Dario Amodei Flipped His Jobs Prediction. That Tells You Something About the Window.

The anxiety about AI and jobs has been running so hot for so long that it's easy to miss when the signal actually changes. It changed on May 5. Dario Amodei — the CEO of Anthropic, the company building Claude — publicly walked back his previous position that AI could eliminate 50% of entry-level knowledge work. What he said instead matters more than the reversal: it depends on whether markets have "time to adjust." The window is real. It's finite. And it's still open — but the people who will come through it well are not the ones still arguing about whether it exists.

What happened

Amodei had spent most of 2025 warning that AI could displace half of entry-level knowledge work. This was not a fringe take — it was the CEO of a leading AI lab giving a specific number to a scenario most people preferred not to think about.

On May 5, at a New York event alongside JPMorgan CEO Jamie Dimon, Amodei invoked a different framework: the Jevons Paradox. The economic idea — that greater efficiency in using a resource tends to increase total consumption of that resource, not decrease it — has been used to argue that AI won't destroy jobs because it will create demand for human work in adjacent areas. Amodei applied it directly, while adding a constraint the optimists tend to skip: the mechanism only works if markets have time to adjust.

The same week, The Neuron published analysis on what becomes scarce when AI handles generation cheaply (The Neuron, May 4). The answer: judgment calls, relationships, and taste — the things that can't be automated because they require knowing what good looks like in a specific context, for a specific person, with specific stakes.

What to do differently Monday morning

Stop waiting for the jobs question to resolve before deciding whether to build AI skills. It won't resolve cleanly, and the wait is itself a decision.

The Jevons argument says: your workload may grow, not shrink, because AI makes you and your team faster. But faster only creates more value if you can direct, evaluate, and decide on what the AI produces. Speed without judgment is just more output that still needs checking.

The skill that compounds right now isn't prompting — it's task framing. Knowing which problems are worth giving to AI, and how to hand them off in a way that gets useful output back, is what separates professionals who get leverage from AI from professionals who get drafts they spend as long editing as they would have writing.

Judgment on AI output is becoming a premium skill. When generation is cheap, the person who can tell good from acceptable from wrong — quickly, reliably, in context — has something that scales. That's not a new skill; it's the editing, critical thinking, and domain knowledge you already have, applied earlier in the process.

A head of people at a 200-person company

Priya runs HR at a 200-person technology services firm. For most of 2025 she watched colleagues debate whether AI would replace recruiters, HR coordinators, and L&D roles. She didn't resolve that question. She decided it wasn't the right question for the next six months.

Instead she spent those six months learning which parts of her own work AI could handle adequately and which parts it couldn't. AI drafts job descriptions, generates first-cut interview questions, and summarises candidate notes. Priya reviews, edits, and makes the final calls — on candidates, on language, on which role actually fits the team's gap.

When her company ran a hiring push in Q1 and doubled the volume of open roles, her team handled it with the same headcount. The colleagues who spent that time waiting for the jobs question to resolve are now applying for the role she grew into.

When the wrong question wastes the window

James manages vendor relationships at a 40-person logistics company. He spent much of the past year in a defensive crouch — waiting to see what AI would do to his role before deciding whether to engage with it.

His concern wasn't irrational. Some of what he does — tracking invoice status, chasing confirmations, summarising supplier calls — is exactly the kind of work that AI handles adequately. He was right to notice that.

What he missed was the question Amodei's framing implies: if the efficiency gains from AI create more demand, where does that demand land? For James, it landed on the analysis and negotiation work he'd been doing less of because the administrative work consumed most of his time. AI handles the administrative layer now. He's doing more of the work that actually required him.

The jobs question turned out to be the wrong frame. The right question was: if AI takes the parts of my job that don't require me, what do I do with the time?

The actual window

Amodei's Jevons Paradox invocation is not a guarantee. It's a conditional: if markets adjust in time. The adjustment is individual before it's organisational. Companies will not build AI fluency for you. The window that's open now — where building these skills makes you measurably more effective before the tools are table stakes — is what makes the conditional worth acting on.

The jobs question will keep being asked. Answer it by closing it: pick one thing you do regularly that AI could handle adequately, hand it off clearly, and use the time for the work that requires you.