How did 'Project OT,' Meta's attempt to replace a large number of employees with AI, end in failure?
Reuters has reported that Meta CEO Mark Zuckerberg had planned to implement a plan in 2026 to 'lay off and redeem a large number of employees and replace them with AI.' Reuters has summarized the origins of this plan, codenamed 'Project OT,' the backlash within the company when the plan was leaked, and the events leading up to its collapse.
Mark Zuckerberg had a bold plan to replace Meta staff with AI. Here's how it imploded. | Reuters
Since ChatGPT was released at the end of 2022, Silicon Valley executives have been focusing on the potential that generative AI can bring to the way we work, and in recent years, 'AI agents' that autonomously perform various tasks have been attracting attention. Recently, Meta's management team has also been inspired by the 'AI native' management philosophy, which involves integrating AI tools and AI agents to automate workflows and integrating AI into the core of processes.
In 2025, executives including Meta's Chief Data Officer Alex Schultz and Head of Product Naomi Gleight visited Asia and were impressed by how startups were building their organizations around AI. The executives commissioned their own research into the organizational structures of AI startups and launched a pilot project to assess the potential of 'AI native' companies, a source familiar with the company told Reuters.
Then, in January 2026, Zuckerberg and some of his close associates held a retreat at their villa in Hawaii to refine 'Project OT,' which aims to make Meta more AI-native. According to internal documents and sources obtained by Reuters, Project OT also considered scenarios in which AI would take over the work of thousands of employees, potentially reducing the size of many internal teams by up to 60%.
Project OT's organizational restructuring was planned to be implemented in two phases, with the first phase of layoffs scheduled for May 2026 and a second, larger-scale layoff in November. This would include not only simple dismissals but also the closure of vacant positions and reassignment to new departments, and one HR executive predicted that the scale of the layoffs would exceed 25%.
Around the same time as the planning for Project OT, Meta also launched a program to incorporate AI into product development. In July 2025, a pilot project led by Yemei Archiborn, Vice President of Product Management, created a 'small technology pod' consisting of 2-3 engineers and 1 designer, replacing the traditional team of 10-20 people. The goal was for all members to use AI tools to build product prototypes in a short period of time.
In an internal post titled 'AI-Native Playbook' published by Archiborn's team, it is stated that traditional roles such as product designer and engineer will be abolished in pods. Members will be given a new title called 'Builder,' and there will be a single managerial position called 'Pod Leader' to oversee everyone.
When Zuckerberg and others launched 'Project OT' in early 2026, they were instructed to move forward with changes to the management structure, and by June, at least 11 departments, including engineering and research teams, had adopted a pod system. However, it has been reported that pod leaders were not given formal management authority such as performance evaluations, and that they had not received sufficient manager training, which caused confusion among employees.
Meanwhile, in March, before many executives at the vice-president level were briefed on Project OT, Reuters reported that 'Meta is planning to lay off up to 20% of its employees.' This report caused considerable anxiety among many ordinary employees, but executives were unprepared for the backlash and had no choice but to refuse to talk to them.
Meta plans to lay off up to 20% of its employees due to rising AI costs - GIGAZINE
Then in April, Reuters reported with more details that 'approximately 8,000 people, or 10% of the total workforce, will be laid off in May 2025.' At this stage, Zuckerberg also officially announced the layoff plan, explaining to employees that the reductions were due to significant capital investments.
Meta, the company that operates Instagram and Facebook, is reportedly planning to lay off 8,000 employees, the first wave of a large-scale layoff plan for 2026 - GIGAZINE
While moving forward with its May layoff plan, Meta has reassigned some engineers to its 'Applied AI Engineering Department,' where they will be tasked with creating puzzles to be used as training data to improve the coding capabilities of AI models.
In late April, we also started an initiative to collect data on employees' mouse movements and keystrokes and use it as training data for AI.
Meta begins collecting employee mouse movements and keystrokes as AI training data - GIGAZINE
The initiative to train AI using employee data led to widespread suspicion among employees that they were being used to train AI that would replace them. Furthermore, frustration over the lack of transparency regarding the layoff plan resulted in a flood of angry and self-deprecating posts on Meta's internal communication tool, Workplace. As a result of these events, employee morale has declined, and in the semi-annual employee satisfaction survey, Meta's employee satisfaction rate plummeted from 74% to 55%.
Furthermore, other internal data indicated that the AI at the core of Project OT was not performing as expected. Internal documents noted that while the amount of code produced by employees using generative AI increased significantly, the actual increase in productivity is questionable. According to one internal post, code changes made to the internal software platform and infrastructure used by employees increased by 220% year-over-year, while changes leading to new features or upgrades increased by only 36%.
In March, the infrastructure team issued a 'reliability warning due to increased AI coding,' and in April, 'large-scale, disruptive activity by uncontrolled AI agents that would be unlikely to be carried out by humans' was reported. As a result of these issues, critical technical and security incidents, such as service disruptions and potential data breaches, increased by 40% year-over-year, and the time staff spent resolving problems increased by 70%.
Furthermore, in June, it was reported that hackers had successfully tricked Meta AI's support chatbot and stolen the Instagram accounts of celebrities. The compromised Instagram accounts included the White House account of former President Barack Obama and the account of a Space Force Chief Sergeant.
Hackers deceive Meta AI's support chatbot to steal celebrity Instagram accounts - GIGAZINE
Amid increasing internal pressure, Zuckerberg and his staff consulted with their aides just hours before implementing the May layoffs and decided to cancel the second round of layoffs planned for November. Although the May layoffs were carried out, the remaining employees were told that 'there are no further company-wide layoffs planned for this year.'
Furthermore, management has been working to restore employee morale, temporarily suspending the mouse tracking program for AI training and returning some employees from the newly formed Applied AI Engineering department to their original teams. In addition, Workplace has seen a series of empathetic posts from management, as well as promises to improve the quality of snacks in the office kitchenette and increase spending on business trips and social events.
Zuckerberg attended an internal meeting in July and acknowledged that the large-scale organizational restructuring was not going well, saying, 'At least the progress in AI agent development over the past four months has not accelerated as much as we had hoped.'
'The progress of AI agents hasn't been as fast as we'd hoped,' says Meta CEO Mark Zuckerberg - GIGAZINE
However, when Zuckerberg speaks to employees about layoffs, he uses phrases like 'company-wide' and 'this year.' As a result, some employees are concerned that individual layoffs may occur on a team-by-team basis, or that another large-scale layoff may take place in 2027.
Related Posts:
in AI, Posted by log1h_ik