In 2025, the two most influential voices in artificial intelligence told us our jobs were in danger. Sam Altman warned that whole categories of entry-level white-collar work could disappear. Dario Amodei went further, suggesting AI could eliminate up to half of those jobs within a few years and push unemployment into the double digits — and framing it as his duty to say so out loud.

A year later, the message has reversed. Altman now says he was "pretty wrong" — even "delighted to be wrong" — about how fast those jobs would vanish. Amodei has recast automation as a multiplier of output rather than a destroyer of jobs: automate most of a task, the argument now goes, and people simply move up to the part that remains.

It is tempting to read this as two honest forecasters updating on new evidence. The displacement they feared has not shown up in the data, and that gives them cover to walk it back. A second reading points to the public mood — booed commencement speakers, protests outside data centers, a hardening hostility toward AI. Both readings contain some truth. But neither is the driver. The data offered cover, the backlash created pressure. What turned cover and pressure into a change of message was something more fundamental.

The tone did not soften because these two men changed their minds. It softened because their audience and their business model changed — and put enterprise customers at the center.

Why doom paid in 2025

A year ago, both companies were raising enormous sums in a scaling-law mindset: the bigger the model, the bigger the leap, the bigger the cheque. In that world, the more transformative — and the more dangerous — AI appears, the easier it is to justify the valuations, the capital expenditure, and the regulatory attention. Disruption was not a warning against the pitch. It *was* the pitch.

For Anthropic in particular, danger framing did double duty. A company that warns loudest about the risks positions itself as the responsible actor that regulators should trust to build. Fear, in 2025, was a strategic asset.

Why reassurance pays in 2026

Now the customer has changed. Both companies are moving toward public listings at roughly trillion-dollar valuations. "We may cause mass unemployment" is not a story you tell cautious institutional investors — public hostility adds a reputational risk that they have to neutralize. More importantly, the revenue that has to justify those valuations no longer comes from venture rounds. It comes from enterprises.

And enterprises buy on different values. In the corporate world, "trust" and "reliability" are what customers actually pay for. No CIO — and no employee asked to adopt these tools — buys a product sold as "this will replace you". So the message is being optimized for the buyer. The same technology that was sold as a revolution is now sold as a dependable productivity tool.

The same pivot, two different exits

What makes this moment revealing is that Altman and Amodei are not softening in the same way.

Altman's brand was always abundance — a future of plenty made possible by AI. For him, a soft climbdown costs almost nothing. He can simply say he was wrong about the timing and move on; the optimism survives intact.

Amodei's brand *was* the warning. The white-collar reckoning was his signature claim, the thing that set him apart from peers who only whispered it. He cannot quietly retract it without losing the credibility it bought him. So instead of retracting, he reframes: automation that once "destroyed" jobs now "multiplies" output. Same destination, different exit — keep "AI is transformative," drop "it is coming for your job".

We have seen this movie before

None of this should surprise anyone who has watched a technology hype cycle run its course. As I argued in my book *Making Sense of Generative AI*, the same patterns repeat from the Dot-Com bust to the green-tech bubble to the smartphone era. Two of those patterns are on full display here.

First, timelines are always wrong. As Bill Gates observed, we overestimate the change of the next two years and underestimate the change of the next ten. The 2025 predictions were classic hype-cycle narratives — the promised upheaval was supposed to arrive too fast and too clean. What we are watching now is the predictable correction of the *timing*. It is not evidence that the long-run impact will be small; it may yet be larger than anyone is claiming today.

Second, business models matter more than technology. The companies that ultimately win these cycles are rarely the loudest. They are the ones with a credible path to a sustainable business. The shift in how Altman and Amodei talk is simply that path becoming visible — the moment a research lab starts behaving like a company that has to sell something.

What to watch

The lesson for anyone trying to make sense of AI's leaders is to follow the money, not the mood. When the tone changes, look at who they are trying to sell to and what they need to raise. The forecast is downstream of the business model, and it always has been.

So the next time these narratives shift again — and they will — the useful question is not "what do they now believe?" It is "what do they now need to sell?"

References: