Meta Tried to Shrink Teams Around AI. Why Did the Plan Break Down?

Meta was trying to redesign the organization, not merely deploy AI tools

In early 2026, Meta launched an internal effort known as Project OT, short for Organization Transformation. According to Reuters, the goal went beyond giving employees AI tools. The company explored smaller teams that could supervise AI agents and handle more work with fewer people.

Scenarios included reducing some teams by as much as 60%. That figure did not mean cutting 60% of Meta's total workforce; it referred to selected team-level scenarios that could include layoffs, redeployment and eliminating open positions.

The plan stalled when the expected productivity case was not yet strong enough

Meta cut about 10% of its workforce in May, but cancelled a second restructuring round that had been planned. Reuters reported that internal data showing AI technology was not delivering the expected productivity gains, together with employee alarm, contributed to the pullback.

More AI-assisted activity is not automatically the same as more business output. Changes in code or work volume still have to translate into features delivered, reliability maintained, incidents avoided and useful outcomes for customers.

Faster generation can increase the cost of verification and accountability

Reuters also described reliability and security problems as AI agents and coding tools were used more heavily. The lesson is not that AI lacks value. It is that faster generation creates a greater need to decide who validates the output, who owns failures and which tasks are safe to delegate.

A job rarely maps cleanly to one AI system. Most roles mix generation, judgment, approval, coordination, security, customer context and exception handling — tasks with very different automation profiles.

When employees distrust the transition, the AI experiment itself changes

Project OT also ran into anxiety around further layoffs and opaque communication. If employees believe AI-usage data may become evidence for eliminating their jobs, the conditions for experimentation, knowledge sharing and adoption change.

AI transformation is therefore a trust problem as well as a technology problem. Companies need to explain what will be automated, what remains human-owned, how productivity will be measured and where saved time is supposed to move.

The first question may not be 'How many people can we remove?'

Labor-cost reduction is a legitimate part of the AI investment case. But organization design works better when companies first identify which tasks can be automated, which tasks become more valuable with AI assistance, and where human judgment remains non-negotiable.

If the headcount target comes first, accountability can disappear before the technology is ready. If tasks come first, the same technology can instead increase capacity or enable work that teams previously could not perform.

AI may still make teams smaller. The sequence matters.

Meta's pullback is not evidence that AI will never reduce employment. As the technology improves, some teams may become smaller and some roles may decline. Meta itself continues to invest heavily in AI infrastructure.

But organizations do not transform automatically because the tool changes. Work changes first, roles change next, and only then does the sustainable size of the organization become clearer.

The question ‘How many people can AI replace?’ sounds technical. Inside a company, it quickly becomes a harder question: which work can safely be owned by whom?