Zuckerberg Halted Meta’s Second AI Layoff Wave After Productivity Problems Emerged
Thursday, 10 September 2026, 18:41
Inside Meta, an ambitious AI experiment exposed a widening gap between the amount of code produced and the value it delivered.
In January, Mark Zuckerberg and his closest advisers gathered for their annual retreat at the Meta chief’s estate in Hawaii. There, they developed a radical plan to restructure the company’s operations for the age of artificial intelligence.
The project was code-named “OT,” short for “organizational transformation.” It envisioned turning Meta, the company behind Facebook and Instagram, into an organization where artificial intelligence would underpin most work processes.
According to internal documents, AI was expected to perform a significant share of the daily tasks handled by thousands of employees. The work of virtual employees would be overseen by small, highly specialized teams.
During the planning process, executives considered reducing the size of many teams by 60%. Some employees could be transferred to new divisions, while others could be laid off. One human resources executive predicted that the scale of the cuts could match or even exceed the previous round of layoffs. Three years ago, Meta cut roughly a quarter of its workforce.
The restructuring was planned in two phases: the first was to begin in May, and the second in November. In addition to layoffs, the plan called for eliminating vacant positions and dismissing employees the company considered underperformers.
Artificial intelligence was expected to perform a significant share of the daily work done by thousands of people, while virtual employees would be overseen by smaller teams of specialists.
– Meta’s internal documents
However, on the evening of May 19, just hours before the first round of layoffs was set to begin, Zuckerberg changed course. The following day, Meta cut 10% of its workforce but abandoned plans for the November phase.
Meta confirmed that the “OT” project existed and described it as a yearlong program to cut costs, change team structures, and move employees into priority areas. These included preparing training data for artificial intelligence models.
The company acknowledged that some scenarios involved reducing teams by up to 60%, but stressed that this did not mean cutting 60% of its entire workforce. According to Meta, executives canceled the second phase before determining the total number of potential layoffs.
Earlier this year, as part of our restructuring, we asked certain teams to model the possible effects of transferring employees, eliminating vacant positions, and making cuts. As a result, thousands of employees moved into priority roles across several newly created teams. We did not implement every scenario considered during this analysis, nor did we ever assume that they would be carried out in full.
How Meta Built an “AI-First” Model
After ChatGPT was released in late 2022, technology executives began actively discussing the impact of generative AI on the future of work. Particular attention focused on so-called agentic systems – programs capable not only of responding to prompts but also of taking actions independently, such as purchasing goods, booking travel, or creating applications.
Meta executives became interested in an approach under which the company would not simply add AI to existing processes, but would build its operations around the technology from the outset. One document related to the “OT” project described automated workflows, AI-ready tools, and the development of new products with artificial intelligence as a priority.
Another area was to involve selling AI agents to other companies that could perform tasks such as scheduling meetings and handling sales.
Last year, data executive Alex Schultz and product executive Naomi Gleit visited countries across Asia. According to people familiar with the matter, they studied startups that had built their organizational structures around artificial intelligence from the beginning.
Gleit said she spent an extended period working from Meta’s Singapore office. According to her, local approaches inspired teams in California and New York. She also noted that some of the ideas emerged directly from employees who had begun changing their workflows on their own.
One of the first experiments was a project led by Ime Archibong. His team created five small technical groups, each consisting of two or three engineers and a designer. They were expected to abandon traditional six-month planning cycles and build prototypes in four-week sprints.
In basketball, a fast break allows you to take more shots and make them better. We expect the same effect from AI tools: they allow us to explore more ideas at lower cost and with greater precision.
In October, Archibong’s team presented a broader plan to expand the model. Traditional product designer and engineer roles could disappear, while members of the small groups would receive the universal title of “builders.” The number of middle-management layers was to be reduced, with the groups reporting directly to the heads of large divisions.
At the beginning of this year, Zuckerberg launched the “OT” project and instructed executives to change the management structure. By June, at least 11 divisions, including engineering and research teams, had adopted the small-group format.
One division called the model a “village approach” to managing personnel. Performance reviews and promotion decisions were to be made by division leaders with support from human resources specialists and unspecified “AI systems.” Each leader was expected to oversee between 30 and 50 employees, while small-group leaders would be responsible for daily operations without formal management authority.
Meta said its teams had experimented in different ways with increasing flexibility. The company also stressed that decisions about employee evaluations and promotions were made – and would continue to be made – by people, not artificial intelligence.
At the same time, Meta created a talent tool to identify “irreplaceable talent.” One document referred to a hypothetical “10-times productivity employee” – someone capable of doing the work of an entire team. The company planned to use some of the money saved through layoffs to offer large salaries to attract and retain leading AI engineers.
Employees Push Back Against the Transformation
In March, reports emerged of possible cuts that could affect 20% or more of Meta’s workforce. The news alarmed employees, although most of them were not yet supposed to know that layoffs were being planned.
