Meta's AI Productivity Paradox: More Code, Fewer Features

Meta's ambitious push to become an "AI-native" organization has surfaced a counterintuitive challenge that should concern every technology leader: more employee productivity doesn't automatically translate to better business outcomes.

According to a Reuters investigation, Meta explored scenarios this year under an internal initiative called Project OT that examined how smaller teams could oversee AI systems performing significant portions of work. Some scenarios considered workforce reductions of up to 60% in certain teams. The company proceeded with cuts affecting roughly 10% of its workforce—about 8,000 employees—in May, but canceled a planned second round.

The productivity disconnect

The internal data tells a revealing story. Code changes across Meta's platforms and infrastructure jumped 220% in a single year. Yet the number of those changes that actually became new or improved features for users increased by only 36%. The company also reportedly saw more significant technology and security incidents, along with increased time spent addressing them.

These figures don't indicate AI underperformance. They demonstrate something more fundamental: AI has moved the bottleneck rather than eliminated it. When ten potential solutions can be generated in minutes instead of one, someone still must decide which solution is correct. When vastly more code can be written, someone must determine how it integrates into the product.

Why it matters

This gap between individual output and organizational throughput represents a critical challenge for 2025 planning cycles. Companies investing heavily in AI tooling may find that their constraint isn't generation capacity—it's judgment, prioritization, and the ability to absorb increased output. The scarce resource has shifted from execution to decision-making.

What becomes valuable

The shift has immediate implications for talent strategy. By late 2026, the ability to work with AI will likely be a baseline requirement across many technology roles rather than a differentiator. What will distinguish high-value employees is their capacity to convert AI-generated abundance into business results—defining the right problems, distinguishing signal from noise, understanding broader implications, and connecting across disciplines.

This context makes Bill Gates' recent commentary on "Human Reserved" jobs particularly relevant. The Microsoft founder warned last week that many jobs could disappear permanently due to AI and suggested society may need to designate certain roles as deliberately human-staffed even when machines become capable. But Meta's experience suggests organizations first need to identify what remains "Human Required"—the judgment and synthesis work that becomes more critical as execution costs approach zero.

The non-linear transition

Meta's case challenges the assumption that doubling employee productivity enables cutting teams in half. AI will certainly make tasks redundant and reshape professions, but the transition appears far less linear than early projections suggested. Companies that succeed may not be those replacing the most employees, but rather those first recognizing that building organizational capacity to handle increased output is now the binding constraint.

The details were first reported by Reuters in their investigation of Meta's internal AI transformation efforts.