Meta engineers describe Applied AI unit as forced draft

A three-month-old Meta unit with roughly 6,500 engineers and product managers has become the latest flashpoint in the company’s AI push, according to TechCrunch. Employees inside the Applied AI team describe being moved there with little choice — “join or quit,” in one account — and spending their days producing puzzles and coding problems meant to generate training data for Meta’s models.

The complaints land in a company that has paired repeated rounds of layoffs with billion-dollar AI spending. Meta’s internal logic, as described in a leaked audio recording from an April meeting cited by TechCrunch, is that its own engineers produce higher-quality examples than outside contractors. That view sits awkwardly alongside an internal announcement in April that Meta’s models still lacked the knowledge to outperform humans at technical tasks like coding, and that the company needed real examples of how people complete everyday computer work. If the models need authentic human workflows, the cheapest reliable source is the workforce already on payroll — and the cost of that choice shows up as reassignment, monitoring, and morale.

TechCrunch reports that more than 1,600 employees have signed a petition protesting a program that monitors clicks and keystrokes to collect training data. When the same company that can deactivate user accounts without meaningful appeal (as Meta’s Oversight Board recently complained) turns similar automation inward, the boundary between product governance and workplace governance starts to blur. The Applied AI group’s original management structure reportedly had as many as 50 employees reporting to one manager, a ratio that can make individual grievances hard to surface and easy to route into generic HR processes.

The internal temperature became visible this week when a livestreamed, employee-only presentation was hijacked by someone who launched into an expletive-laden outburst aimed at a senior Meta AI executive, according to TechCrunch. The incident is notable less for the profanity than for the venue: a company channel, during a controlled internal event, in a unit that exists to supply the raw material for Meta’s next-generation AI products.

Meta has begun acknowledging the strain. Chief product officer Chris Cox described the environment as “brutal” on a call with employees this week, TechCrunch reports. CEO Mark Zuckerberg followed with an internal memo on Friday saying recent changes had “caused distress,” conceding mistakes and promising fixes — while restating that Meta’s “north star” is to be the best place for top talent to make an impact.

The unit is led by Maher Saba and reports up through CTO Andrew Bosworth, according to TechCrunch. It is also operating in an AI organization reshaped by big-ticket deals, including Meta’s acquisition of Scale AI for $14.3 billion and the appointment of Alexandr Wang as chief AI officer to lead Meta Superintelligence Labs.

For now, the work product is not a new consumer feature, but thousands of employees producing training examples under monitoring tools. The company’s pitch is superintelligence; the immediate deliverable is keystrokes.