What 'Project OT' Actually Asked of Workers

Meta called it Project OT: an internal plan to cut some teams by as much as 60 percent and replace the removed roles with AI agents working under a small group of human supervisors. Framed as efficiency, the disclosed mechanism required more than swapping software for headcount — it required a continuous supply of exactly the kind of task data that only the workers being replaced could generate.[1]

That supply came from tracking software installed to monitor employees' day-to-day work, which staff discovered was feeding the very systems meant to replace them. The reaction was immediate: internal channels filled with angry posts, mocking memes went up in US offices, and sentiment scores on the affected teams fell from 74 percent to 55 percent.[1]

Two Failures, Not One

Meta canceled a planned November layoff wave once the backlash peaked. Mark Zuckerberg later acknowledged, separately, that the AI agents had not sped up work as fast as he had expected. The reversal reflects two compounding failures rather than one: agents that underperformed on their own terms, and a legitimacy breach once workers learned their labor was training data for their own replacement. It remains possible that the technical shortfall alone would eventually have forced a reversal even without the tracking discovery — but the disclosed timeline shows the legitimacy breach arriving first.[1]

The Policy Lever Neither Story Mentions

Bill Gates's new call for a 'robot tax' targets exactly the incentive gap Project OT exposed. Gates argues that a business can write off a robot's cost immediately while payroll taxes apply to every human employee, tilting the ledger toward automation regardless of whether it actually works — and he wants a tax on automation to fund retraining instead of leaving the transition to each company's internal messaging.[2]

Meta's reversal and Gates's proposal describe the same gap from opposite ends: one shows what happens when a company's internal case for replacing labor outruns both proof and consent, the other names the policy setting that rewards moving fast anyway. The workable test of either claim sits in one place: whether Meta discloses which employees' monitored work trained the agents in Project OT, and what it owes them for it.[1], [2]