Meta Planned to Fire Thousands Until AI Agents Failed to Deliver

Meta spent the first half of 2026 developing a plan to replace sizeable amounts of employee work with AI agents and potentially reduce the size of some teams by as much as 60%.

Key points

• According to a June internal post from Chief Technology Officer Andrew Bosworth, code changes to Meta's internal platforms and infrastructure rose 220% year over year.
• Meta still went ahead with cuts the after day, eliminating around 8, 000 jobs, or roughly 10% of its workforce.
• Nevertheless, on the night of May 19, just hours before the first round began, Zuckerberg cancelled planning for the November phase.
• In June, hackers additionally exploited Meta's new AI customer-support bot to gain access to high-profile Instagram accounts, including the inactive Obama White House account.
• While labor organizing activity rose, the company's employee sentiment score fell from 74% favorable to 55% in its half-year Pulse survey.

Meanwhile, the internal project, code-named Project OT, short for Organization Transformation, was eventually scaled back after Meta's AI systems failed to deliver the improvements executives had projected.

Reuters documented the details after reviewing scores of internal documents, posts and recordings and speaking with more than 20 residents familiar with the firm. Meta Planned Two Waves of Restructuring.

Mark Zuckerberg and senior Meta executives developed the plan at his Hawaii compound in January.

Meanwhile, the idea was to let AI agents take over routine work performed by thousands of employees, leaving smaller groups of what an internal planning document described as "talent-dense" workers to supervise the AI systems. Project OT included two major restructuring waves.

Meanwhile, the first was planned for May, followed by another in November. Layoffs would be combined with closing open positions and removing employees Meta considered poor performers.

One HR executive estimated that the overall reduction could become as sizeable as, or larger than, the roughly 25% workforce reduction Meta carried out three years earlier.

Meta still went ahead with cuts the after day, eliminating around 8, 000 jobs, or roughly 10% of its workforce. During the same week, the firm moved concerning 7, 000 employees into AI-related roles. Reuters could not determine exactly why Zuckerberg cancelled the second round. Instagram is Limiting AI Slop on the Platform With New Restrictions. Meta Confirms Project OT. Meta acknowledged that Project OT existed.

For context, the firm described it as a year-long effort to lower costs, redesign team structures and move employees into priority areas, including the production of AI training data.

According to Meta additionally, planners considered reductions of up to 60%, but stated those scenarios applied only to certain teams and were never intended to cover 60% of the entire firm. A number of major business units were excluded.

Meta remarked the second restructuring wave was cancelled before executives had determined how numerous extra employees would have lost their jobs.

Meanwhile, the firm stated the project ultimately resulted in thousands of employees moving to newly created priority teams and that not every restructuring scenario under consideration was implemented. AI Agents Did Not Produce Much Useful Output.

Meta's internal performance data indicated a major gap between the amount of work AI agents generated and the amount that resulted in useful products.

Nevertheless, changes that actually reached users as new or improved features climbed only 36%.

Internal documents additionally indicated that the climbed AI activity created reliability and security challenges.

Notably, an April post remarked unsupervised AI agents were carrying out "large-scale, disruptive actions" that humans would be unlikely to perform.

While the amount of time employees spent dealing with those incidents climbed 70%, major technical and security incidents, including service disruptions and potential data leaks, rose 40% from the previous year. Infrastructure teams had reportedly raised reliability reservations as early as March.

In June, hackers additionally exploited Meta's new AI customer-support bot to gain access to high-profile Instagram accounts, including the inactive Obama White House account. Meta declined to comment on the internal performance figures. Employees Feared They Were Training Their Replacements.

Reservations additionally grew among Meta employees that they were helping build systems that could eventually replace them.

For context, the firm reassigned engineers to create software difficulties applied as AI training data. Some employees complained internally that the work was repetitive.

Meta additionally introduced tracking software on devices employed by U.S. Employees that recorded keyboard and mouse activity so AI agents could learn how humans operate computers. The firm paused that program in June.

Since then, 26 employees have sued Meta, alleging that its AI systems unfairly targeted workers who were on medical leave during the layoffs.

Meta employees additionally began responding to executive posts with elephant images as internal frustration climbed.

While labor organizing activity rose, the company's employee sentiment score fell from 74% favorable to 55% in its half-year Pulse survey. Meta Tested Much Smaller Teams.

Meanwhile, an internal document called the AI-Native Playbook outlined how Meta imagined its future workforce.

Traditional product teams containing 10 to 20 specialists could shrink into smaller groups of only three to five individuals.

Engineers, designers and product managers would additionally share a single job title: "builder."

Meta planned to remove layers of middle management and employ "agent-assisted analysis" to assist determine daily priorities.

By June, at least 11 units had adopted the smaller pod structure.

Under the model, unit leaders could manage 30 to 50 employees with backing from human-resources teams and unspecified "AI systems."

Meta declined to explain exactly what those AI systems would do but remarked decisions involving performance ratings and promotions were and remain produced by humans. This AI Search Engine Can Find Anything Remarked in Podcasts. Zuckerberg Admits AI Agents Progressed Gradually.

At a July firm town hall, Zuckerberg acknowledged that AI agents had not advanced as swiftly as Meta projected.

He informed employees that the "trajectory of the agentic development" during the previous four months had not accelerated in the way executives anticipated.

Nevertheless, Zuckerberg stated he still projected the technology to deliver significant benefits within three to six months.

In August, Bosworth additionally informed employees not to assume that AI productivity improvements should translate into extra time off. Meta's public messaging has since shifted.

While Zuckerberg lately published a 6, 500-word essay predicting that AI could create an abundance of jobs even if individual businesses employ fewer workers, the firm is now running emphasizing its investment in residents.

"Company sizes may shrink, just as they did in the transition from industrial giants to tech companies, " Zuckerberg wrote. Meta Still Spending Heavily on AI.

Despite scaling back Project OT, Meta has not reduced its investment in artificial intelligence.

Taken together, the developments around meta Planned to Fire Thousands Until AI Agents Failed to Deliver point to a situation that is still moving, and the coming days should bring more clarity.