Meta's three-month-old Applied AI unit is experiencing a staff uprising, with thousands of engineers describing their mandatory reassignments as forced labor and calling the work "soul-crushing." The company packed roughly 6,500 engineers and product managers into the division following a May overhaul that cut 8,000 jobs and shifted 7,000 more people into AI roles, affecting close to one-fifth of Meta's workforce.

What Triggered the Breakdown at Meta's AI Unit?

The breaking point arrived this week when someone hijacked a livestreamed employee presentation and launched into an expletive-laden meltdown, demanding that attendees tell a senior Meta AI executive he was "a piece of sh*t." One presenter reportedly covered their face with both hands. The outburst exposed deeper frustration simmering across the Applied AI team, where workers say Meta gave them no real choice: join the unit or quit.

Many employees now call themselves "draftees," a term that captures the involuntary nature of their reassignment. Their daily work is narrow and repetitive: writing puzzles and coding problems used to train Meta's AI models. "It's literally the gulag," one employee told Wired. "Most people find the work soul-crushing," said another. The anger intensified because many workers discovered their fate through a sudden email, with no advance notice or consultation.

An internal announcement reviewed by Wired spelled out Meta's reasoning: the company's models still could not beat humans at technical work like coding. "For agents to understand how people actually complete everyday tasks using computers, we need to train our models on real examples," the note read. Rather than hire outside contractors, Meta decided to repurpose its own workforce.

Why Did Zuckerberg Choose Internal Staff Over Outside Contractors?

In a leaked audio recording, Mark Zuckerberg defended the decision to draft staff instead of hiring external data-labeling firms. He pointed to his chief AI officer, Alexandr Wang, who sold his data-labeling startup Scale AI to Meta for $14.3 billion before taking over Meta Superintelligence Labs. Wang knows the data-labeling trade well, Zuckerberg argued. More importantly, Zuckerberg claimed the average Meta employee carries "significantly higher" intelligence than a third-party contractor, making staff the better pick for training advanced AI models.

Zuckerberg

The decision reflects a broader shift in how AI companies approach model training. Brute computing power alone is no longer enough; models learn from human-written examples and judgment. This is why a company paying top engineers handsomely has some of them solving coding puzzles by hand, and why Meta spent $14.3 billion acquiring a data-labeling firm in the first place.

How Is Meta Trying to Fix the Morale Crisis?

By Friday, Zuckerberg was in damage-control mode. In an internal memo, he admitted the recent changes had "caused distress" and owned the missteps the company now plans to fix. "Given the complexity of these changes, we've made mistakes and will almost certainly make more," he wrote. The memo included concrete commitments aimed at rebuilding trust and improving working conditions across the company.

• Team Resources: Meta promised bigger budgets for team offsites and a company-wide hackathon scheduled for July to foster collaboration and creativity.
• Workspace Changes: The company will return assigned desks in many offices and reduce the number of direct reports per manager, addressing structural issues that made the initial rollout "punishing" with up to 50 employees answering to a single manager.
• Career Pathways: Zuckerberg pledged to find new roles for staff stuck doing model-training work, acknowledging that the current assignments are not sustainable long-term.
• Job Security: He vowed no more company-wide layoffs for the rest of 2026, providing some stability after the May upheaval.

The discontent extends beyond the Applied AI group. More than 1,600 employees across Meta have reportedly signed a petition against a program that tracks their clicks and keystrokes to gather AI training data, one piece of a broader push to weave AI into everyday work. The mood ran dark enough that chief product officer Chris Cox felt compelled to address the "brutal" environment on a call with employees this week.

What Does This Reveal About AI's Real Costs?

The Applied AI revolt exposes a fundamental tension in the AI industry: you can buy chips, but you cannot order morale. Meta is betting heavily on AI, lifting its 2026 capital-spending plan to between $125 billion and $145 billion, nearly double last year's outlay, with most aimed at AI data centers and training clusters. That wager looks easier to stomach when the same overhaul that erased roughly 8,000 jobs also frees cash to spend on infrastructure.

Yet the unrest inside Applied AI exposes a ceiling. Investors have begun sorting tech firms by whether all that AI spending actually pays off, and on that question Meta still owes a clearer answer than a memo can provide. The company's ability to retain talent and maintain morale while pursuing aggressive AI expansion will likely determine whether its massive capital investment translates into competitive advantage or becomes a cautionary tale about the human cost of scaling artificial intelligence.

Maher Saba leads the Applied AI team, a 12-year Meta veteran who once served as a vice president in Reality Labs, the division that burned through $83 billion on the metaverse before the company pivoted to AI. His challenge now is to rebuild a team that feels conscripted rather than inspired, a task that no amount of hackathons or desk assignments may fully resolve.