Invest2 publishers2 min readPublished Updated
Zuckerberg cancelled Meta's second AI layoff wave hours before the first one landed
Project OT scoped cutting some teams by as much as 60 per cent and handing their work to AI agents. Meta executed a single 10 per cent layoff instead, after internal data showed the agents were not delivering the hoped-for productivity gains.
The Investor · Invest desk
What happened
- A January leadership retreat at Zuckerberg's Hawaii compound produced Project OT, short for Organization Transformation, a plan for an "AI native" Meta.
- Scenario-planning exercises under the project explored slashing the size of many teams across the company by as much as 60 per cent, with the work absorbed by AI.
- The restructuring was designed as two waves, a first purge in May and a second shake-up in November, topped up by closing open roles and pushing out staff judged poor performers.
- Meta called off the November planning on the night of May 19, hours before the first wave, and laid off 10 per cent of its employees the following day.
- By then internal data was suggesting the autonomous agent technology at the heart of the strategy was failing to deliver the productivity gains the plan assumed.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint With no structural payroll reduction to show, the case for Meta's AI spending has to be made out of revenue and product rather than a smaller wage bill, at a moment when some investors were already asking what the spending had bought.
- precedent The most aggressive adopter running the exercise and stopping after wave one makes "we scoped it" the respectable position for other boards, and pushes the execution decision out to whenever the productivity numbers arrive.
- exposure Any forecast that books agent adoption as a headcount saving now has to argue against a large buyer's own internal measurement of the same technology.
- contradiction Meta characterises Project OT as ordinary scenario planning that always included redeployment, which sits awkwardly against an internal projection that the culling would match or exceed a 25 per cent cut.
A substitution trade only clears if the buyer can measure what the machine absorbed, and on Meta's own internal numbers the autonomous agents at the centre of Project OT were failing to deliver the productivity gains the plan had been built on [9]. So compare what shipped with what was scoped. A company-wide layoff of 10 per cent [8] against scenarios that took some teams down by as much as 60 per cent [4] is roughly a sixth of the ceiling, and the comparison is loose in Meta's favour, since one figure is company-wide and the other is team-level [1]. Against the cleaner yardstick, the roughly 25 per cent reduction an HR executive said Project OT would match or exceed [5], the delivered cut comes in at about 40 per cent of it [2]. The second wave never ran [7].
The more interesting term, or rather the one that runs the thesis backwards, is where the freed labour went. Meta says the exercise ultimately moved thousands of employees onto several newly established teams doing priority work [16], and it lists producing training data for its AI models among those priorities [12]. Headcount became an input to the models rather than a cost the models removed, which is a different business than the one the January retreat sketched, where virtual workers would be supervised by small talent-dense cadres of humans [3].
Three readings survive the evidence and they are not equally flattering: that agents cannot yet hold a job; that employees were in open revolt, convinced the AI programme was aimed at them [10], and a chief executive facing that does not need the technology to fail before he blinks; or that this was routine scenario planning that always contemplated redeployment as well as cuts [14]. Reuters, working from scores of internal documents and more than 20 people with knowledge of Meta's inner workings [18], could not determine what changed Zuckerberg's mind [17]. This is probably wrong, but I weight the first reading, for one procedural reason: Meta cancelled the November wave before it had determined how many people overall would lose their jobs [15], which is how a buyer behaves when he has stopped believing the delivery date, not when he is negotiating the size of the order. What would prove it wrong is a second wave reinstated under another name, or an internal agent-productivity series that turns, and the second would show up quietly well before the first.
What to watch
- A second restructuring wave reappearing under a different name, and any figure for the total job losses Meta says it never determined.
- Any internal or disclosed agent-productivity measure that reverses the failing read behind Project OT.
- How large the training-data teams get: if that brief keeps absorbing redeployed staff, headcount is an AI input rather than an AI saving.