Zuckerberg killed Meta's November AI reorg hours before the May layoffs ran

A 60 percent team reduction is a claim about a substitution ratio. It asserts that an agent, plus a human to supervise it, covers work that used to take some larger number of people, at quality you would ship. According to Reuters, Meta's OT documents put that ratio on a calendar before there was a measurement to support it: wave one in May, wave two in November ^[6].

The measurement went the other way. Internal indicators showed the autonomous agents Meta was betting on were not producing the expected productivity growth ^[9]. Employees had already objected openly, reading the transformation as partly an exercise in replacing them ^[8]. Zuckerberg reversed on the evening of May 19, hours before the first wave, and the 10 percent reduction ran the next day with no November phase behind it ^[7]. Phase two was dropped before phase one had produced a single day of outcome data ^[18].

Meta's account of the same documents is narrower. The company confirms OT existed and calls it a year-long program of cost reduction, team restructuring and moving people onto priority work, including preparing training data for AI models ^[10]. It says some scenarios did reach 60 percent, but per team, not across the whole company, and that leaders cancelled the second phase before any total layoff number was set ^[11]. In its statement, Meta says teams were asked to model the consequences of transfers, closed roles and cuts, that thousands of people moved into several newly created teams, and that it "did not implement all scenarios considered during this analysis and never assumed they would be fully implemented" ^[12]. That reading makes the 60 percent an input a team was handed, arrived at before anyone measured what agents could actually absorb. Also worth noting: the priority work people were moved onto included feeding the models their training data ^[10].

One HR leader expected the total to match or exceed the wave three years ago, when Meta cut roughly a quarter of staff ^[5]. After a 10 percent reduction, reaching a cumulative quarter takes about 17 percent of whoever is left: 1 - 0.75/0.90 = 0.167 ^[19]. That is what came off the calendar, and it came off on the strength of internal indicators this account does not quantify ^[20].

The reversible version of this experiment is already in the record. Ime Archibong's team built five small technical groups, two or three engineers and a designer each, which were to abandon the traditional six-month cycle ^[17]. Alex Schultz and Naomi Gleit went to Asia to study startups whose org structure was built around AI from the start ^[15], and Gleit said practices from Meta's Singapore office inspired teams in California and New York, with some ideas coming from employees rewriting their own workflows ^[16]. A four-person pod can be judged inside a quarter; a 60 percent cut across many teams cannot be assessed on anything like that timescale.

For the 60 percent to transfer to your org, you would need per-step agent success rates high enough that reviewing the output costs less than doing the task, measured on your workflows rather than a vendor's. Meta measured on its own and stopped. Copy the stopping, hours out.