the swift reversal on device monitoring and two separate executives publicly acknowledging strain suggest Meta views employee trust as an active constraint on execution speed, not just a communications problem to manage after the fact.

Few companies can reroute thousands of their own engineers into AI data work instead of leaning entirely on vendors like Scale AI; Meta's choice to do both at once is a preview of a trade-off other AI-first reorganizations will face as they build agents that need real interaction data to learn workplace tasks. The speed of Meta's climbdown on monitoring software, weeks after visible resistance, suggests that consent and instrumentation design, not just data volume, are becoming a real constraint on how AI labs collect this kind of data, particularly given the EU regulatory exposure that Reuters and The Hindu both flagged. For teams elsewhere planning similar AI-first reorganizations, the more durable lesson is less about the 10 percent headline and more about the operational and trust cost of converting existing staff into AI infrastructure without their buy-in.