Après 8 000 licenciements, Zuckerberg admet que Meta a mal évalué le progrès de l'IA

Mark Zuckerberg, CEO of Meta Platforms Inc., recently acknowledged during an internal town hall meeting that the company had miscalculated the pace of progress in artificial intelligence (AI). This admission came amid widespread discontent following a major restructuring effort that resulted in approximately 8,000 layoffs, representing roughly 10% of Meta’s global workforce. The restructuring aimed to shift focus toward AI development, with plans to allocate up to $145 billion toward AI infrastructure by 2026. However, the move has led to significant employee dissatisfaction and raised questions about the effectiveness of the strategy thus far.

Zuckerberg noted that the trajectory of AI agent development over the past four months had not met expectations. While the company had initially expressed optimism about advancements in AI tools such as Anthropic’s Claude Code, these hopes have not materialized into tangible outcomes. Despite the challenges, Zuckerberg remains cautiously optimistic, stating that he anticipates seeing more substantial returns from Meta’s AI investments within the next three to six months.

The controversy surrounding Meta’s AI initiatives extends beyond strategic miscalculations. Concerns were raised regarding the company’s mouse-tracking software, which had been temporarily suspended due to potential data privacy issues. In response, Meta’s Chief Technology Officer, Andrew Bosworth, confirmed that a recent review found no evidence of employee data being used in AI training. Should the program be reinstated, it would operate on an opt-in basis, allowing users to choose whether they wish to participate.

Bosworth also took responsibility for the rollout of Meta’s newly established Applied AI division, describing the implementation as an “atrocious” performance. Formed in March with 6,500 engineers and product managers, the division was intended to drive forward Meta’s generative AI projects. However, the transition was marked by confusion and disorganization, with some employees referring to the environment as “a gulag.” Bosworth emphasized the need to restore trust by clarifying the company’s vision and stabilizing teams affected by rapid restructuring. To improve the workplace atmosphere, he proposed measures including enhanced communication, better career development opportunities, and improved office amenities such as microkitchens and increased travel budgets.

Amidst these corporate challenges, an intriguing narrative emerged from outside the company. Rishabh Agarwal, an IIT Bombay alumnus and accomplished AI researcher, declined a lucrative job offer from Meta, opting instead to pursue independent research. Agarwal, known for his contributions to reinforcement learning and large language models, had previously worked at prestigious institutions such as Google Brain, Google DeepMind, and Waymo. He joined Meta’s Superintelligence Labs in April 2025 but left after about five months, citing the desire to explore different risks in a rapidly evolving technological landscape.

Agarwal’s decision to leave Meta was influenced by his experiences across multiple leading AI organizations. Following his departure, he co-founded Periodic Labs, an AI startup dedicated to developing an AI scientist capable of formulating hypotheses for real-world scientific experiments. The startup aims to expedite discoveries in fields ranging from new materials to medicine and physics. Notably, Periodic Labs has garnered support from notable figures such as Nvidia and Jeff Bezos, positioning it as a promising entity in the AI research domain.

As Meta continues to navigate its AI ambitions, the company faces both internal and external scrutiny. With high-profile departures and strategic recalibrations, the path ahead remains uncertain. The outcome of Meta’s current strategies will likely depend on how effectively the company addresses ongoing challenges and adapts to the dynamic nature of AI innovation. Meanwhile, individuals like Agarwal highlight the growing trend of professionals seeking alternative avenues to shape the future of technology beyond traditional corporate structures.