OpenAI's Sam Altman and Anthropic's Dario Amodei spent the past year warning that AI would hollow out white-collar work. This week both of them said they got it wrong. Speaking to a Commonwealth Bank of Australia conference in Sydney on Tuesday, Altman said he was wrong about the economic fallout, while Amodei has swapped his talk of mass unemployment for a story about workers becoming more productive.
The timing is doing a lot of work
Both companies are reportedly heading toward IPOs this year, each at a valuation Fortune pegs near a trillion dollars. A jobs apocalypse is a hard thing to pitch to public investors, so the sudden optimism is tough to take at face value. Fortune first reported the reversal, lining up last year's warnings against this week's mea culpas.
A year ago Altman was telling his brother Jack on a podcast that a lot of jobs would simply go away. Amodei went further, claiming AI could eliminate up to half of all white-collar roles within five years. Neither wants to own those numbers now.
Altman's inbox changed his mind
What moved Altman wasn't a study. It was his own email. In the interview, he described handing his Slack and inbox to AI, then quietly going back to answering everything himself.
His own communication, Altman said, is "not something that I can imagine outsourcing to an AI anytime soon," which is a striking line from the man whose company sells it.
He also wanted some credit for the fear. People tell him he could have spared the world a lot of doom and gloom, he said, but the risk looked real to him at the time. Maybe. It also made for useful press during a fundraising cycle.
What Amodei is selling now
Amodei's new framing leans on an old idea. He argues that automating 90% of a job just pushes workers onto the harder remaining tenth, lifting output rather than cutting jobs. He and others tie this to Jevons paradox, the nineteenth-century observation that cheaper coal led to more coal burned, not less.
Apollo economist Torsten Slok makes the same bet from the data side. "Lower cost per interaction does not mean fewer interactions," he wrote in a recent blog post, citing call centers and radiology as fields that kept hiring straight through automation. Clean story. Also one that can't really be disproven on the timescale a roadshow cares about.
So what does the data say?
Messier than either side wants. Tech layoffs passed 115,000 through May, already near the 124,000 logged across all of 2025, with Meta and Amazon among the companies naming AI as a reason. That figure is self-reported, though, and AI is a tidy label for cuts a company already wanted to make.
Pull the other thread and it frays too. The Yale Budget Lab has found no meaningful shift in the occupational mix or unemployment duration for the jobs most exposed to AI since ChatGPT arrived in late 2022. Its researchers stress the work isn't predictive and that they're still watching month to month. Nobody has the displacement signal yet, which is not the same as saying it will never come.
Goldman Sachs CEO David Solomon, who never bought the panic, made the long-view case in an op-ed: US civilian employment has grown 145% since 1962 despite wave after wave of automation. He didn't reverse anything, because he never sounded the alarm.
Yale's team plans to keep publishing monthly. The real test of whether any of this is happening shows up in their next reading, not in whatever the two CEOs tell investors on the way to the public markets.
