Meta Builds New AI ‘Data Engine’ Teams to Train Smarter Models

Meta is creating a new AI engineering organization as the company expands its efforts to develop more advanced artificial intelligence systems. The new group will focus on building the engineering infrastructure that supports Meta’s training, evaluation, and improvement of its AI models. 

The team will work alongside Meta Superintelligence Labs, the company’s research division responsible for developing Meta’s latest frontier AI models. According to an internal memo, the organization will focus on building tooling, data pipelines, and evaluation systems designed to help models improve through real-world data and feedback. 

Meta builds new applied AI engineering teams

Here is what we currently know about the new teams:

• The Wall Street Journal reported that Meta is forming a new applied AI engineering organization designed to support the development of its advanced AI systems and broader superintelligence initiative. 
• The group will be led by Maher Saba, currently a vice president in Meta’s Reality Labs division, and will report directly to Meta Chief Technology Officer Andrew Bosworth. 
• Business Insider noted that the team will work closely with Meta Superintelligence Labs, headed by former Scale AI CEO Alexandr Wang.
• According to the memo quoted by The Wall Street Journal, the engineering team will build “the data engine that helps our models get better, faster.”

Saba also emphasized that improving AI models requires more than breakthroughs in research or advances in computing power. 

“Building great models isn’t just about researchers and compute; it requires real-world data, feedback, and evals,” he wrote in the memo.

Two teams will support tooling and data pipelines

The applied AI engineering organization will consist of two primary teams focused on supporting Meta’s model development pipeline. 

“It will be made up of two teams, one responsible for building interfaces and tooling and a second responsible for executing tasks, generating data, and providing evaluations that flow back to their modeling teams,” Saba told The Wall Street Journal. 

These systems are intended to create a continuous feedback loop between model performance and the real-world data used to improve it. According to the internal memo, the process helps transform capable AI models into more competitive systems through improved training data and evaluation workflows.

A flatter structure for the engineering teams

Business Insider highlighted that the new engineering organization will operate with an unusually flat structure. Teams could have manager-to-employee ratios of up to 50 individual contributors to one manager, according to employees familiar with the plan. 

This structure aligns with comments Meta CEO Mark Zuckerberg recently made about elevating individual contributors and flattening engineering teams. In recent remarks to investors, Zuckerberg said the company is seeing some projects that once required large teams, now being completed by a single highly skilled engineer. 

Some other technology companies also use relatively flat leadership structures. Nvidia, for example, is known for having a large number of direct reports under CEO Jensen Huang.

Learn more about how AI is influencing tech workforce changes after Jack Dorsey reduced Block’s workforce by nearly half.

The post Meta Builds New AI ‘Data Engine’ Teams to Train Smarter Models appeared first on eWEEK.