The Prophecy and the Prospectus

The prophecy

For two years the people building the large language models said the work was ending. In spring 25 the head of Anthropic told the world AI would be writing ninety percent of code inside three to six months, maybe all of it within a year. Sam Altman spent years warning that whole tiers of white-collar work were about to vanish, entry-level desks first. The message did its job. It sold an intelligence so total that resisting it was pointless and funding it was the only smart move left on the table.

The numbers came in

The real share of AI-written code landed near thirty percent. IBM's chief executive had called it: he put his estimate at twenty to thirty back in Mar 25, and took heat for being a pessimist. The entry-level desks stayed full. This past May, Altman said he was "delighted to be wrong" about the jobs apocalypse he'd forecast, and the head of Anthropic reversed in the same window, now saying automation might widen human work instead of ending it. Nobody resigned over the missed forecast. The prediction just changed, quietly, like a menu.

The calendar

Now line the reversals up against the filings. Anthropic filed confidentially for a public offering on 1 Jun 26, fresh off a raise at a $965 billion valuation. OpenAI filed on 8 Jun 26, carrying its own forecast of a fourteen-billion-dollar loss this year and no profit before 2029. The scary version of the story ran while these companies were raising private money. The softer version arrived when the regulator entered the room, because a prospectus has legal consequences and a prediction on a podcast has none.

The apocalypse got walked back the same season the offering paperwork went in.

Karen Hao's Empire of AI is the book on how those predictions worked as fundraising, and her reporting shows who paid for the belief along the way: data workers in Nairobi and Caracas earning a dollar or two an hour to clean the training data behind the product. The people at both ends of that pipeline, the ones who wrote the data and the ones who scrubbed it, were told the whole time that machines were about to replace them.

What to do with that

The tools are real and worth learning. I use them, and the operator still decides what they're worth. What failed was the forecast that you were obsolete. The companies that preached the end of the coder are hiring software developers as fast as they can right now, while their own products ship bugs. So next time a chief executive predicts the end of your trade, do one cheap thing before you believe him. Check whether he's raising money that week.

Next: the full review of Empire of AI, with the labor reporting that earns the book its title.