Meta explored rebuilding parts of its workforce around smaller teams and AI agents — but internal data reviewed by Reuters suggested that dramatically increasing AI-assisted output did not automatically translate into dramatically greater productivity.
WHAT’S HAPPENING
Earlier this year, Meta launched an internal initiative called Project OT — Organization Transformation — aimed at redesigning parts of the company for an “AI-native” workplace.
The idea went considerably further than simply giving employees AI tools.
According to a Reuters investigation based on internal Meta documents and more than 20 people familiar with the effort, some planning scenarios envisioned replacing traditional teams with much smaller groups supported by AI agents. In the most aggressive scenarios, the size of certain teams could have been reduced by as much as 60% through a combination of layoffs, redeployments and eliminating open positions.
Meta confirmed Project OT existed and confirmed those scenarios, while stressing that it never planned to eliminate 60% of its entire workforce.
Meta proceeded with roughly a 10% workforce reduction in May, but plans for a second company-wide restructuring wave later in the year were abandoned. Reuters reported that internal concerns were building at the same time about whether AI agents were delivering the productivity improvements executives expected.
WHY IT MATTERS
This may be one of the more important AI workplace experiments happening inside a major corporation because it tests a popular assumption:
If AI lets employees produce dramatically more work, the company should need dramatically fewer employees.
Meta’s internal experience suggests the equation may not be that simple.
Reuters reported that internal code changes increased 220% year over year as employees used more AI.
But changes resulting in new or improved features actually reaching users increased only 36%.
More production was happening.
The useful output was not increasing at the same rate.
That’s a major distinction.
AI can increase activity without increasing value at the same speed.
WHO BENEFITS
Workers who know how to combine AI with judgment, product knowledge and accountability may become more valuable — not less.
If companies discover that simply replacing people with agents creates more code, more reports or more activity without equivalent improvements in results, then the advantage could shift toward employees who can determine:
What should AI actually be doing?
That makes human judgment, supervision and domain expertise part of the productivity equation rather than something automatically removed from it.
WHO LOSES
Companies that mistake AI output for AI productivity could find themselves cutting organizational knowledge faster than the technology can replace it.
Reuters reported that Meta’s internal systems also showed warning signs from increased AI-generated activity. Internal posts said major technical and security incidents rose 40%, while time spent responding to those problems increased 70%.
Meta declined to comment to Reuters on those specific internal figures.
The risk is straightforward:
A machine capable of producing work faster can also produce mistakes, complexity and cleanup work faster.
WHAT HAPPENS NEXT
Meta hasn’t abandoned AI-driven organizational change.
Mark Zuckerberg said internally in July that AI-agent technology had not accelerated as quickly as he expected, while saying he believed improvements could begin producing greater benefits within months.
And Meta’s broader public position remains strongly optimistic. Zuckerberg has argued that AI can expand individual capabilities, enable smaller teams and businesses to accomplish more, and potentially create more employment over time even if individual companies become smaller.
So Project OT isn’t evidence that AI won’t change employment.
It may be evidence that the transition is going to be messier than simply replacing workers with agents.
The first major workplace lesson of the AI era may be this:
Doing more work with AI is easy to measure. Proving that the extra work actually matters is much harder.
