AI SkillsJune 1, 2026·5 min read

Both AI CEOs Just Walked Back Their Jobs Apocalypse. Here's What Actually Changed.

Sam Altman and Dario Amodei reversed their most alarming AI jobs predictions in the same week. The data underneath the reversals is more specific — and more actionable — than the headlines.

By Forge Team

The question is no longer whether AI will change your job. It already has. The more useful question, after a week in which both leading AI lab CEOs publicly reversed their most alarming predictions, is: which part of your job is becoming more valuable, and which part are you still doing because you haven't thought about it?

What the reversals actually said

At a Commonwealth Bank of Australia conference in Sydney on May 26, Sam Altman said he had been "pretty wrong" about AI job displacement, describing expecting far more entry-level white-collar elimination by now. He mentioned trying to delegate his own email and Slack to AI and reverting, concluding: "We really do care about our interactions with people."

The same day, Fortune reported that Dario Amodei has also shifted, now citing the Jevons Paradox: "If you automate 90% of the job, then everyone does the 10%, and the 10% expands to be 100%." Fortune noted that both reversals come ahead of trillion-dollar IPOs — worth factoring into how much interpretive weight you give them.

The complicating data sits right beside the walk-backs. Tech Journal reported 113,000+ tech layoffs in 2026 across 179 companies, with 48% explicitly citing AI as a driver. Meta cut 8,000 roles on May 20. Oracle eliminated between 10,000 and 30,000 positions. Job titles are not disappearing wholesale. Headcounts are shrinking.

Demis Hassabis, speaking at Stanford GSB (reported May 29), described current AI agents as "a little bit like a practice run" for what arrives in the next 12–18 months, and tightened his AGI timeline to possibly 2029.

What this means by Monday

The Amodei framing is the most actionable piece from the week. If AI automates 90% of a role, the remaining 10% becomes the entire job. The question is whether you know what your 10% is.

For knowledge workers, that 10% tends to be judgment, relationships, and problem-framing: deciding which customer problem actually matters, not generating the analysis of it. The conversation with the nervous client, not the summary of that conversation. Knowing when the AI output is wrong before anyone else in the room does.

If AI is making you faster at generating output, that is useful. If it is doing the thinking you used to do — and you're signing off without really reading it — that is the opposite of the direction these reversals point.

Evaluate three tasks you currently send to AI and identify which ones require your judgment at every step — and which ones don't.

An operations manager at a logistics firm

A senior operations manager at a 180-person logistics company uses AI to process carrier performance data, summarise weekly exception reports, and draft supplier communications. That work used to take 12 hours a week; it now takes four. The freed-up eight hours moved into client escalation calls and route redesign decisions — the conversations where relationships and domain knowledge determine whether a contract renews.

That is the Jevons Paradox playing out correctly. The 10% expanded. But it only expanded because she actively chose where to redirect her attention, rather than absorbing more data processing.

The risk is the reverse: using AI to speed up output generation without redirecting time to the judgment work underneath. That produces faster-looking work, not better decisions.

Practice the habits that keep your reasoning sharp when AI is handling more of the mechanics.

The counterpoint: where the 10% can erode

A content strategist at a 45-person SaaS company spent three years building specific judgment about what their technical audience actually reads, clicks, and shares. She now uses AI to write first drafts, generate outlines, and repurpose posts across formats. The output is faster than anything she produced manually. It is also, gradually, less specifically hers.

The risk is not that her role disappears — it is that her judgment about this audience, built over years of watching what lands, goes unexercised long enough that she can no longer trust it. Hassabis's phrase at Stanford — "maintain your sense of meaning" rather than becoming passive technology users — is not about resisting AI. It is about using it in ways that keep the 10% from quietly becoming 5%.

Altman's own reversion on delegated email maps to the same dynamic. The AI handled the mechanics. The relationship signal was lost.

The one thing worth doing this week

Write down the 10% of your role that AI cannot currently do — not "the human stuff" in the abstract, but the specific decisions and relationships where your judgment makes the outcome different. Then check how much of your actual working week you're spending on it.

That gap — between where your judgment matters and where your time goes — is the thing both CEOs were circling around, even if neither said it plainly.

Map one task you regularly hand to AI and identify exactly which decision point you need to stay inside.

Like this post?

Get the next one in your inbox. Practical AI skills, no filler.