What was said
From the moment ChatGPT launched, "mass AI unemployment" was a headline fixture: investment-bank reports of 300 million jobs exposed, warnings that half of entry-level white-collar jobs would vanish within five to ten years, and every round of tech layoffs narrated as AI's doing. The tone sharpened every year, as if the wave were scheduled for next year.
What actually happened
Three years on, the aggregate ledger hasn't moved: US unemployment has stayed around 4%, with no statistically visible "AI job apocalypse." What actually happened is slow, local displacement. The loudest exhibit is Klarna: in 2024 it announced that AI was doing the work of 700 support agents and that hiring had stopped; in 2025 it publicly admitted quality had suffered and started hiring humans again — the most famous single claim of "replacement," retracted by its own claimant. Stanford's 2025 research drew the true shape: in AI-exposed occupations, employment for 22–25-year-olds declined noticeably in relative terms. The harm is real, but its shape is the entry narrowing, not the occupation disappearing.
Why it didn't come true
A job is a bundle of tasks: AI eats thirty percent of it first, and the rest — signing, answering for outcomes, soothing clients, cleaning up — stays human, so AI enters as a tool and rewrites the job from inside instead of carrying it off whole. Second, replacement needs near-100% reliability while assistance only needs 70% to be useful, and diffusion always takes the cheaper road first. Add institutional friction — regulation, unions, contracts, who is liable when something goes wrong — an entire layer that moves an era slower than the technology. Three brakes stacked together turned the tsunami into a seep.
The unexpected part
"Everyone is about to be unemployed" became a business in itself: layoffs dressed as AI strategy read as keeping up with the times rather than failing at business, so companies love saying it; fear drives traffic, so media love reporting it. The prophecy renews daily while reality arrives late, every day. And the people actually hurt — the newcomers who can't get in — don't show up in unemployment statistics or headlines. The lesson cuts both ways: those who cried doom on the aggregate were wrong, and so were those who cried "no impact at all." The truth grew in the layer nobody watches.