Meta abandoned its most aggressive AI-driven staffing cuts after checking its own coding data

Meta measured inputs. The comparison inside the company, as Fortune describes it, ran from dramatically increased AI-assisted coding to gains in user-facing work, and the data raised questions about whether the second followed the first. The most aggressive workforce-reduction plans were then dropped ^[5]. Any delayering case built on agent productivity has to show the same link.

Fortune gives no headcount, no cost, no date, and no account of what became of the rest of Project OT ^[6]. The design is on the record: agents taking over much of the work performed by thousands of employees, teams becoming dramatically smaller, some layers of management disappearing, on the premise that smaller, more "talent-dense" groups could do what much larger teams did ^[1][2][7]. Those details come from a Reuters investigation, as reported by Fortune ^[1].

The one operating change the account documents is at the top. Zuckerberg has been using what Meta calls a "CEO agent" to retrieve answers that previously required going through several layers of staff ^[3]. Fortune's authors split what those layers were doing into coordination friction, the scheduling and documentation and data reconciliation and approvals that AI can compress, and cognitive friction, the resistance between ideas held by people who read the same evidence differently ^[4]. They wrote that for an individual thinker, AI "should function more like a sparring partner than a mirror" ^[8]. Their worry about the agent is specific: faster access to the organisation, less exposure to the people who would have challenged its assumptions ^[9].

Meta abandoned the most aggressive plans, so a moderate version of Project OT may still be running and simply has not surfaced ^[5]. Coding output may also lead shipped output by a couple of quarters, in which case the internal data was read too early to settle anything.

I would still take the retreat at face value, because of who ran it. The company with the models, the instrumentation and every incentive to prove that agents can replace management layers looked at its own numbers and kept the layers ^[5]. For now the agent capacity Project OT assumed sits inside existing teams. What would move me: a disclosed link between AI-assisted code volume and shipped user-facing work, or a Meta headcount reduction that Meta itself attributes to agents ^[11].