Meta intended to reduce certain teams by 60% and substitute their tasks with AI.

      Meta dedicated the first half of this year to planning the replacement of much of its employees' roles with AI agents, but later retracted the most ambitious aspects of the initiative when the technology fell short. This scheme was code-named Project OT, short for Organization Transformation, and it considered reducing the size of many teams by up to 60%.

      Katie Paul reported on this plan for Reuters, drawing from numerous internal documents, posts, recordings, and interviews with over 20 individuals familiar with the company.

      Mark Zuckerberg and his executives devised this plan at his Hawaii residence in January, envisioning AI taking over daily tasks performed by thousands of employees. Smaller groups of what a planning document referred to as “talent-dense” personnel would oversee the virtual workers.

      The initiative was structured in two phases, with the first round expected in May and a second in November. Alongside layoffs, there would be the closing of open positions and the dismissal of employees considered poor performers by Meta.

      One human resources executive anticipated cuts that could equal or exceed the approximately 25% reduction Meta made three years ago, as indicated in an internal document.

      On the evening of May 19, just hours before the first round was set to begin, Zuckerberg called off the November phase. The following day, Meta went ahead with approximately 8,000 job cuts, around 10% of its workforce, while moving 7,000 employees into AI positions within the same week.

      Reuters was unable to determine what prompted Zuckerberg's change of heart.

      Meta has confirmed the existence of Project OT, describing it as a year-long initiative aimed at reducing costs, restructuring teams, and reallocating staff to key areas such as training data production. The company also validated the two-phase structure and the 60% reduction figure but clarified that this percentage applied to specific teams and not the entire company. Significant divisions were excluded from this consideration, and the second phase was cancelled before the number of anticipated layoffs was finalized.

      “This ultimately resulted in the reassignment of thousands of employees to focus on priority work in several newly-formed teams, as previously reported. Ultimately, not every scenario from the exercise was implemented, and it was never assumed we would proceed with all of them,” Meta stated.

      The concept stemmed from Meta executives’ analysis of how AI startups organize. Chief Data Officer Alex Schultz and Head of Product Naomi Gleit made trips to Asia, where several individuals noted their admiration for how these startups had structured their organizations around AI.

      Gleit explained to Reuters in June that she had spent “considerable time” in Meta's Singapore office, and the practices observed there had “inspired some teams in California and New York.” She remarked that many concepts emerged from bottom-up initiatives.

      Meta's own data indicated that the agents were generating quantity rather than quality. Code modifications to internal platforms and infrastructure surged by 220% year-over-year, according to a post from Chief Technology Officer Andrew Bosworth in June, while changes that produced new or enhanced features for users increased by 36%.

      This gap became problematic. An April post indicated that unregulated AI agents were executing “large-scale, disruptive actions that humans would likely not undertake.” Significant technical and security incidents, such as service disruptions and potential data breaches, rose by 40% compared to the previous year, with the time spent addressing these issues increasing by 70%.

      Infrastructure teams had already raised concerns about reliability by March. In June, hackers took advantage of Meta’s new AI customer support bot to access prominent Instagram accounts, including the inactive Obama White House page.

      Meta declined to comment on the internal data.

      Employees started to believe they were training their replacements, as engineers were reassigned to develop software puzzles for training data, a task many described as monotonous in internal communications. Additionally, tracking software was mandated on the devices of U.S. employees to monitor keystrokes and mouse movements, enabling agents to learn to use computers like humans.

      This program was paused in June. Since then, 26 employees have filed a lawsuit against the company, claiming its AI systems targeted workers on medical leave for layoffs.

      Staff responded to executive communications with pictures of elephants, and Meta's internal sentiment score declined from 74% favorable to 55% in its biannual Pulse survey. Efforts for labor organization gained momentum.

      An internal post referred to the AI-Native Playbook, which outlined the intended structure. Product teams consisting of 10 to 20 specialists were to be reduced to smaller pods of three to five. Engineers, designers, and product managers would share the title “builder,” and middle management levels would be eliminated, with “agent-assisted analysis” determining daily priorities.

      By June, at least 11 units had adopted the pod structure. Unit heads would manage groups of 30 to 50 employees with support from human resources and unspecified “AI systems.” Meta declined to clarify what this term meant but stated that performance evaluations and promotions were “and are made by people, not AI.”

      Zuckerberg