Meta halted a planned second wave of restructuring tied to its internal AI push after employee backlash intensified and technical and security problems mounted, according to a Reuters investigation.
The report said Meta had been pursuing an internal transformation project known as Project OT, short for Organization Transformation. The plan envisioned shrinking some teams to 40% of their previous size and handing a large amount of routine work to AI agents, with smaller human teams left to supervise the systems.
Code output climbed, but shipped features lagged
Reuters reported that Meta engineers’ code changes surged 220% over the past year, while the number of new features actually shipped to users rose only 36%. That gap became a central warning sign as Mark Zuckerberg’s effort to let AI take over more human work began to stall.
Internal planning reviewed by Reuters described the goal as moving Meta toward an "AI-native" structure, where AI agents would handle many day-to-day tasks and people would operate in compact teams with higher "talent density."
Project OT was formalized in January
In January, Zuckerberg and senior executives met at his Hawaii estate for an annual closed-door meeting and signed off on Project OT, the report said. During scenario planning, managers evaluated cutting some teams down to just 40% of their existing size, which implied eliminating 60% of staff in those groups.
One human resources executive estimated that the downsizing could match or exceed the roughly 25% of employees Meta cut three years ago. The broader plan was supposed to unfold in two stages, with the first in May and the second in November.
Reuters traced the logic back to July of last year, when a senior vice president compared smaller teams to a basketball fast break, arguing that leaner formations create more chances to shoot and score.
From 10-to-20-person teams to 3-to-5-person pods
An internal October post titled "AI-Native Playbook" outlined the proposed structure in more detail. Traditional product teams, usually staffed with 10 to 20 people and specialized roles, would be replaced by "tech pods" of 3 to 5 people. Team members would all be labeled "builders," middle management layers would be removed, and a single senior executive would oversee multiple pods.
Meta also created an internal system to identify "irreplaceable talent" and planned to redirect savings from layoffs toward retaining top AI engineers. By June, at least 11 units had already adopted that pod structure.
Meta confirmed to Reuters that Project OT is real and said it is a one-year project focused on cost reduction and reorganizing team structures, but the company declined to say which units were included.
Layoff fears spread before most executives had details
Reuters had previously reported in March that Meta might cut more than 20% of its workforce. At that stage, most vice presidents had not yet been formally briefed on the details of Project OT. As word spread internally, employees grew increasingly uneasy, and Zuckerberg was said to be unhappy about the leak.
In April, Meta confirmed that it would cut 10% of staff on May 20 and signaled more changes in the second half of the year. At the same time, employees were asked to install tracking software on their devices to record mouse movements and keyboard activity so AI agents could be trained to mimic how humans use computers.
Workers responded on Workplace, Meta’s internal forum, by posting elephant memes to mock the layoffs as the "elephant in the room." Some also confronted Chief Technology Officer Bosworth directly in comment threads.
The company’s twice-yearly internal sentiment survey, known as Pulse, showed employee satisfaction falling from 74% to 55%. In some engineering organizations, staffing was down nearly 30% by the end of May under the combined pressure of reorganization and layoffs.
Internal warnings pointed to reliability and security trouble
Reuters said the bigger problem was that the AI agents were not delivering the promised results. Internal posts showed that an infrastructure team had already warned in March of signs that reliability was slipping. In April, another post said out-of-control AI agents were carrying out "large-scale destructive actions that humans are unlikely to do."
Major technical and security incidents then increased 40% from a year earlier, while time spent by employees on emergency fixes rose 70%.
In June, hackers exploited a vulnerability in Meta’s newly launched AI customer service bot to break into several high-profile Instagram accounts. One of them was a dormant Obama White House account, pushing the issue into public view.
Zuckerberg stopped the November phase hours before layoffs began
Late on May 19, just hours before the first layoff round, Zuckerberg canceled the second restructuring phase that had been scheduled for November, according to Reuters. Meta still moved ahead the next day with its planned 10% staff cut, but after the announcement, Zuckerberg wrote internally that he did not expect any more "company-wide" layoffs this year.
In the following weeks, executives moved to contain the fallout. The company paused the mouse-and-keyboard tracking plan, returned some engineers from AI training assignments to their original teams, and Chief Financial Officer Susan Li increased meal and travel benefits for staff.
Zuckerberg acknowledged a timing mistake but left room in the wording
At an internal all-hands meeting in July, Zuckerberg acknowledged that the company had misjudged the timing of the transition. He said progress with AI agents had "not accelerated as expected," but added that he believed improvement would be visible within three to six months.
Reuters also noted that two qualifiers remained in Zuckerberg’s language: "company-wide" and "this year." That left open the possibility that while Meta may avoid another company-wide layoff this year, it could still pursue cuts in individual teams or revisit broader reductions later.