Meta sized a 60% team cut on an assumption it had not yet measured

The two reductions landed on the same engineering teams, and they compound rather than add. Take out a tenth, then take a fifth to a third of what remains, and an infra or product team keeps 63 to 72 of every 100 engineers it began the year with, a cumulative loss of 28 to 37 percent ^[17]. That bracket is why the company-wide figure understates the operative one: Gergely Orosz reports some teams absorbed 30 to 40 percent cuts and then struggled to cope with their workloads ^[14]. The number that decides whether an on-call rota holds is the team number, not the corporate one.

The two halves of Project OT also cost differently, and only one of them saves money in the quarter it happens. A layoff takes salary off the books. Moving an infra engineer to data labeling keeps the salary and removes the capability, buying training data with domain knowledge; Orosz says he spoke with teams whose key engineers went to labeling and whose critical domain knowledge went with them ^[15]. Meta's plan folded layoffs and reallocation into the same 60 percent ^[7], which means part of the reduction target was to be met by moves that produced no payroll saving at all.

The board-deck version of this is coherent. AI raises output per engineer, a team of 40 does what 100 did, so you sequence one restructuring in May and another in November ^[10] and bank the difference. It is incomplete in one specific place: the 60 percent rested on the assumption that AI would let the smaller teams operate as well as before ^[8], and that assumption was an input to the number rather than a finding that followed it. Orosz, citing Reuters, traces part of the appetite to executives including chief data officer Alex Schultz and head of product Naomi Gleit, who visited Asia last year and admired how startups there had built their org charts around AI ^[16]. Meta's HR function had already sized the outcome as a bigger layoff than the 2022-2023 rounds that removed a quarter of staff ^[9].

A skeptic will say the reversal is the control working: Zuckerberg called off the November wave hours before the first one ran ^[12], the thesis was tested and pulled, and nothing here travels. What was pulled is narrower than that. The May cut of about 10 percent went ahead ^[2], the labeling reassignments went ahead ^[3], and the thing that stopped November, on the account as reported, was employees in open revolt over exactly this suspicion ^[13] rather than a productivity readout from the teams that had just been cut. Nothing in the record shows a measurement that retired the 60 percent, which leaves the same reasoning available to the next planning cycle.

The causal claim in this story is thinner than the planning documents behind it. Orosz attributes the low morale and the run of outages, including an Instagram zero-auth password reset in which an account could be taken over by asking the AI bot to swap the email address, with Barack Obama's given as the example ^[4], to the same period. That reading is his; the Reuters material he quotes covers the Hawaii retreat, the plan and the cancellation ^[6][5], not the cause of any incident, and no figure is offered for what the outages cost. As a measurement of what AI-driven shrinkage does to reliability, it does not carry the weight being asked of it. As one engineering leader's account of teams that lost a third of their people, it is the most detailed account on offer.

For an executive holding a similar proposal this quarter, the transferable part is not the percentage. Meta's number was fixed at a January retreat ^[6] and the evidence that would have justified it was still outstanding on the night of 19 May, when the second wave was dropped ^[12].