Meta Cut 8,000 Workers to Build AI Agents Faster. Zuckerberg Now Says the Plan Didn't Work.
Four months after cutting 8,000 employees and moving 7,000 more onto three new AI teams, Zuckerberg told employees the agentic development trajectory has not moved as expected.
Meta CEO Mark Zuckerberg told employees at an internal town hall on July 3 that the company's Meta AI agents program had not moved as fast as expected. Four months passed. "The kind of trajectory of the agentic development over at least the last four months hasn't really accelerated in the way that we expected," he said, according to a Reuters recording of the meeting. Meta has committed up to $145 billion in AI infrastructure spending for 2026 - all of it predicated on agentic AI becoming a core product layer.
8,000 Layoffs, 7,000 Reassigned - Three New AI Teams, Still Not Fast Enough
Planning for the restructuring began in January and February, driven by what Meta described internally as anxiety about whether the company was moving fast enough to compete. Layoffs came in May - approximately 8,000 workers, about 10% of Meta's corporate workforce - while another 7,000 employees shifted into three newly formed AI divisions: Applied AI Engineering, Agent Transformation Accelerator XFN, and Central Analytics. Zuckerberg called it messy. He acknowledged on Thursday that the reorg was not as "clean" as it should have been, and that executives had "miscalculated on the timing."
Zuckerberg told employees he still expects Meta to see "more significant benefits" from AI investments in the next three to six months. Timing, not strategy - that was his framing. Executives understood the direction, he said, but underestimated how quickly a restructuring of 15,000 people would translate into product output.
Engineers Were Already Calling It a "Soul-Crushing Gulag" in June
Reuters was not the first to surface the internal dysfunction. Engineers said it first. A June 12 TechCrunch investigation described Meta's restructured AI unit as a "soul-crushing gulag" - based on accounts from engineers who reported lack of direction, constant leadership churn, and pressure to demonstrate AI productivity gains the technology could not yet deliver. Zuckerberg's July 3 admission confirms what those engineers said six weeks earlier: the timeline was wrong.
$145 Billion Spent. Four Months In. Nothing Accelerated Yet.
Meta's compute cloud strategy adds context. Reuters reported on July 1 that Meta is exploring selling excess AI computing capacity to third parties - a sign the company built infrastructure ahead of what its own teams can currently absorb. Spending $145 billion on AI while admitting the software execution has not kept pace is the kind of gap analysts will want explained on October earnings calls. Zuckerberg choosing to say it to employees first - before analysts hear it - is either honest leadership or an indication the Q3 numbers will say it regardless.
He expects the next 90 days to look different. Whether that prediction holds - and whether $145 billion in AI infrastructure investment starts generating the agent velocity Zuckerberg originally planned around - is what Meta's Q3 earnings report, due in October, will say.
