Project OT aimed to replace up to 60% of certain teams with AI, but disappointing tool performance and employee resistance forced Mark Zuckerberg to abandon the broader restructuring.
Mark Zuckerberg wanted to build the leanest version of Meta that AI could enable. Instead, he got a case study in how ambition without execution can backfire spectacularly.
Reuters journalist Katie Paul revealed the details of Meta’s internal initiative, dubbed Project OT (Organization Transformation), which aimed to transition the company to what leadership called an “AI-native” workforce. The plan envisioned cutting up to 60% of certain teams across two layoff waves, replacing human workers with AI tools operating under smaller human oversight groups. The second wave never happened.
What Project OT actually looked like #
Zuckerberg first laid out the vision at his Hawaii retreat in January 2026. The concept was straightforward in theory: deploy AI across daily operations, shrink teams dramatically, and pocket the cost savings. Two rounds of cuts were scheduled for May and November to execute this transformation over the course of the year.
Meta confirmed the initiative as a year-long planning exercise focused on cost-cutting and shifting teams toward AI. Not every department was in the crosshairs, but the scale of the ambition was hard to miss.
The May layoff proceeded. On May 20, approximately 8,000 employees, roughly 10% of Meta’s workforce, lost their jobs. Nearly 7,000 others were reassigned to AI-focused roles. That’s about 15,000 people whose work lives changed in a single month.
Then came the plot twist. On May 19, just hours before the first round of cuts began, Zuckerberg canceled the November layoffs entirely. The broader restructuring targets were abandoned. The CEO acknowledged mistakes in the rollout and pledged no further company-wide layoffs.
The numbers that killed the plan #
The internal data told a split story that ultimately undermined the whole premise.
On the surface, one metric looked impressive: AI-assisted coding increased by 220% year-over-year. Engineers were writing code faster than ever.
But user-facing product improvements grew by just 36%. Writing code faster doesn’t help much if the output doesn’t translate into better products.
Even worse, security incidents reportedly climbed by approximately 40% during the AI transition. That meant engineering teams were spending more time putting out fires than building new features, essentially creating a productivity tax that offset whatever gains the AI tools were generating.
Employee resistance compounded the technical shortcomings. When you tell people that AI is coming for their jobs, and then the AI tools underperform, you get a workforce that’s both demoralized and skeptical.
Why this matters beyond Meta #
The gap between AI-assisted coding productivity and actual user-facing improvements is particularly telling. It suggests that current AI tools are better at augmenting narrow technical tasks than at replacing the broader judgment, creativity, and cross-functional coordination that goes into shipping products people want to use.
Katie Paul is scheduled to host a Reddit AMA on September 3, 2026, to discuss her reporting on Project OT in more depth. Given the internal complexity of what she uncovered, including canceled layoff waves, mixed productivity data, and a CEO walking back his own initiative, the session should offer a closer look at how one of tech’s boldest workforce experiments ran headfirst into the gap between AI’s promise and its current capabilities.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our