In a candid internal memo that highlights the growing pains of a Silicon Valley giant in the midst of a radical pivot, Andrew Bosworth, Meta’s Chief Technology Officer, admitted that the company performed an "atrocious" job of launching its new Applied AI division. The post, shared with employees on Monday and subsequently reviewed by news outlets, represents a rare moment of executive contrition following months of internal turmoil, reported worker dissatisfaction, and a culture that some employees have likened to a "gulag." Bosworth’s message signaled a desperate attempt to "rekindle" the company’s legendary culture through a combination of management restructuring, career development initiatives, and the return of office perks like improved snacks and social budgets.
The admission follows a period of intense friction within Meta’s AI engineering ranks. In March, Meta consolidated approximately 6,500 engineers and product managers into a new Applied AI unit, tasked with the high-stakes mission of integrating generative AI across the company’s vast ecosystem of apps, including Facebook, Instagram, and WhatsApp. However, the transition was far from seamless. Reports emerged of engineers being "drafted" into the unit against their will, assigned to repetitive or menial tasks that left high-level talent feeling underutilized and undervalued. Bosworth’s memo acknowledged these failures directly, noting that the company had "undermined the trust" of its workforce by prioritizing speed over the professional well-being of its staff.
The Applied AI Friction: From Engineering to "Janitorial" Work
The core of the dissatisfaction stems from the nature of the work assigned to the 6,500 members of the Applied AI team. As Meta races to compete with rivals like OpenAI, Google, and Anthropic, the demand for high-quality training data and the refinement of large language models (LLMs) has become paramount. For many veteran engineers who were previously working on autonomous projects or high-level product development, the shift to Applied AI felt like a demotion to what some described as "janitorial" tasks.
Internal reports indicated that workers were tasked with manual data labeling, basic testing of coding agents, and repetitive debugging to improve the "agentic" capabilities of Meta’s frontier models. One employee famously described the environment as a "gulag," a term that underscored the feeling of being trapped in a high-pressure, low-fulfillment environment. Bosworth addressed this sentiment by admitting that the executive team had lost sight of the employee perspective while rushing to address broader strategic imperatives.
"We obviously did an atrocious job explaining the vision," Bosworth wrote. He noted that the company failed to provide a clear picture of how employees would be supported in their careers during the shift. While he maintained that the decision to draft people onto the AI team was the correct move for the sake of the company’s survival in the AI race, he acknowledged that it came at a significant cost to morale.
Structural Reforms and Management Restructuring
To address the erosion of trust, Bosworth outlined a series of structural changes intended to provide more stability for engineers. One of the most significant changes is a new cap on the number of direct reports for managers. Meta plans to limit managers to approximately 20 direct reports, a move designed to ensure that employees receive more personalized attention and career coaching. In the wake of Meta’s "Year of Efficiency," many management layers were stripped away, leaving some supervisors overwhelmed with 30 or 40 subordinates, leading to a breakdown in mentorship and professional guidance.
Furthermore, the company aims to limit the frequency of manager changes during internal restructurings. Frequent reorgs have been a staple of Meta’s operation over the last two years, often leaving teams "in the lurch" as strategies shifted overnight. Under the new guidelines, managers will be encouraged to focus primarily on people management rather than their own independent technical work, ensuring that they are available to advocate for their team members’ career growth.
Bosworth also introduced the concept of "AI coaching" tools. These tools are intended to help workers navigate the rapidly changing landscape of AI development, providing them with the resources to upgrade their skills and ensure they remain competitive. This move aligns with Bosworth’s stated philosophy that "AI won’t take your job, but someone who knows AI might."
The Cultural Context: Rebuilding After the "Year of Efficiency"
The current crisis within the AI division does not exist in a vacuum. It is the latest chapter in a broader downward swing in morale that began in late 2022. Since then, Meta has undergone several rounds of mass layoffs, resulting in the departure of more than 21,000 employees. CEO Mark Zuckerberg dubbed 2023 the "Year of Efficiency," a period characterized by cost-cutting, the flattening of management hierarchies, and a move away from the "perk-heavy" culture that once defined the company.
While the Year of Efficiency was cheered by Wall Street—Meta’s stock price has recovered significantly and hit record highs in 2024—the internal toll has been heavy. Employees have reported increased surveillance, including the tracking of office badge-ins and mouse movements, alongside a general feeling of job insecurity. The aggressive pivot toward AI, following the expensive and largely unproven multi-billion-dollar bet on the Metaverse, has added a new layer of stress to the workforce.
In addition to Bosworth’s memo, CEO Mark Zuckerberg has also reportedly been active on internal message boards, acknowledging the "bad vibes" within the company and promising to address concerns. The executive team appears to be pivoting toward a "hearts and minds" campaign to retain the top-tier talent necessary to win the AI arms race.
