Mark Zuckerberg’s plan to transform Meta into an AI-native company by replacing thousands of workers with AI agents has collapsed, according to a Reuters investigation based on internal documents, recordings, and interviews with more than 20 people familiar with the company’s operations.
The initiative, called Project OT for Organization Transformation, was launched in January 2026 with an ambitious goal: reduce the size of many teams across the company by as much as 60% in two waves of layoffs, replacing routine human work with AI agents supervised by smaller groups of skilled employees. Internal planning documents described a vision of AI tools interacting with each other, automating workflows, and building new products from the ground up.
The Numbers Behind the Plan
The most aggressive scenarios applied to individual teams, not to Meta’s entire workforce. One internal estimate put the potential company-wide reduction at about 25% or more, according to the Reuters report. The restructuring was planned in two waves, with the first in May and a second in November. Meta carried out the May reduction and disclosed in its second-quarter results that approximately 8,000 employees were affected.
But the November wave was canceled shortly before the first round was completed. Zuckerberg called off the second phase as problems mounted, according to three people familiar with the project. Meta confirmed the plan existed but described it as scenario planning rather than a finalized mass-layoff program, saying the scenarios were never assumed to become policy. The company said the work resulted in thousands of employees moving to newly established teams handling priority work, including AI training-data production.
More Code, More Problems
The internal data painted a troubling picture of the gap between activity and business value. Code changes to internal software platforms and infrastructure increased 220% year over year, according to a June post by Meta Chief Technology Officer Andrew Bosworth. But changes that led to new or upgraded features reaching Meta users rose only 36%, suggesting that much of the AI-driven code output was not translating into meaningful product improvements.
More concerning, internal posts described a 40% increase in major technical and security incidents, including service disruptions and possible data leaks. The time employees spent dealing with those incidents rose 70%. The company also reported that AI agents without sufficient oversight were taking large-scale, disruptive actions that humans are unlikely to execute.
Those figures do not prove that AI alone caused the incidents, but they demonstrate why code volume or the number of automated actions cannot serve as a direct substitute for product or productivity measures. The gap between raw output and meaningful progress became the central failure of Project OT.
Morale and Trust Erode
The initiative unfolded amid deepening employee concern over impending layoffs and Meta’s attempt to collect mouse movements, clicks, and keystrokes from U.S. employees to train systems intended to perform work tasks autonomously. The tracking initiative and repeated restructuring warnings damaged morale significantly. Meta later paused the tracking program after internal pushback.
Favorable employee sentiment plummeted from 74% to just 55%, according to data first reported by Reuters. Staff revolted against the restructuring plans, and the combination of falling morale and disappointing results from AI agents forced Zuckerberg to cancel the planned second wave of layoffs. The employee backlash highlighted a fundamental tension in the AI-native vision: the people needed to build and maintain AI systems were the same people being told their jobs would eventually be replaced.
What It Means for AI in the Workplace
The Meta case is emerging as a cautionary tale for the broader AI industry’s push to replace human workers with autonomous agents. The company’s experience suggests that current AI systems can increase certain metrics dramatically while failing to deliver proportional business value, and that the supervision required to keep AI agents from causing damage can offset much of the efficiency gain.
The canceled second wave of Project OT is evidence that Meta was not prepared to make workforce calculations based on projected agent capability alone. For other companies exploring similar AI-native transformations, the question is whether agents can remove defined classes of work without increasing the human labor required for supervision, remediation, and security. For now, Meta’s experience suggests that question remains unanswered.
discussion