Meta CEO Mark Zuckerberg ran a secret project to make the company ‘AI native’; asked teams to study impact of…
Mark Zuckerberg spent the first half of 2026 quietly redesigning Meta around a single bet: that artificial intelligence could absorb the daily work of thousands of its employees. The plan had a code name, Project OT, short for Organization Transformation, and it was drawn up in January at the chief executive’s Hawaii compound during Meta’s annual leadership retreat. Reuters, which reviewed internal planning documents and spoke to more than 20 people familiar with the company, reported that executives ran scenario exercises examining what would happen if many teams across Meta shrank by as much as 60%. What remained would be small, “talent-dense” groups of humans overseeing virtual workers.The restructuring was meant to arrive in two waves, one in May and another in November. On the night of May 19, hours before the first wave landed, Zuckerberg pulled the plug on the second. Meta went ahead the next morning with a 10% cut, roughly 8,000 jobs, and then stopped. Four months later, the company is doing something odder: asking engineers in its new Applied AI division whether they would like to be managers again.
Meta’s Project OT began at a leadership retreat and never reached its second act
Meta has since confirmed Project OT, describing it as a year-long effort focused on cost cutting, redesigning team structures and shifting staff into priority work such as producing training data for its AI models. It also accepted that the most drastic scenarios involved reducing some teams by up to 60%, while insisting it never intended to cut 60% of the entire workforce and that the exercise blended redeployments with layoffs. The November wave was cancelled before leaders had worked out how many people would lose their jobs.
Mark Zuckerberg’s AI native playbook wanted builders and pods, not managers
The blueprint was unusually specific. Product teams of 10 to 20 people would give way to pods of three to five. Engineers, designers and researchers would share one generic title, “builder”. Middle management layers would disappear, leaving unit heads responsible for 30 to 50 people and pod leads running daily work with no formal authority. An early pilot under product management vice president Ime Archibong created five small pods that built prototypes in four-week sprints rather than six-month planning cycles. Meta also built an HR tool to identify irreplaceable talent, the kind of engineer Silicon Valley likes to call a 10X performer, and planned to spend part of its layoff savings on keeping them.
Employee anger and Meta’s own internal data slowed the AI transformation
Staff revolted once tracking software began capturing keystrokes and mouse clicks on US laptops to teach AI agents how humans use computers. More than 1,000 employees signed a petition against it, flyers went up in offices, and internal posts filled with elephant images to mark the layoffs nobody would discuss. Meta’s half-year Pulse survey showed employee sentiment falling from 74% favourable to 55%. The technology was struggling too. Code changes to internal systems jumped 220% year on year, yet changes that reached actual users rose only 36%. Major technical and security incidents climbed 40%, and time spent firefighting them rose 70%. In June, attackers exploited Meta’s AI customer support bot to reach high-profile Instagram accounts, including the dormant Obama White House page.
Meta is now rebuilding the management layer it spent a year removing
Around 7,000 employees were moved into Applied AI this year, some of them former managers handed individual contributor roles. That is the group now being offered a voluntary route back into management. Meta closed the second quarter with 75,472 employees, down 3%, on revenue of $60.8 billion and expenses up 55% to $42 billion, with at least $130 billion earmarked for AI infrastructure this year. At a July town hall, Zuckerberg conceded the reorganisation had been mistimed and that agent technology had not moved as fast as he expected. The bet was that AI could run the work with fewer people watching over it. Meta is discovering how many people that still takes.
