Jobs

There is no real disagreement about the destination. Ilya Sutskever, who helped build the engine, will tell a room of students flatly that "the day will come when AI will do all of our jobs... not just some of them, but all of them" (Singularity Watch, Jan 2026), and his reasoning is one sentence long: the brain is a biological computer, so a digital one can do the same things. Dario Amodei thinks "a country of geniuses in a data center" is one or two years out (AI Nutshell, Feb 2026). Shane Legg has held a 50-50 bet on minimal AGI by 2028 since 2009 (IA XYZ, Dec 2025). The frontier leaders are not arguing about whether. They are arguing about speed, sequence, and what's left when the dust settles ... and on those questions the consensus shatters.

The order of disappearance is now a technical claim, not a hunch

Every one of these people has a theory of which jobs go first, and they have converged on a brutal one: the safe jobs are the physical ones, and only because the robots aren't ready.

Shane Legg gives it a name. The "laptop rule": if your whole job runs through a screen and a keyboard, you are in the exact domain AGI hits first (IA XYZ, Dec 2025). Software, finance, law are "right on the front lines"; plumbing and hands-on care are protected, "simply because robotics is moving at a much slower pace." His number is specific and grim: a project needing 100 engineers today "could potentially only need 20 who are mostly just supervising."

Amodei agrees and then removes the comfort. Asked whether physical reality is fundamentally different, he says no: "I don't think so." Claude was used to pilot the Mars rover; piloting a robot is "not different in kind. It's different in complexity." The blue-collar reprieve, in his telling, is a hardware delay, not a safe harbor. "The brain of the robot will be made in the next couple years. The question is making the robot body."

The genuine surprise in the evidence is Amodei reversing himself on the record. "Six months ago I would have said the first thing to be disrupted is entry-level white collar jobs... I still think those are going pretty fast. But I actually think software might go even faster" (AI Nutshell, Feb 2026). Not because the models are inherently better at code, but because developers "adopt things quickly" and are "socially adjacent to the AI world." Diffusion, not capability, sets the clock. Customer service, banking, manufacturing are further from the frontier socially, so they get a few more months. That's the whole reprieve.

The fault line: bloodbath versus abundance

Here the two big labs split hard, and the split is temperamental as much as analytic.

Amodei's frame is disruption faster than society can absorb. He owns the word the market threw at him ... "the bloodbath headlines, oh my god are the entry-level pipelines going to dry up" ... and then explains why it isn't hysteria. The historical comfort ("we were farmers, then industry, then knowledge work, people adapted") fails because that took "centuries or decades. This is happening over low single-digit numbers of years." His actual fear, stated plainly: "the normal adaptive mechanisms will be overwhelmed." That single phrase, he notes, is what spooked the market the week he said it.

Sam Altman, same month, different planet. At IIT Delhi he tells students "you all will enter adulthood with superintelligence" as an "incredible opportunity," and frames the whole thing as scarcity giving way to abundance: "We built up instincts and institutions to deal with a world of scarcity. Almost none of that applies well to a world of abundance" (IIT Delhi, Feb 2026). Where Amodei sees a feedback loop with no natural break, Altman sees deflation as the good news: "GDP is going to turn out to be a terrible metric because AI is so deflationary." He concedes the disruption is real ... Codex, he admits, blindsided even OpenAI: "what it means to be an engineer changed so much so quickly" ... but the structure of his answer is opportunity, not overwhelm.

Greg Brockman, six months earlier, is the bluntest of the OpenAI camp and the least reassuring: asked "is my job in danger," he says "AI is going to change a lot of jobs," and then the real line ... "we are going to change a lot of fundamentals of the social contract" (Matthew Berman, Oct 2025). No abundance gloss. Just: the contract is getting rewritten, and he's not promising you like the new terms.

Anthropic's own house is divided

The most interesting tension isn't between labs. It's between the Amodeis.

While Dario is narrating a bloodbath, his sister and co-founder Daniela Amodei is running the optimism campaign. The number of jobs AI can do without humans, she says, is "vanishingly small" (Fortune, Feb 2026). Her bet is augmentation: "humans plus AI together actually create more meaningful work, more challenging work, high-productivity jobs." And her advice cuts against the entire STEM-forever narrative: "studying the humanities is going to be more important than ever," because the models are already good at STEM, and what's scarce is "understanding what makes us tick" and "the ability to interact with other people." Anthropic even ran a campaign, "keep thinking," built on the premise that you shouldn't outsource your higher-level thinking to Claude.

Read side by side, it's not a contradiction so much as a division of labor: Dario describes the macroeconomic wave, Daniela describes the individual surviving it. But notice the seam. Dario says the centaur phase ... human-plus-AI, demand for engineers possibly rising ... "may be very brief," then "it's just the machine," invoking how the human-checks-the-AI era in chess ended. Daniela's "humans plus AI" is exactly that centaur, sold as durable. One of them is wrong about how long the partnership lasts.

Jack Clark and the politics of jobs nobody needs

Jack Clark, Anthropic's third voice, skips the economics and goes straight to the part everyone else flinches from: what happens politically when the work runs out.

His prediction (Conversations with Tyler, Feb 2026) is the sharpest in the whole file. There is "a high chance for a political movement to arrive which tries to freeze a load of human jobs in bureaucratic amber" ... not by reason, but "driven by the chaotic winds of political forces." When Cowen offers the cozy version (everyone keeps a job, AI does the hard parts, we're richer so we can afford it ... a welfare state for service workers), Clark refuses the bargain on grounds of meaning: "I'm not confident that you'll pick a load of jobs which naturally create their own meaning."

He's also funny and unnervingly concrete about the jobs that survive: the number-one growth role might be "laundering the information that comes from AIs into human systems that are not predisposed to that information going in directly." He does it himself ... consults Claude when his toddler bonks their head, then dials the human advice nurse, because he "can't take that Claude assessment and give it to Kaiser Permanente." The future of work, in Clark's telling, is partly humans acting as legal and emotional adapters for machines that already know the answer.

What actually shifted in 2025-2026

The honest reading of the evidence: 2026 is when the abstraction became a stock chart. Amodei's interviews are bracketed by real market carnage ... the AI Nutshell framing has Anthropic announcements knocking eight-figure-percentages off CrowdStrike and IBM, and in India, where "$200 billion of exports is IT services," Demis Hassabis concedes "a lot of areas are going to get disrupted" (India Today, Feb 2026) while Amodei spends a CNBC-TV18 hit insisting Anthropic is "not trying to replace the existing IT industry," pitching Claude as "Lego building blocks." That's the tell: the leaders are now doing diplomacy, because the disruption stopped being a slide and started being a layoff.

Shane Legg's analogy is the one that lingers. We are in "March 2020, when all the experts were shouting about an exponential curve, but most people were just going about their daily lives." The builders all see the curve. They cannot agree on whether it ends in abundance or amber.

So watch the verbs. Altman says opportunity. Amodei says overwhelmed. Brockman says rewritten. They're describing the same automation ... and quietly betting their reputations on three incompatible endings.