Mark Zuckerberg Admits Meta Made Mistakes in AI Workforce Shift
mark zuckerberg said in an internal memo on Friday that Meta’s recent AI workforce changes had caused distress. He also said the company had made mistakes it plans to address, after a three-month-old Applied AI unit became the center of mounting employee anger.
Meta’s Applied AI unit
The unit has roughly 6,500 engineers and product managers, and employees assigned there are generating puzzles and coding problems to train AI models. That puts a large slice of Meta’s AI staffing on routine data work rather than visible product launches, and it helps explain why workers inside the group are treating the transfer as a career choice they were not given.
Employees said they were forced into the group with no real choice and had to join or quit. One employee told Wired, “It’s literally the gulag,” and another said, “Most people find the work soul-crushing.”
Chris Cox and the backlash
Meta chief product officer Chris Cox addressed the “brutal” environment on a call with employees this week. More than 1,600 Meta employees company-wide have also signed a petition protesting a program that monitors their clicks and keystrokes for AI training data, showing that the backlash now reaches beyond the Applied AI team itself.
The unit is led by Maher Saba, a 12-year Meta veteran who previously served as a vice president in Reality Labs. The new organization reports to Meta CTO Andrew Bosworth, while Meta has shifted billions toward AI after years of layoffs and an $83 billion burn in Reality Labs.
Zuckerberg, Wang and the gap
At an internal meeting last month, Zuckerberg said the average Meta employee has “significantly higher” intelligence than third-party contractors. In a leaked audio recording from that meeting, he also said Alexandr Wang knows the data-labeling world well, even as Meta’s own internal announcement said its AI models still lacked the knowledge to outperform humans at technical tasks like coding.
That gap between ambition and capability is now visible in the work itself. Meta said, “For agents to understand how people actually complete everyday tasks using computers, we need to train our models on real examples,” which explains why the company is leaning so hard on human-generated training data instead of the performance it says it does not yet have.
The unresolved issue is whether Meta can keep recruiting and retaining AI staff while the company keeps pushing the same training model that has already triggered petitions, a webcast meltdown, and public complaints from inside the team.
