Capacity planning is arithmetic before it is strategy. Cut a team by 60% and you keep 40% of the people [2], so output per remaining person has to reach 2.5 times its old level before team output holds flat [1]. Project OT carried that multiple as its premise: according to Reuters, the plan was underpinned by the assumption that AI would let the smaller teams operate as well as before [3]. That's a projection standing in for a measurement.
The version that shipped gives you something to check it against. Meta laid off circa 10% of engineers in May [7], and around the same time infra and product teams handed over a further 20-30% of their engineers to data labeling for AI training [8]. Compound those and an affected infra team is running on roughly 63 to 72 percent of its original engineers [2], which is between 47 and 62 percent of the reduction Project OT wanted [3]. Gergely Orosz reports that teams which took 30-40% cuts struggled to cope with their workloads [10].
Reuters attributes part of the enthusiasm to travel: Chief Data Officer Alex Schultz and Head of Product Naomi Gleit visited Asia last year and admired how startups there had built their organizational charts around AI, and Meta commissioned its own research into it [13].
For a 60% number to transfer to your org, several things have to hold at once. The work removed has to be work a model actually finishes, not just drafts. Reviewing the model's output has to cost less than doing the work itself, including the reviews that catch nothing. And whatever's left of the 40% still has to cover the on-call rotation and the people who know why a subsystem is shaped the way it is. Orosz says he spoke with teams whose key engineers went to labeling, developers with critical domain knowledge that was lost when they left [11]. The AI capacity Meta was building got paid for out of the same headcount the AI was supposed to give back.
The Instagram failure is the legible part. Orosz writes that the period produced low morale and a string of embarrassing outages, including a zero-auth password reset bug that let anyone's account be taken over just by asking the AI bot to swap in a different email address [9]. That causal link is his reading, not something Reuters established, and outage attribution is hard to prove from outside. The shape of the bug still tells you where the exposure sits. An email change on an arbitrary account was honoured with no evidence that the requester owned it. Authorization code is the worst place to be short of reviewers, because the failure mode there is takeover rather than latency.
The May 19 sequence is the part other executives will end up citing. Hours before the first wave, Zuckerberg called off planning for the November round, and the 10% layoff went ahead the next day [6]. Employees by then were in open revolt, convinced the AI transformation initiatives were partly aimed at replacing them [12]. The account does not give a reason for the reversal [6]. Orosz's own estimate is that a 30-40% company-wide layoff had been planned [14], and Meta's HR projected something larger than the 2022-2023 round that removed 25% of staff [4]. Two months before the Reuters report, Orosz had already asked why Meta appeared intent on dismantling an engineering organization at a company posting record revenue and profits [15]. The Project OT plan reads like the answer to that question.
