Anthropic’s own data shows ai is over-hyped: top-paid jobs fear bots that still can’t flip burgers
The robots aren’t coming for your job—at least not today, and not if you cook fries for a living. A study released by Anthropic, the ChatGPT rival founded by OpenAI defector Dario Amodei, reveals a canyon-sized gap between what large language models could do in theory and what they actually do inside real companies.
The numbers that deflate the bubble
Across 22 occupational clusters, the median “theoretical exposure” to ai sits at 57 %. Flip to the “observed” column and the figure collapses to 12 %. Software engineers top the fear chart—94 % of their tasks look automatable on paper—yet only one third of those tasks have been handed to Claude or its cousins. Lawyers, bankers and data-entry clerks follow the same pattern: high on hype, low on deployment. Meanwhile, line cooks, motorcycle mechanics and farmhands register near zero in both columns. The machines, it seems, still can’t julienne a carrot.
Why the mismatch? Tokens, not talent, drive revenue. Every prompt sent to Claude consumes cloud credits that Anthropic sells by the million. The more existential the press release, the faster corporate budgets open. Amodei’s Davos warning that ai will “write all software within a year” landed one week before Anthropic closed a funding round at a $ 60 billion valuation. “Fear is a monetisable API,” says Jonatan Amenedo, whose fintech clients track cloud invoices. “Generate FOMO in the C-suite and the POs sign themselves.”
Deskilling in slow motion
The hidden casualty isn’t unemployment—it’s eroding expertise. Junior analysts now paste spreadsheets into a chat window instead of learning pivot tables themselves. Managers who once caught errors by instinct outsource that muscle to a summary button. Bruno Marcial, ai director at Madrid’s EBIS school, calls it “silent deskilling”: the worker keeps the title, loses the craft. U.S. labour data already show entry-level hiring for 22- to 25-year-olds stalling even as senior postings rise. The pipeline is clogging at the source.
Anthropic’s report ends on a careful note: no statistical spike in lay-offs since ChatGPT arrived. But absence of evidence is not evidence of absence. Wages for senior coders are flat in real terms, and contingent-work platforms are filling once-permanent roles with freelancers who rent ai by the hour. The job didn’t disappear; it was liquefied.
Revolution or racket? The answer depends on which spreadsheet you zoom in on. Wall Street sees a $ 4 trillion generative-ai TAM. The fry cook sees a thermostat that still can’t tell when the oil is too hot. Somewhere between those two truths lies the future of work—half automated, half make-believe, and already priced into next quarter’s earnings.