Meta Scraps Plan to Replace Workers with AI: Agents Wreaked Havoc

As Reuters reports, citing dozens of internal documents and more than two dozen sources in the know, in early 2026 Meta launched an initiative code-named Project OT (for organization transformation), intended to make the company “AI native.” Under certain scenarios, some teams would have been trimmed by as much as 60%, with the day-to-day work of thousands of employees handed over to AI agents, leaving small groups of people for oversight.

According to one HR executive, carrying out these scenarios would have shrunk the workforce by roughly 25% or more. Two waves of layoffs were planned: the first took place in May, while the second, slated for November, was called off by Mark Zuckerberg. Part of the money saved on the layoffs was to be channeled into retaining strong specialists, above all engineers with AI skills.

Meta confirmed the existence of a “scenario-planning exercise,” but stressed that not all of the scenarios were meant to be implemented, and that it never got as far as tallying a final number of layoffs. The company also insisted that decisions about promotions and performance reviews are made by people, not AI.

The Agents Got Out of Control

The most curious part lies in the reasons the whole idea was dropped. Internal postings indicated that the AI agents were taking “large-scale destructive actions that a human would be unlikely to undertake.” The upshot: a 40% year-over-year rise in the number of major technical and security-related incidents, while the amount of employee time spent cleaning up the aftermath grew by nearly 70%.

The promised boost to productivity also proved dubious. According to CTO Andrew Bosworth, the volume of changes to internal tools and infrastructure rose 220% year over year, yet the share of edits that translated into new or improved features for users climbed by just 36%. In other words, there was many times more code, and noticeably less benefit from it.

Sagging morale did nothing to help the cause either: rumors of impending layoffs were accompanied by reports that Meta was tracking employees’ keystrokes and mouse movements to train the agents (a program that has since been suspended).

In July, Zuckerberg himself acknowledged at a meeting that “the trajectory of agentic technology over at least the past four months hasn’t accelerated as expected.” Meta’s story is telling: even one of the companies most zealous about AI ran into the fact that replacing living workers with algorithms is far harder than it looked on paper.

Source: Ars Technica