Chronology of Meta’s Strategic Shifts
To understand the current tension, it is necessary to look at the timeline of Meta’s recent corporate evolution:
- November 2022: Meta announces its first-ever mass layoff, cutting 11,000 jobs (approx. 13% of its workforce) as the post-pandemic digital ad boom cooled.
- February 2023: Mark Zuckerberg announces the "Year of Efficiency," signaling a shift from growth-at-all-costs to a leaner, more disciplined corporate structure.
- March 2023: A second round of layoffs is announced, affecting another 10,000 workers.
- Late 2023: Meta releases Llama 2, signaling its serious entry into the open-source generative AI market. The company begins a massive hiring spree and internal reallocation of talent toward AI.
- March 2024: The Applied AI unit is formed, consolidating 6,500 staff members. Friction begins almost immediately as engineers from diverse backgrounds are "drafted" into standardized AI tasks.
- June 2024: Internal dissent reaches a boiling point, with the "gulag" comments leaking to the press. Bosworth and Saba issue memos promising cultural and structural changes.
Strategic Competition and the Compute Resource War
The pressure on the Applied AI team is exacerbated by the sheer scale of the competition. Meta is currently spending tens of billions of dollars on NVIDIA H100 GPUs to build the infrastructure necessary for Llama 4 and beyond. In its Q1 2024 earnings call, Meta raised its capital expenditure forecast to between $35 billion and $40 billion, citing AI infrastructure as the primary driver.
However, even with these massive investments, internal resources are strained. Bosworth’s memo touched on "tough trade-offs" regarding the amount of compute available to different teams. In the world of AI development, compute—the processing power needed to train and run models—is the most precious currency. The fact that Meta, one of the world’s wealthiest companies, is facing "bottlenecks" highlights the intensity of the current technological race.
Engineers within the Applied AI team are not only dealing with menial tasks but are also competing with other internal units for the hardware resources needed to execute their projects. This scarcity has led to further frustration, as teams are held to high performance standards while lacking the tools to achieve their goals.
Snack Diplomacy and the Return to Legacy Perks
Perhaps the most surprising element of Bosworth’s memo was the emphasis on "microkitchens" and social events. During the Year of Efficiency, many of the famous Silicon Valley perks—free gourmet meals, laundry services, and extensive travel budgets—were scaled back or eliminated.
Bosworth’s vow to "improve microkitchens" and increase budgets for social events and travel suggests a return to "snack diplomacy." The theory is that by improving the physical environment and encouraging in-person social interaction, the company can foster a more "cheerful" and collaborative culture.
"I hope we can rekindle the best of the culture we joined," Bosworth wrote. While snacks and happy hours may seem trivial compared to the structural issues of career growth and job security, they serve as a symbolic olive branch to a workforce that has felt increasingly alienated from the corporate mission.
Official Response and the Path Forward: "Fixing Forward"
Maher Saba, a Vice President leading the Applied AI team, also addressed the workforce in a separate post. He introduced a new mantra for the unit: "moving fast and fixing forward." This is a calculated evolution of Meta’s original and now-retired motto, "move fast and break things."
Saba acknowledged the dynamic nature of AI work, noting that traditional six-month engineering roadmaps no longer apply because the technology evolves too quickly. To mitigate the frustration of those who were forced into the unit, Saba announced that employees would now be allowed to apply for other roles within Meta if they could secure them. This return to "business as usual" regarding internal mobility is intended to restore a sense of agency to the staff.
The Applied AI group will continue its work on increasing the coding and agentic capabilities of Meta’s models, with plans to expand into cybersecurity, debugging, and automated product development. The goal is to move beyond the "menial" phase and into a period where the 6,500-strong team is driving the next generation of AI-powered features.
Analysis: Can Meta Bridge the Cultural Divide?
The success of Meta’s AI ambitions depends entirely on its ability to retain and motivate high-level engineering talent. While the company has the capital and the hardware, it is currently facing a "human capital" crisis. The admission by Bosworth is a significant first step, but the path to recovery is complex.
The tension between "speed" and "employee fulfillment" is a classic dilemma in high-tech industries, but it is magnified at Meta due to the scale of the AI transition. By capping management reports and allowing for more internal mobility, Meta is attempting to create a more sustainable version of its "efficiency" model. However, the underlying pressure remains: Meta must prove that it can turn its massive investment in AI into a product that defines the next era of the internet.
Whether "snack diplomacy" and management tweaks can heal the wounds of the last two years remains to be seen. For now, the leadership at Meta is betting that by admitting their mistakes and returning to some of the cultural foundations that built the company, they can convince their engineers that the "gulag" days are over and a new, more fulfilling era of AI development has begun.
