Mark Zuckerberg AI Staffing Strategy: How Project OT Imploded

The Mark Zuckerberg AI staffing strategy had a code name, a blueprint and a timetable. A Reuters special report published on August 26, 2026 says Meta planned to hand much of its daily work to AI agents and shrink some teams by up to 60%. Then, hours before the second round of cuts, Zuckerberg called the whole thing off.

The goal was an “AI native” Meta. Software agents would handle much of the work thousands of employees did by hand. Meanwhile, smaller groups of staff, described internally as talent dense, would supervise them. Above all, the plan assumed those agents could absorb real work at scale.

Reuters built its account from internal planning documents and people close to the project. Indeed, the full Reuters special report traces how fast the plan moved from retreat to layoff notice.

It began in January 2026, at Zuckerberg’s annual leadership retreat on his Hawaii compound. Then his top lieutenants drew up Project OT, short for Organization Transformation.

Of course, money was part of the logic. Meta meant to spend some of the savings from layoffs on very large pay packages for a much smaller bench. Meanwhile the infrastructure bills kept climbing. Capital spending for 2026 is guided at 130 to 145 billion dollars, against 72.2 billion in 2025.

The org chart was redrawn from scratch. Work would sit with pods of two or three engineers and designers, all carrying one generic title: builder. So a Pod Lead would run the day to day, without formal management authority.

Above them, Org Leads would each oversee 30 to 50 people, and layers of middle management would vanish. Internal documents also show executives modeling cuts of up to 60% for many teams. Meta later confirmed the figure, saying the most drastic scenarios involved reducing the size of some teams by up to 60 per cent.

In April, Reuters reported that Meta had mandated tracking software on the devices of its US staff. It captured keystrokes and mouse clicks, and that data fed the training of AI agents.

Employees drew the obvious conclusion. Many said openly that they were training their own replacements. As a result, labor organizing gained momentum, and favorable sentiment in the half-year Pulse survey slid from 74% to 55%.

That timing mattered. The tracking arrived while the first layoff wave was still being modeled, so every internal message about efficiency landed on a workforce that had already done the arithmetic.

Inside the company, the numbers told a different story from the pitch deck. By March, infrastructure teams were flagging reliability warning signs. In fact, agents were taking disruptive actions that humans were unlikely to perform. By July, Zuckerberg was conceding in public that AI development was moving slower than he had expected.

That gap between code written and features shipped echoes the wider evidence on AI coding tools productivity impact. In short, more output is not the same thing as more value.

Wave one landed on May 20, 2026, when Meta cut 10% of its workforce. Before that, HR executives had braced for a program that matched or beat the roughly 25% reduction of 2022 and 2023.

However, wave two never came. Hours before that May announcement, Zuckerberg canceled the November restructuring. His follow-up memo told staff he did not expect other company-wide layoffs this year, and it leaned hard on one word: stability.

Nevertheless, Meta’s public line stays measured. The company says it asked some teams to run a scenario planning exercise, that thousands of employees moved to priority work on newly established teams, and that it never assumed every scenario would go ahead.

Plenty of companies are running some version of this experiment. Meta may simply be the first to publish the post mortem by accident. Still, its lesson is narrow and useful: agents can raise activity long before they raise output.

A financial lesson sits underneath. Free cash flow fell 91% year on year in the second quarter of 2026, to 784 million dollars, even though revenue grew 28% to 60.8 billion. Because of that squeeze, the pressure to show a return on AI has not gone anywhere.

Curious how these agents actually get built? Our roundup of the top agentic AI frameworks walks through the tooling Meta was betting on. For the wider picture, read our report on AI technology job cuts in 2026.

Sources

*Sources: Reuters, The Globe and Mail, The Business Standard, TimesLIVE, IBTimes, The Free Press Journal, CNBC, The Motley Fool, best-ai.news.* *Photos: Mark Zuckerberg (2025) by Jeff Sainlar, image provided by Meta, CC BY-SA 4.0, resized and placed on a blurred background. Menlo Park sign: Meta Platforms Headquarters Menlo Park California, CC0, cropped.*