In a move that signals Meta’s serious commitment to accelerating artificial intelligence, CEO Mark Zuckerberg is launching a dedicated applied AI engineering arm. This new unit is designed to speed up the company’s pursuit of more capable, even “superintelligent,” systems by rethinking how Meta organizes its AI work. Rather than a single, centralized machine-learning team, Meta will operate with a constellation of specialized squads, each with a clear mandate and a faster path from idea to impact.
What makes this shift noteworthy is not just the creation of a new department, but the way it’s structured. The data-driven engine at the core of Meta’s strategy will be built under the leadership of Maher Saba, a veteran executive from Reality Labs. The team will report to Chief Technology Officer Andrew Bosworth, but what stands out is the organization’s unusually flat hierarchy. The plan envisions up to 50 individual contributors for every manager, a setup aimed at slashing red tape and speeding decision-making as the company scales its AI ambitions.
The scope of Saba’s unit centers on three core pillars: data processing, tooling, and model evaluations. Put plainly, this team is tasked with refining the raw inputs that feed Meta’s models, creating robust software tools to support researchers, and rigorously assessing how well the models perform. In practical terms, this is the operational backbone that makes models better, faster—an essential ingredient in the broader race toward advanced AI capabilities.
This evolution follows Meta’s earlier foray into a more centralized “Superintelligence Labs” approach under a different leadership, led by Alexandr Wang. The current reshuffling distributes responsibility across multiple leaders and teams, including Wang’s research lab and Zuckerberg’s broader technology strategy group. Projects that were previously housed under one umbrella—codenamed Avocado and Mango among them—will now be pursued in parallel by distinct squads with their own roadmaps and milestones.
What’s striking here is the explicit emphasis on organizational redundancy as a strategic hedge. By dispersing AI work across several specialized teams and leaders, Meta reduces the risk that a single bottleneck could derail progress. If one path runs into friction, others continue advancing. It’s a pragmatic acknowledgment that breakthroughs in AI are as much about execution and governance as they are about technical prowess. This mirrors how Zuckerberg has approached other large-scale initiatives: spread risk, empower multiple leaders, and keep the organization agile enough to pivot when needed.
From a broader perspective, the move signals Meta’s intent to scale its AI operations beyond a single executive’s bandwidth. In a field where speed and adaptability often determine competitive advantage, a flatter, multi-team structure could help Meta move more nimbly through experimentation, testing, and deployment at scale. What many people don’t realize is how significant infrastructure decisions—like data pipelines, evaluation frameworks, and developer tooling—are to the pace of AI progress. Meta’s focus on these elements may prove just as important as the model architectures themselves.
As Meta bets on this distributed model, observers may wonder how sustained coordination will be maintained across the various squads. The key, experts suggest, will lie in robust governance without overbearing oversight—aligning goals with clear, measurable outcomes while preserving the autonomy that accelerates progress. If Meta gets that balance right, the company could unlock rapid iteration cycles that outpace competitors relying on more centralized approaches.
In conclusion, Meta’s reorganization reflects a mature, risk-aware strategy to scale AI responsibly and efficiently. By building a network of focused teams led by seasoned engineers and keeping a lean, flat structure, the company aims to push forward on its AI roadmap while maintaining agility in a fast-changing landscape. What makes this development particularly interesting is not just the ambition to reach “superintelligence,” but the deliberate emphasis on infrastructure, governance, and multi-threaded execution as the real levers of progress.
