Meta’s AI-Driven Restructuring Plan Put on Hold Due to Code and Employee Issues

Original author: Claude, Shenchao TechFlow

Shenchao Summary: A Reuters investigation published on August 26 revealed that Meta internally simulated an "AI-native" restructuring in early 2026, proposing to cut up to 60% of certain teams’ staff, with agents taking over routine tasks and smaller, high-density talent groups overseeing virtual employees. On the night of May 19, Zuckerberg halted the second wave of cuts originally scheduled for November. The next day, the company still laid off approximately 10% of its workforce. Internal data showed a 220% year-over-year increase in code changes, while only a 36% increase in features delivered to users; major technical and security incidents rose by 40% year-over-year, and employee satisfaction dropped from 74% to 55%.

Reuters published an investigative report on August 26: At the beginning of the year, Zuckerberg established an organizational transformation initiative codenamed Project OT (Organization Transformation) to replace a large number of routine human roles with AI. In the most aggressive scenario, many teams could be reduced to just 40% of their original size. The plan was scheduled to be implemented in two phases, in May and November. On the night of May 19, he halted planning for the second phase.

Reuters reviewed dozens of internal documents, posts, and recordings, and spoke with more than twenty informed individuals. Meta acknowledged the existence of the year-long project but characterized it as a "scenario planning" exercise, emphasizing that it did not aim to cut 60% of the company’s workforce and that "not every scenario was pursued."

In January 2026, Zuckerberg and his core team held an annual leadership offsite at their Hawaiian estate. According to Reuters, citing sources familiar with the matter, the meeting set a strategic direction: to transform Meta into an “AI-native” company. An internal planning document stated that AI would take over large volumes of routine work performed by thousands of employees, with virtual workers overseen by smaller, more “talent-dense” human teams within the company.

In the scenario simulation, up to 60% of multiple teams could be cut. Some employees will be transferred to new units, while others will leave. A human resources executive noted in another internal document that the scale of this round of reductions could match or even exceed the company’s previous layoff of approximately 25% about three years ago. The restructuring plan will occur in two phases: the first wave in May, followed by a second in November, during which vacant positions will be eliminated and underperforming employees will be removed.

These scale details had not been previously reported. Meta’s written response to Reuters was clear: the most severe scenario did involve reducing certain teams by up to 60%, with paths including both layoffs and reassignments; the company never intended to cut 60% of the entire group, and several key business units were not included. The official statement read: “As part of the company’s restructuring earlier this year, we asked certain teams to conduct scenario planning to assess the potential impacts of reassignments, closing open positions, and layoffs.” “This ultimately led to thousands of employees transitioning to newly formed teams working on priority initiatives—a development that has been publicly reported. We did not implement every scenario in the planning, nor did we assume all would be enacted.”

Reuters was unable to confirm a single trigger for Zuckerberg's change of heart that night, and the company declined to arrange an interview with him.

After ChatGPT launched at the end of 2022, Silicon Valley executives became particularly fascinated by “agents”: they shouldn’t just answer questions, but should be able to place orders, book flights, and write applications themselves. Internally, Meta codified this philosophy as “AI-native”—one sentence from a Project OT document reads: “AI-ready tools and agents collaborate, workflows are automated, and new projects default to AI-first.” Externally, the company is also preparing to sell business agents capable of scheduling and closing deals to other enterprises, with a business version already launched in June.

Product lead Naomi Gleit spent a "considerable amount of time" at the Singapore office last year. In June, she told Reuters that practices there "inspired teams in California and New York," with many changes being "bottom-up." Chief Data Officer Alex Schultz also visited Asia last year; according to three informed sources, the team greatly admired how local startups were reorganizing their structures around AI.

An earlier pilot took place in July 2025. Ime Archibong, long-time Vice President of Product Management, announced in an internal post that his team would first form five "small technical squads," each consisting of two to three engineers and one designer, equipped with AI tools, replacing the previous six-month product cycles with four-week sprints. He used basketball as a metaphor: "In basketball, embracing fast breaks lets you take more shots—and better ones. We have the same expectation for AI tools: they enable us to explore more ideas at lower cost and higher fidelity." The next sentence read: "This isn’t just more fun—in an AI-first era, we believe this is the winning strategy."

In October 2025, someone on his team posted the “Native AI Handbook,” providing other teams with guidance on “how to cross over”: traditional roles of product designers and engineers disappeared, and team members were uniformly renamed “builders”; middle management was eliminated; teams reported directly to a single senior unit lead; and daily priorities were assigned by “agent-assisted analysis.”

