Sam Altman now says he’s “fairly confident” AI has created more jobs than it has eliminated so far. It’s a sharp reversal: just months ago he warned that entire professions would disappear “fast enough to be scary.” Dario Amodei at Anthropic has also walked back similar claims, shifting from “AI will take a large share of entry-level office jobs in a very short timeframe” to “automation is a productivity multiplier, not a job killer.”

Why this matters to you. If you work with AI or need to explain it to your team, this isn’t reassurance: it’s a signal that the public framing is changing without the numbers to back it up. Recent studies (Yale Budget Lab, multi-university research on programmers and copywriters) show no significant impact from AI on measured productivity or aggregate employment. The downturn among programmers and copywriters began in early 2022, months before ChatGPT. AI-related layoffs did happen—in some cases budget allocated to people went to GPUs, in others AI was the justification shareholders preferred—but they didn’t produce the mass displacement that was announced.

What to do. When managing expectations or fears on your team, bring real numbers, not CEO statements. The lesson Converse, don’t command shows you how to use AI to increase work quality rather than replace the people doing it.

In depth

The context

In 2023 and through early 2024, Altman and Amodei painted scenarios of rapid replacement: Altman spoke of professions disappearing “fast enough to potentially be scary,” Amodei of a large share of entry-level office jobs taken by AI “in a very short timeframe.” Those statements fueled both hype and anxiety, and gave many companies a specific angle for internal discussions. Now both have shifted position, moving toward more cautious framing: Altman says “that’s not what I expected,” and Amodei now describes AI as a productivity multiplier, not a substitute.

What the numbers say

Recent studies don’t confirm the original narrative. Multi-university research on programmers and copywriters traced the start of the employment downturn to early 2022, months before ChatGPT’s launch. The Yale Budget Lab found no measurable labor market shifts tied to AI. Layoffs declared “because of AI” did happen, but in some cases the savings went to GPUs, in others AI was the most convenient justification for cuts already decided.

No study so far shows significant AI impact on measured aggregate productivity or total employment. That doesn’t mean AI won’t change work—it already is, selectively, on specific tasks—but mass predictions haven’t held up.

The implications

If you’re in a management role, or simply need to explain AI to colleagues, the lesson is: CEO proclamations change, numbers less so. The narrative shift says little about facts and much about public pressure. When you need to decide how to use AI on your team, build your own measurement: time saved on concrete tasks, work quality before and after, real costs. The lesson Capstone: is it working? Measuring time and quality gives you the method.

What we don’t know

Altman cites no specific data, and “fairly confident” leaves room. The sources reporting this statement cite no underlying studies: it’s a public statement, not a measured result. That doesn’t make it false, but it makes it a position, not a verifiable conclusion.