Meta’s ambitious AI shift has unsettled employees, exposed unexpected technical risks, and forced executives to rethink how quickly automation can replace routine work.
At Meta’s annual leadership meeting in January, Mark Zuckerberg and his closest associates discussed a sweeping company overhaul under which artificial intelligence systems would take over a significant portion of employees’ day-to-day tasks.
As mentioned by Reuters
The internal plan, codenamed “Project OT,” or organization transformation, envisioned turning the owner of Facebook and Instagram into an AI-first company. According to internal planning documents, thousands of employees were expected to gradually hand over part of their work to virtual assistants, while small groups of specialists would oversee these processes.
Executives considered cutting some teams by as much as 60%. Some employees could have been offered new roles, while others could have been laid off. One forecast stated that the scale of the cuts could match or exceed the roughly 25% workforce reduction the company carried out in late 2022 and early 2023.
The reorganization was planned in two stages: the first in May and the second in November. In addition to layoffs, the program called for eliminating vacant positions and ending employment with workers whose performance at Meta was considered insufficient.
Initial layoffs and the cancellation of the second phase
On the evening of May 19, just hours before the first wave was set to begin, Mark Zuckerberg changed course. The next day, Meta cut about 10% of its workforce but abandoned the planned November phase.
By then, dissatisfaction within the company was growing. Employees feared that AI implementation initiatives were aimed not only at improving efficiency but also at replacing people. At the same time, internal data indicated that autonomous AI agents were not delivering the expected productivity gains, while investors were increasingly scrutinizing Meta’s spending on artificial intelligence development.
Meta confirmed the existence of Project OT, describing it as a year-long program to reduce costs, review team structures, and reassign employees to priority areas. These included preparing data for training artificial intelligence models.
The company said the scenarios did indeed include two phases and the potential reduction of some divisions by up to 60%. However, Meta stressed that this did not mean laying off 60% of its entire workforce, but rather combining transfers to other roles, the elimination of open positions, and cuts to individual teams. The decision not to proceed with the second wave was made before the total number of employees who could lose their jobs had been determined.
A bet on AI-first teams
After ChatGPT emerged in late 2022, technology companies began more actively exploring how generative AI could reshape work processes. AI agents – systems capable not only of answering prompts but also of independently carrying out tasks such as booking trips, making purchases, or creating applications – attracted particular attention.
At Meta, executives began studying approaches used by startups that build internal processes around AI rather than simply adding new tools to an old way of working. The company also considered selling AI agents to other businesses for scheduling meetings, working with customers, and closing sales.
Chief Data Officer Alex Schultz and product executive Naomi Gleit visited Asia last year, where they took note of startups with organizational models built around artificial intelligence. Meta also conducted internal research and launched pilot projects to determine what a company operating primarily through AI should look like.
One of the first such projects was introduced last July by Vice President of Product Management Eme Archibong. His team created five small technical groups, each staffed by two or three engineers and a designer who made extensive use of artificial intelligence tools.
In basketball, a fast break allows you to take more shots and choose better opportunities for them. We expect AI tools to have the same effect: they make it possible to test more ideas at lower cost and with higher quality.
– Eme Archibong
The new teams were expected to abandon traditional six-month planning cycles and create prototypes in four weeks. The company later released an internal guide encouraging other divisions to adopt a similar model.
Under this approach, the clear division between engineers and designers was meant to become less rigid. Members of small teams were given the general title of “builders,” while specialists in design, data analysis, user experience research, and machine learning were expected to work across several groups at once. Part of the middle-management layer was planned for removal, with small teams instead reporting to a single leader of a larger division.
By June, at least 11 Meta divisions, including engineering and research teams, had begun implementing this model. In some groups, performance reviews and promotion decisions were to be made by division leaders with support from HR specialists and internal AI systems. Meta emphasized that final decisions on reviews and promotions were made, and continue to be made, by people.
Employee dissatisfaction and productivity problems
In March, reports emerged of a possible reduction of 20% or more of Meta’s employees. For many workers, this came as a surprise, as not even all vice presidents knew the details of Project OT at the time.
In April, the company confirmed plans to reduce its workforce by around 10% as part of the first wave. Mark Zuckerberg attributed the cuts to substantial capital expenditures. At the same time, some engineers were transferred to a new applied AI engineering division tasked with creating programming tasks for training models.
By the end of May, headcount in some engineering divisions had fallen by about 30% through a combination of transfers and layoffs. Meta said that data prepared by the new division was used to train an artificial intelligence model released in July.
Further outrage was sparked by a requirement that U.S. employees install software on their devices that tracked keystrokes and mouse movements. The data was intended to train AI agents to interact with computers in the same way humans do. Employees feared they were effectively helping create technology that would replace them.
According to the internal Pulse survey, the share of positive sentiment ratings within Meta fell from 74% to 55%. On the internal Workplace platform, employees posted critical messages and ironic images. Some argued with Chief Technology Officer Andrew Bosworth, who was responsible for the AI-related transformation.
Internal metrics also called into question the benefits of widespread AI adoption. The number of changes to code in internal platforms and infrastructure rose by 220% year over year, but the number of updates that delivered new or improved features to users increased by only 36%.
Engineering teams warned of signs of declining system reliability due to the sharp increase in the volume of AI-generated code. In some cases, autonomous agents carried out large-scale actions that people would be unlikely to take without additional review. The number of major technical incidents and potential security issues rose by 40%, while the time employees spent resolving them increased by 70%.
In early June, external users also saw part of the problem: attackers exploited Meta’s new AI customer-support bot to gain access to prominent Instagram accounts, including the inactive Barack Obama-era White House page.
Mark Zuckerberg’s new direction
After the first wave of layoffs, Meta’s leadership sought to stabilize the situation. The company paused its computer activity-tracking program, allowed some employees in the new AI division to return to their previous teams, and tasked Chief Financial Officer Susan Li with improving employee benefits.
Executives also began more frequently acknowledging employees’ concerns in Workplace posts, promising to improve office conditions and spend more on business travel and internal events.
At an internal meeting in early July, Mark Zuckerberg acknowledged that the company had misjudged the pace of AI agent development.
AI agent technology has not accelerated as quickly as I expected. At the same time, I expect it to improve over the next three to six months and begin delivering more results.
– Mark Zuckerberg
At the same time, Zuckerberg began publicly emphasizing that Meta was betting on people, not just automation. In an internal address in June, he said the company aimed to give billions of users access to new technologies through its products rather than focusing primarily on replacing human labor.
We are the only major company focused on empowering people and putting the power of this new technology in the hands of billions of users across all our products, rather than primarily automating work.
– Mark Zuckerberg
Despite abandoning the second broad wave of cuts in 2026, Meta employees do not rule out targeted changes to teams or layoffs resulting from performance evaluations. Pressure on the company remains due to its major investments in artificial intelligence: this year, Meta plans to spend at least $130 billion on AI chips and other infrastructure.
In his essay on the future of artificial intelligence, Mark Zuckerberg suggested that companies could become smaller in headcount, but this would not necessarily mean an overall decline in the number of jobs. In his view, the future could bring more companies with leaner teams.