At the beginning of this year, Zuckerberg pushed for the implementation of Project OT and demanded that management restructure. By June, at least 11 units—including engineering and research—had been converted into team-based structures. An internal memo stated that performance ratings and promotions would now follow the “village approach”: unit leaders overseeing 30 to 50 people would make the final decisions, with HR and an anonymous “AI system” providing support; team leads would handle day-to-day operations but hold no formal managerial authority. One employee assigned to lead a team wrote on an internal message board: “I haven’t undergone management training, nor do I have access to rating tools or management resources.” Meta emphasized to Reuters that ratings and promotions have “always been and continue to be human decisions, not AI-driven.”

Traditional product teams typically consist of 10 to 20 people: seven to fourteen engineers, plus one each in product, design, data science, experience research, and data engineering. The AI-native team in the design mockup has only three to five members. Human resources is simultaneously seeking “irreplaceable talent” and “10x performers”—theoretically, one person augmented by AI can match ten, and the savings from layoffs are being used to increase their salaries.

The central axis of the plan is autonomous agents. Internal data has not supported this axis.

Chief Technology Officer Andrew Bosworth wrote in an internal post in early June: Code changes in the internal software platforms and infrastructure used daily by employees increased by 220% year-over-year; changes that actually became new features or upgrades delivered to Meta users increased by only 36% year-over-year. The infrastructure team had warned as early as March that the surge in AI-generated code posed a “reliability alert.” Another internal post in April put it more bluntly: Unconstrained agents are performing “large-scale, destructive actions that humans would be unlikely to carry out.”

The results were posted in the same internal thread. Major technical and security incidents (such as service outages and potential data breaches) increased by 40% year-over-year, and employees spent 70% more time on "firefighting" compared to the same period last year. In early June, attackers exploited Meta’s AI customer service bot to gain access to a batch of high-profile Instagram accounts, with Reuters specifically naming the previously deactivated Obama White House account. Meta declined to comment on these internal incident figures.

At an internal town hall in early July, Zuckerberg admitted the timeline was off. In a recording heard by Reuters, he said, “Over the past four months, the trajectory of agent development has not accelerated as we expected,” and the company’s bet on its new architecture “has not yet paid off.” He also expressed hope for more noticeable returns over the next three to six months. At the same event, he acknowledged that the restructuring was “not clean enough,” and that during planning in January and February, senior leadership’s concern was that “we weren’t adapting quickly enough.”

Investors are asking about something else. Meta has provided a capital expenditure guidance of $130 billion to $145 billion for this year, primarily for AI chips and infrastructure. Reuters’ report states “at least $130 billion” and cites LSEG estimates suggesting this spending could consume the company’s operating cash flow in 2026. The second-quarter earnings report subsequently confirmed the pressure: free cash flow declined by 91% year-over-year.

When the axe fell on May 20, employees no longer believed that "transformation" was merely about switching tools. Reuters first reported on March 13 that the company was discussing layoffs of 20% or more, and later in April detailed the first wave of approximately 10% in May and potential additional cuts within the year. Insiders told Reuters that the reporting disrupted the management’s original communication timeline.

Alongside layoffs, a large number of engineers were reassigned to new teams focused on AI engineering and other initiatives to work on training data. Earlier Reuters reports stated that in May, approximately 10% of global staff—around 8,000 people—were laid off, while another 7,000 were transferred to new teams related to AI workflows. By the end of the second quarter, the company had approximately 75,500 employees. Some employees forcibly reassigned internally referred to themselves as "coolies." On U.S. devices, employees were required to install keylogging and mouse-tracking software to collect keystrokes and mouse movements for training AI agents capable of mimicking human computer operations. Employees expressed frustration and posted sarcastic jokes on the internal Workplace platform; some replied to executives’ posts with elephant images, alluding to the unspoken issue of layoffs as “the elephant in the room.” Earlier reports indicated that over 1,600 employees had signed a petition opposing surveillance.

In the semi-annual Pulse survey, employee sentiment dropped from 74% to 55%. Labor organizing efforts are also intensifying. On the night of May 19, the second wave of layoffs was halted; the first wave proceeded as planned the next day. Zuckerberg subsequently told those who remained that he “does not expect any further company-wide layoffs this year,” hoping to provide more “stability.” The company later paused mouse and keyboard tracking, allowed some individuals who had been removed to return to their original teams, and increased budgets for travel, team-building, and snacks. Employees closely scrutinized in internal communications the two qualifiers he repeatedly used—“company-wide” and “this year”—speculating that headcount reductions would continue through team-level cuts or performance-based eliminations, or that the next round would be pushed to next year.

In June, Zuckerberg posted internally to refocus the external narrative on “betting on people”: “We are the only major company that prioritizes empowering individuals and putting this new technology into the hands of billions of users, rather than primarily automating work.”

The Reuters draft placed this statement after the plan had already been scaled back. Currently, Meta's official stance on this survey remains: "The scenarios have not been fully implemented."