For the past year Mark Zuckerberg has been Silicon Valley's most aggressive convert to the superintelligence cause — pouring billions into compute, poaching researchers with eye-watering pay packages, and tearing Meta's AI organisation apart to rebuild it. In an internal memo to staff on 12 June, he conceded the obvious: a lot of it has hurt, and not all of it has worked.

"Given the complexity of these changes, we've made mistakes and will almost certainly make more," Zuckerberg wrote, in a memo first reported by Reuters and Bloomberg. The headline reassurance for nervous employees: he does not expect more company-wide layoffs in 2026, and is "focused on providing as much stability as possible" from here.

The upheaval he's apologising for

The candour lands against a brutal backdrop. Meta cut roughly 8,000 jobs in May, part of a sweeping reorganisation that pulled staff into a new Applied AI Engineering unit — a posting some employees reportedly nicknamed "literally the gulag." Researchers were reassigned to the grind of training models; teams were reshuffled; the culture, by the memo's own implicit admission, frayed.

So the rest of the memo reads like a morale-repair kit. Zuckerberg promised to increase budgets for team offsites and company events, run a company-wide hackathon in July, and restore assigned desks in many offices by the end of the year — a small, telling detail after a period in which even a fixed place to sit had apparently become a casualty. He also said Meta would try to find new roles for staff who'd been moved onto model-training work.

Why it matters beyond Meta

It's easy to read this as HR housekeeping. It isn't. It's the first real admission from one of the "spend whatever it takes" camp that the human and organisational cost of an all-in AI pivot is showing — and that you can't restructure your way to superintelligence without breaking things you then have to fix.

Zuckerberg isn't retreating from the AI bet; the spending and the ambition remain enormous. What he's conceding is the execution — that moving this fast, this hard, generated genuine mistakes and genuine damage, and that stability now has a value of its own. For a company that spent the year signalling relentless momentum, "we got things wrong and people are exhausted" is a notable change of tune.

The honest takeaway: the AI talent-and-compute arms race has a back office, and the back office is tired. Meta is the biggest player to say so out loud — and probably won't be the last.