Meta’s AI Plan Hits Snags as Large Scale Actions by Agents Disrupt Operations
Earlier this year, Meta initiated a plan to reduce team headcounts by up to 60 percent to transform the company into an “AI native” entity. Reuters reported on the challenges Meta faced in implementing this plan, which was reportedly codenamed Project OT (organization transformation).
According to Reuters, Meta asked some teams to conduct a scenario planning exercise that explored reducing headcounts by 60 percent and resulted in the relocation of thousands of employees to newly established teams. While Meta confirmed the existence of the plan, it stated that not all scenarios were implemented and that the plan was never presumed to be fully executed.
Meta CEO Mark Zuckerberg directed the push for Project OT earlier this year, Reuters noted. The first round of layoffs occurred in May, with the second round being canceled. Reuters based its report on internal documents, posts, and recordings, as well as interviews with more than 20 people familiar with the company’s inner workings.
Project OT aimed to use AI to handle much of the daily work previously performed by thousands of human employees. Small teams would oversee the AI, with one internal document suggesting that Meta could reduce its headcount by about 25 percent or more. However, Meta canceled the plan before determining the exact number of layoffs.
To support the plan, Meta intended to reallocate savings from layoffs to high performing employees, particularly those with AI engineering skills. The company also launched a pilot program in engineering, research, and at least eight other teams, restructuring them into smaller groups. An internal post called “AI Native Playbook” outlined how the pilot would remove middle management and use “agent assisted analysis” to prioritize daily tasks.
While Meta envisioned AI agents improving productivity and decision making, internal documents revealed concerns. For instance, code changes made to the internal software platforms and infrastructure increased by 220 percent year over year, but changes leading to new or upgraded features for users were only up 36 percent. Moreover, AI agents reportedly made “large scale, disruptive actions” that humans were unlikely to execute, leading to a 40 percent increase in major technical and security incidents. Employee time spent resolving these issues increased by up to 70 percent.
These internal concerns prompted Zuckerberg to cancel the November round of layoffs. Reuters noted that reports of impending layoffs and Meta’s tracking of employee keyboard and mouse input to train AI agents since paused had also dampened employee morale.
In July, Zuckerberg acknowledged during a company meeting that the trajectory of agentic development had not accelerated as expected over the past four months.
Meta’s experience underscores the challenges other organizations face in integrating AI. While some firms are exploring ways to leverage AI to save time, drive revenue, and reduce costs, Meta’s case highlights the risks of overzealous AI projects. The company’s struggles suggest that even eager organizations may struggle to fully replace human workloads with AI, emphasizing the importance of carefully considering the potential impacts of AI on both operations and employees.
Thinking about building an AI product?
Get in Touch