Where the AI headcount bet actually breaks

Substituting AI for headcount in routine work has a ceiling, and it is closer than the loudest replacement claims admit. The interesting story is not that the bet failed. It is that the company most exposed to it ran the experiment, watched the productivity numbers, and shrank the plan rather than the workforce twice over.

Mark Zuckerberg's Meta explored cutting some teams by as much as 60% under an internal plan Reuters and Yahoo Finance identify as Project OT. The first round landed: 8,000 jobs, roughly 10% of the workforce, gone in May. The second round did not. Internal metrics said the AI tools were not delivering the expected productivity gains, reliability issues piled up, and employees pushed back. Meta shelved the follow-up.

The mechanism is portable. A serious AI substitution bet tests a routine-work threshold under capex pressure and management will. When the productivity numbers arrive, the agency-expanding move is to read them and shrink the plan. Most companies facing the same test will not, because the cost of admitting the ceiling exceeds the cost of pretending it isn't there. Meta's 8,000 cuts and the shelved follow-up together draw the line the AI replacement story keeps talking past: AI can compress the routine. It cannot yet replace the routine worker, and the difference shows up as hiring that slows, not layoffs that announce themselves.

Zuckerberg did not retreat from AI. He retreated from the part of the bet the metrics said no to. That is the model worth carrying forward.