Executives did not openly discuss the situation with rank-and-file employees. Instead, managers were advised to tell their teams that their roles would “evolve” under the influence of artificial intelligence.
In April, the company confirmed plans for a first round of layoffs affecting 10% of its workforce. Mark Zuckerberg attributed the cuts to significant capital expenditures.
At the same time, Meta was moving engineers into a new applied AI engineering division. Its task was to create software puzzles for training artificial intelligence models in coding skills. Some employees considered the work monotonous and boring.
As a result of transfers and layoffs, the number of employees in some engineering divisions had fallen by 30% by the end of May.
The company also installed software on employees’ devices in the United States that recorded keystrokes and computer-mouse movements. The data was intended to help AI agents replicate how people interact with computers.
Employees feared they were effectively training a system that would eventually replace them. On Meta’s internal communications platform, they posted angry messages and sarcastic jokes. Responses to executives featured images of elephants – as a symbol of a problem everyone knew about but no one discussed.
Some employees argued with Meta’s chief technology officer, Andrew Bosworth, who was responsible for the AI-driven transformation. Against the backdrop of the conflict, employees’ efforts to organize a union movement also gained momentum.
Morale within the company deteriorated sharply: according to the results of a half-year survey, employee loyalty to Meta fell from 74% to 55%.
AI Fell Short of Expectations
Meta’s internal data showed that greater use of AI did not guarantee a corresponding increase in productivity. The amount of code produced by employees with the help of artificial intelligence had increased significantly, but the practical results were much more modest.
At the beginning of June, Bosworth said that the number of changes to internal software platforms and infrastructure had risen 220% year over year. At the same time, the number of changes that resulted in new or improved features for Meta users had increased by only 36%.
Infrastructure teams had warned as early as March about reliability problems linked to the rapid growth in the volume of AI-generated code. Another internal post said that uncontrolled agents had carried out large-scale destructive actions that people would probably never have attempted.
The number of major technical and security incidents, including service outages and potential data leaks, rose 40% year over year. The time employees spent resolving incidents increased by 70%.
At the beginning of June, hackers exploited a vulnerability in Meta’s new AI-powered chatbot and gained access to verified Instagram accounts, including the inactive White House account of Barack Obama.
Meta declined to comment on internal data concerning AI-related failures.
Several hours before the first round of layoffs on May 20, Zuckerberg again discussed the situation with his closest advisers and canceled the November phase of the restructuring. The following morning, the company cut 10% of its workforce.
I do not expect any other companywide layoffs this year, and I want to provide employees with greater stability.
After the layoffs, executives attempted to rebuild employees’ trust. The company suspended its program for tracking mouse movements and keystrokes, allowed some engineers to return to their previous teams, and began publishing messages supporting its workforce.
In July, Zuckerberg said at an internal meeting that he had made mistakes in choosing the timing of the reorganization. According to him, AI-agent technology had developed more slowly than he expected. At the same time, the Meta chief predicted that it should prove more useful over the following three to six months.
“Betting on People”
After scaling back the most radical steps of its internal transformation, Meta began publicly positioning itself as a company focused on people. In an advertising campaign, it said it was betting on employees and users.
Zuckerberg also promoted a vision of the future in which artificial intelligence would be accessible to a broad range of people. He emphasized plans to make it easier to create AI agents and portrayed competitors as companies primarily focused on automating work.
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 labor.
At the same time, employees noted that Zuckerberg had spoken only about the absence of “companywide” layoffs “this year.” As a result, they suspected that Meta could continue cutting individual teams, dismissing people based on performance, or postponing major decisions until next year.
Pressure on the company was intensified by massive spending on artificial intelligence and investor scrutiny. This year, Meta plans to spend at least $130 billion on AI chips and other infrastructure. Analysts estimate that these expenses could nearly consume the company’s entire operating cash flow in 2026.
In his vision of the future of AI, Zuckerberg suggested that an excess of jobs could emerge in the future, even though individual companies might employ fewer people.
Companies may become smaller – just as happened during the transition from industrial giants to technology companies. But that does not mean an overall reduction in the number of jobs. Rather, it means there will be more companies, each employing fewer people.
Meta’s plan was intended to make artificial intelligence the foundation of a new model of work, but it encountered employee resistance, security problems, and weaker-than-expected results. Ultimately, the company carried out the first round of layoffs but was forced to abandon the most ambitious part of its planned restructuring.
- Leading companies reduced AI spending per employee in August, reflecting possible seasonal slowdown, completed investments, infrastructure efficiencies, or greater scrutiny of business value.
- OpenAI is urging the US Congress to establish mandatory national AI safety rules, including independent testing, cybersecurity requirements, and incident reporting for advanced systems.
