AI Unit's Internal Strife
Meta's Applied AI Unit Grapples With Internal Turmoil Amid Low Employee Morale
Key Points
During a call open to thousands of employees, an expletive-filled interruption reportedly demanded leaders “write to a specific Meta AI executive” and tell him “he’s a piece of shit,” after which the meeting’s two leaders asked everyone to mute and moved on to their technical talk.
Employees who spoke to WIRED described the Applied AI work as extreme and purposeless—one said, “It’s literally the gulag,” and described tasks such as “generating puzzles” to test how reliably models can solve them, with another saying, “Most people find the work soul-crushing.”
Meta’s AI push includes named deliverables and quantified staffing changes: Meta is shifting “about 7,000 employees into AI-related roles,” and the superintelligence lab has produced an AI model called “Muse Spark” aimed at improving products like Instagram and WhatsApp.
In internal debate over open-source vs. proprietary models, coverage points to timing around the next-generation “Llama” releases—after the anticipated launch of “Llama 3 in 2024,” employees worry Meta could move toward closed-source systems—while crypto-linked observers argue decentralization should replace open-source AI, citing projects including Bittensor and Fetch.ai.
A livestreamed all-hands meeting at Meta turned ugly on June 12, 2026, when an employee unmuted and demanded that leaders tell a top executive "he's a piece of shit." The outburst, heard by thousands, was the flashpoint for a WIRED exposé revealing that Meta's 6,500-person Applied AI unit is on the verge of revolt.
The blow-up caps months of turmoil inside Meta's new Applied AI (AAI) unit, formed in March 2026 to support the company's Superintelligence Labs. Workers describe a "soul-crushing" environment, a strategic U-turn away from open-source AI, and a $145 billion capital spending spree that has reshaped — and rattled — the entire company, according to TechCrunch.
Employees Call the New AI Unit "The Gulag"
The AAI unit was built by "drafting" engineers from other teams, with no option to transfer out. High-paid engineers have been reassigned to tasks like generating coding puzzles to test model reliability. One worker told WIRED: "It's literally the gulag." Another said, "Most people find the work soul-crushing."
Management ratios inside the unit have reached as high as 50 individual contributors per manager, according to TechCrunch. That leaves many workers feeling isolated and purposeless. The June 12 livestream incident — two leaders simply asked everyone to mute and moved on — showed how far trust has broken down between Meta's technical staff and its leadership.
Zuckerberg Admits Mistakes as 7,000 Workers Get Reassigned
In May 2026, Meta laid off 8,000 employees — about 10% of its roughly 78,000-person workforce — while simultaneously moving 7,000 others into AI-focused roles, according to Reuters. The moves were framed as building a "flatter" structure with "smaller teams." Many employees read that as a euphemism for impossible workloads.
On June 12, Zuckerberg issued an internal memo admitting the company had made "mistakes." He wrote: "Given the complexity of these changes, we've made mistakes and will almost certainly make more." Chief AI Officer Alexandr Wang — hired in 2025 after Meta spent $14.3 billion for a 49% stake in his former firm Scale AI — is now tasked with closing the gap with OpenAI and Google, per TechCrunch.
Meta Quietly Abandoned Open-Source AI for a Closed Model
As recently as July 2024, Zuckerberg published a manifesto calling open-source AI "the path forward." By April 8, 2026, Meta had launched Muse Spark — a fully closed-source model, the opposite of its Llama line. The shift followed what insiders called the "Llama 4 debacle," with staff viewing the model as a failure next to rivals like GPT-5 and Gemini, according to finance.yahoo.com.
Muse Spark features a "Contemplating Mode" for deep reasoning and is being woven into Instagram and WhatsApp for personalized, automated features. It currently leads the HealthBench Hard benchmark with a score of 42.8, beating GPT-5.4 and Gemini 3.1 Pro. But critics in the crypto space — pointing to decentralized projects like Bittensor and Fetch.ai — argue the closed pivot proves big tech cannot be trusted with AI, per TechCrunch.
A $145 Billion Bet That Could Hollow Out Meta's Best Engineers
Meta's 2026 capital spending is running at $145 billion annually, driven almost entirely by AI hardware like Nvidia H100 and H200 chips. The company also implemented token usage limits on internal AI tools after employees consumed around 60 trillion tokens — spiking costs to the point where Meta built a new tracking system called AI Gateway, according to finance.yahoo.com.
The bigger risk may be talent. Senior engineers in the AAI unit are expected to leave as soon as their stock vesting allows, according to TechCrunch. That could hollow out the middle layer of Meta's engineering team right as it needs it most. The next public test comes at Meta Connect on September 23–24, 2026, where Zuckerberg must show that the internal pain is producing real products.
Sentiment
Mostly NegativePublishers
13
Articles
11
Reach
24
