Meta Reorganizes AI Team With 'Radically Flat' Structure
Meta is undertaking a sweeping engineering reorg, cutting ties with its highest-paid AI chief Alexandr Wang. Mark Zuckerberg is now championing a new flat-structured applied AI team with as many as 50 engineers per manager to accelerate its path to superintelligence.
The new applied AI group is headed by Maher Saba, a VP from Meta's Reality Labs, and will report directly to CTO Andrew Bosworth. This team is tasked with creating the data engine to improve Meta's AI models, focusing on building interfaces, internal tooling, and the data pipelines necessary to turn capable models into market-leading ones. This move is part of a broader "year of efficiency" at Meta, a strategy Zuckerberg has mentioned for over a year to flatten organizational structure and remove layers of middle management. The goal is to speed up decision-making and decrease the latency of information flow up and down the chain, empowering individual contributors. The reorganization follows several previous shake-ups within Meta's AI divisions. In the past year, Meta dissolved its GenAI and AGI Foundations teams, redistributing staff after the lukewarm reception of its Llama 4 model. These constant changes signal concerns about being outpaced by rivals like OpenAI and Google. Alexandr Wang, who joined Meta after it invested $14.3 billion for a 49% stake in his company Scale AI, was initially tapped to lead the superintelligence efforts. However, reports indicate clashes with longtime Meta executives and a narrowing of his control, with the new applied AI team being structured around him, potentially to increase pressure to perform. Zuckerberg has described the flat structure of his Superintelligence Labs as functioning like a "group science project" to foster innovation and attract top talent who want direct visibility. This model contrasts with traditional hierarchical structures by design, aiming to make Meta more agile. High-performing teams with more autonomy often dedicate more time to new work, which aligns with the goal of accelerating development. The new unit will partner closely with the Meta Superintelligence Labs, led by Wang, to enhance frontier AI models. Saba's team will focus on leveraging reinforcement learning and post-training techniques, which have recently shown significant gains, to accelerate progress. This is a direct effort to build the infrastructure and data flywheel that turns a strong model into a dominant one.
