Meta’s newly formed Applied AI unit, created in March 2026 to support the company’s Meta Superintelligence Labs, is experiencing a full-scale internal revolt. Engineers and product managers assigned to the unit describe being “drafted” with no real choice between joining or quitting. Their primary task, writing puzzles and coding problems to train Meta’s AI models, is described by multiple employees as “soul-crushing” and by one internal message as working in a “gulag.” The unit comprises approximately 6,500 people.

Mark Zuckerberg acknowledged in a Friday memo that the reorganization “caused distress” and that the company had made mistakes. He pledged to find new roles for staff stuck on model-training work and promised no company-wide layoffs for the rest of 2026.

What the Applied AI Unit Is

Meta created the Applied AI unit in March 2026 as a support organization for Meta Superintelligence Labs, the division the company assembled by poaching AI researchers and paying packages reported to exceed $100 million per person for top talent. The 6,500-person Applied AI unit was meant to do the supporting work: generating training data, building evaluation systems, and creating the infrastructure that expensive researchers need.

The problem is that the work assigned to Applied AI is repetitive, narrow, and strips engineers and product managers of the technical challenges that make software jobs valuable professionally. Instead of building products or solving novel engineering problems, Applied AI employees are writing puzzles and coding questions used to evaluate and train Meta’s large language models.

Why Engineers Are Revolting

According to Futurism, employees say Meta gave them no real choice about the transfer. Those who refused to join the unit were effectively told to find other jobs. The perception of being conscripted, combined with the demoralizing nature of the assigned work, produced the revolt.

The “gulag” comparison, used internally according to reports, reflects how engineers view repetitive model-training data generation relative to their previous work building consumer features and infrastructure. Senior engineers who spent years building complex systems found themselves assigned to write math puzzles at scale.

Internal communications cited in the reporting include a message telling an employee he was “a piece of sh*t” for underperformance on the training data metrics, a management approach that compounded the morale crisis.

The Deeper Strategy Problem

Beyond the employee relations crisis, Meta faces a fundamental strategic disagreement about its AI direction. The confusion spans whether Meta’s AI models should remain open-source, a position Zuckerberg has publicly championed, or shift toward proprietary closed systems that can be monetized more directly.

The May reorganization that created the Applied AI unit reshaped approximately one-fifth of the company’s workforce through layoffs and forced transfers. A reorganization affecting 20 percent of employees without a clear communicated strategy creates the exact uncertainty that drives high performers to update their resumes.

Meta has also capped internal AI token spending after costs approached billions of dollars, a sign that even the world’s fifth-largest company by market cap is finding AI infrastructure costs difficult to absorb at the scale of internal usage that AI-first culture demands.

Zuckerberg’s Response

Zuckerberg’s Friday memo acknowledged the “distress” caused by the changes and promised to find new roles for Applied AI employees doing model-training work. He framed the overall direction as non-negotiable while conceding the execution was flawed.

The no-layoffs pledge for the rest of 2026 is a significant commitment given the company has conducted multiple rounds of layoffs since 2022. Whether it holds depends on whether the AI investment strategy produces sufficient results to justify the cost before the end of the year.

Frequently Asked Questions

What is Meta’s Applied AI unit?

Meta’s Applied AI unit was created in March 2026 as a support organization for Meta Superintelligence Labs. It comprises approximately 6,500 engineers and product managers whose primary task is generating training data, including puzzles and coding problems, for Meta’s AI models. Employees describe being transferred into the unit with no real choice between joining or leaving the company.

Why are Meta employees calling their unit a gulag?

Employees use the term because they perceive the Applied AI assignment as a forced transfer to repetitive, professionally unrewarding work with no exit option. Senior engineers who built consumer features and infrastructure were reassigned to write math puzzles for AI training. Combined with aggressive management messages and the lack of choice in the transfer, the experience is described as demoralizing and punitive.

Did Zuckerberg respond to the Meta AI unit revolt?

Yes. Zuckerberg issued a Friday memo acknowledging the reorganization “caused distress” and that mistakes were made. He pledged to find new roles for employees stuck doing model-training work and promised no company-wide layoffs for the rest of 2026. The memo did not reverse the strategic direction but acknowledged the implementation was flawed.