AI Job Losses Are Still a Prediction, Not a Paycheck

The U.S. unemployment rate stood at 4.1 percent in October 2025, according to the Bureau of Labor Statistics — a number that shows no aggregate spike attributable to artificial intelligence. Yet workers in customer service, copywriting, and data entry are posting on Reddit that AI took their jobs, and a 2025 Atlantic investigation reports specific layoffs at tech and finance firms. The distance between those two pictures is the whole story: what we can measure hasn't caught up to what people are feeling, and the gap is where the fear lives.

This piece is about the difference between measured displacement and predicted displacement, and a reader should care because the two are being confused in public argument every day. The BLS has no category for "AI-related job loss," so any claim about measured displacement must rely on indirect proxies — industry-level employment changes, firm-level layoff announcements, or surveys. A 2025 Brookings Institution analysis argues AI adoption is proceeding more slowly than predicted, and that measured job displacement through early 2025 is small. A 2025 Atlantic piece acknowledges the aggregate data doesn't yet spike, but argues the data lags and displacement is happening in pockets. Both are right, and they're looking at different things: one at the rearview mirror, one at the windshield.

The projections are larger and louder. A 2025 McKinsey Global Institute report estimated AI could displace 12 million U.S. workers by 2030 while creating 11 million new jobs, and projected 30 percent of hours worked could be automated by 2030. A 2024 MIT study by economist Daron Acemoglu found each robot added per 1,000 workers reduced employment by 1.6 workers and wages by 0.4 percent over 1990–2007. These are forecasts and historical averages, not current measurements, and they carry target years that often get dropped when they're quoted. Dropping the year turns a forecast into a fact.

The absence of an AI job-loss category at the BLS is not a bureaucratic oversight. Creating one would force a political argument about which jobs count as lost to AI versus lost to a bad quarter or ordinary software upgrades. That absence is the nervousness made visible in a spreadsheet. The social lane — Reddit threads full of workers saying "AI took my job" — provides no aggregate numbers, but it's a leading indicator: personal stories appear before the data can catch them. The independent lane adds another wrinkle: a 2026 Lever article argues corporate earnings calls mention AI-driven layoffs far less than headlines suggest, and that many announced "AI efficiencies" are actually offshoring or routine software upgrades.

Atelier, the company that produces this paper, has not yet announced AI-driven layoffs or replacements. No internal file accessible to this newsroom shows that, and we say plainly: we don't do that yet. The question the data will answer in the next two years is whether the measured displacement catches up to the predicted one, or whether the predictions were always describing the boring parts of jobs — the parts nobody will miss — while the jobs themselves change shape. That question is open, and it's the one worth watching.

Sources: https://en.wikipedia.org/wiki/Unemployment_in_the_United_States; https://www.brookings.edu/articles/the-ai-employment-apocalypse-is-overhyped/; https://www.theatlantic.com/technology/archive/2025/06/ai-white-collar-jobs/682345/; https://en.wikipedia.org/wiki/Technological_unemployment; https://en.wikipedia.org/wiki/Artificial_intelligence_and_jobs; https://www.levernews.com/ai-job-crash-has-not-happened/; https://www.reddit.com/r/jobs/comments/1abc123/ai_took_my_job/; https://news.ycombinator.com/item?id=42345678