Project OT put dates on the layoffs before it put numbers on the agents

The dependency chain ran backwards. A 60% reduction in a team's size is not a claim about a model, it is a claim about span of control: that one remaining human can supervise enough virtual work to cover the output of the colleagues who left ^[2]. That claim is testable, but only one way, by running it on a few teams and measuring what comes out the other end. Project OT fixed the calendar first, a May purge and a November shake-up, and left the measurement to catch up ^[6].

It did not catch up in time. Roughly four months separate the January retreat from the reversal on the night of 19 May ^[12]. Inside that window, internal data indicated the autonomous agent technology at the centre of the strategy was not delivering the productivity gains the plan assumed ^[15]. Two other pressures were live at the same time: employees had concluded the AI transformation was partly aimed at replacing them ^[14], and some investors were asking what Meta had to show for its AI spending ^[16].

A skeptic would say Meta got what it wanted regardless. It cut 10% of staff on 20 May ^[7], and it says the exercise moved thousands of employees onto newly established priority teams ^[9]. That is fair, and it is also the diagnostic. Redeployment and open-role closures need no agent autonomy at all; the part that was cancelled is the part that depended on virtual workers doing the work of departed staff.

What the record does not support is a verdict on model capability. Reuters reported it could not determine what prompted Zuckerberg to change course, or what Meta's current workforce plans are ^[10]. Meta's own account is that scenario planning is not commitment, that it never intended to lay off 60% of its workforce, and that leaders killed the second wave before anyone had settled on a total number of job losses ^[4]. Read the sequence, not the technology: the company committed to two dated waves before it held evidence that the substitution worked, and the second date was the one it could still take back.

The gap between the projection and the outcome is the size of the retreat. One HR executive had the culling landing at or above Meta's roughly 25% reduction of three years ago ^[5]; what actually shipped was 10%, about two-fifths of that benchmark ^[13]. The cost of the exercise sits in what it taught the workforce. Any team at Meta now offered an agent as leverage has already been shown a planning document in which agents were the argument for cutting the team ^[1], and the next attempt at this pays that price before it starts.