There is a tidy story going around: big tech laid people off because AI could do their jobs, and now, having found AI cannot, the companies are quietly hiring them back. It is a satisfying narrative. It is also, for the hyperscalers specifically, not what the primary record says. The truth is more precise, and more useful to anyone making their own AI-and-headcount decisions.

People walking through a dark corridor lit by neon arches, suggesting a workforce in transition.
The layoffs are real and large. What the companies say caused them is not the story the headlines tell.

What happened

Amazon cut about 14,000 corporate roles, announced on 28 October 2025 in a memo from SVP Beth Galetti, with further reductions reported toward 16,000 into early 2026. The striking part is the stated reason. On Amazon’s Q3 2025 earnings call, CEO Andy Jassy said the cut was “not really financially driven, and it’s not even really AI-driven, not right now at least,” and added, “Really, it’s culture.” Amazon’s own framing was too many layers and too much bureaucracy after a hiring surge, not AI replacing those workers. Amazon keeps hiring heavily for AI and data-center roles while trimming corporate: that is reallocation, not AI substitution.

Meta cut about 8,000 roles, roughly 10% of its corporate workforce, in May 2026, framed by a memo from Chief People Officer Janelle Gale around a “flatter structure” of smaller teams. Again, the detail undercuts the simple story: Meta reassigned around 7,000 people into new AI groups (one named Agent Transformation) and cancelled roughly 6,000 planned hires. An earlier October 2025 round cut about 600 in Meta Superintelligence Labs. These were cuts to fund and refocus around AI, not cuts because AI had replaced the work.

The one real admission, framed precisely

There is a genuine “AI is underdelivering” moment, and it is worth quoting exactly. At an internal town hall on 2 July 2026, reported by Reuters from a recording, Mark Zuckerberg told staff that AI agent development over the prior four months “hasn’t really accelerated in the way that we expected,” that the reorganisation had not been as “clean” as hoped, and that the bets “haven’t come to fruition yet,” while saying he expects benefits within three to six months.

Read that carefully. It is an admission that AI progress is slower than hoped. It is not an admission that the layoffs were a mistake, and it is not an announcement of rehiring. Those are different claims, and only the first one is supported.

What the “hiring back” story actually rests on

The genuinely confirmed reversals of AI-for-headcount cuts are not the hyperscalers. They are Klarna, which said it “went too far” and began rehiring human customer-service agents after quality dropped, and Commonwealth Bank of Australia, which reinstated 45 roles after a union showed call volumes were actually rising. Both are documented, with sources, in AI in the real world . The broader “companies are reversing AI layoffs” claim comes mostly from outplacement-firm surveys and analyst forecasts (predictions that some share of AI cuts will be reversed), not from Meta or Amazon confirming they rehired anyone because AI failed.

For context on scale, the MIT Project NANDA report “The GenAI Divide: State of AI in Business 2025” found roughly 95% of enterprise generative-AI efforts show no measurable profit-and-loss return. That is a striking number, and also a self-reported, non-peer-reviewed industry survey, so treat it as a signal of difficulty, not settled science. It too is discussed in AI in the real world .

Why it matters for builders

The useful lesson survives the myth-busting, and is arguably stronger for it. The confirmed reversals, Klarna and Commonwealth Bank, happened because those companies cut human capacity before the AI had proven it could carry the load. That is the mistake to avoid: treat “AI will let us do the same work with fewer people” as a hypothesis to test with real quality metrics, not a fact to reorganise around in advance.

And notice what the hyperscalers are actually doing. They are not betting that AI removes the need for people. They are cutting layers to move faster and pouring the savings, and thousands of redeployed staff, into building AI. Even Meta, mid-build, says the agents are slower than it hoped. If the companies spending hundreds of billions on AI are pacing their own headcount bets on measured results, that is the pattern to copy. See are AI agents replacing SaaS for the augmentation-versus-replacement argument in full.

Sources

Further reading