Economics

There is a strange consensus among the people building frontier AI: the economy is about to be rebuilt, and none of them quite know what it will look like when they are done. Ask Sam Altman and you get deflationary abundance. Ask Dario Amodei and you get a "bloodbath" of entry-level white-collar jobs arriving faster than society can absorb. Ask Ilya Sutskever and you get a shrug at the fact that the models ace the hardest exams and barely move GDP. They are not arguing about whether AI matters. They are arguing about the shape, speed, and distribution of a transformation they all consider inevitable. That is the real story here, and it sharpened considerably over the winter of 2025-26.

Everyone agrees on the size. Nobody agrees on the speed

Start with the agreement, because it is total. Demis Hassabis calls the coming decade "something like 10 times the impact of the industrial revolution but happening at 10 times the speed... unfolding in a matter of a decade rather than a century" (Indian AI summit, Feb 2026). Shane Legg, his co-founder, says superintelligence will "structurally change the economy and society and all kinds of things" and pegs minimal AGI at "probably about two or so" years (Google DeepMind podcast with Hannah Fry, Dec 2025). Altman tells a room at IIT Delhi that "we're going to automate scientific progress. We're going to automate the whole economy" (Feb 2026). Dario Amodei puts 90% odds on a "country of geniuses in a data center" within ten years and "a hunch... within a year or two."

So the magnitude is settled. The fight is about diffusion, and this is where the most interesting hedge in the whole dataset lives. Amodei: "I'm very bullish about the direction of the AI itself... But the diffusion to the economy is going to be a little slower, and that diffusion creates some unpredictability." His evidence is his own product. Code adoption went fast, he argues, not because models are inherently better at code but because developers "are used to fast technological change and they adopt things quickly... If you do customer service or banking or manufacturing the distance is a little greater."

Amodei's quiet retreat from "bloodbath"

The most consequential shift in 2026 is Dario Amodei's. He owns the original doom framing... he is the one who put "bloodbath headlines" and "entry-level white collar labor" into the discourse. But listen to him in February 2026 and you hear a man complicating his own position. "Six months ago I would have said the first thing to be disrupted is these entry-level white collar jobs... data entry, document review for law. I still think those are going pretty fast. But I actually think software might go even faster."

The reframe is from displacement toward speed as the enemy. His worry is no longer simply that jobs vanish. It is that adaptation cannot keep pace: "People talk about previous disruptions... we used to be farmers, then industry, then knowledge work. That happened over centuries or decades. This is happening over low single-digit numbers of years." His verdict, delivered flat: "the normal adaptive mechanisms will be overwhelmed." And he is careful, repeatedly, to add: "I'm not a doomer."

His centaur analogy does the heavy lifting. Borrowing from post-Deep-Blue chess, where human-plus-machine beat any human or machine alone for fifteen or twenty years... before that era simply ended and "then it's just the machine." Amodei thinks software is in its centaur phase now, demand for engineers may even rise, "but the period may be very brief." Same for law: the apprentice tier (paralegals, junior associates) hollows out, leaving the senior partners who talk to juries. Adaptation is possible "if you froze the quality of the technology in place." Nobody is freezing anything.

His sister and co-founder Daniela Amodei holds the optimistic flank from inside the same building. Jobs AI can do with no human at all are "vanishingly small," she told ABC (via Fortune, Feb 2026), and "humans plus AI together actually create more meaningful work." Her counsel: study the humanities, because critical thinking and dealing with other people get more valuable, not less. But she does not pretend the gains are evenly shared. On Anthropic's Economic Index: "Unfortunately the folks that have the most are going to be the first to adopt it. They have the most resources, the most time, the most access." Rich-first, faster than dial-up or Google ever diffused, with whole regions at risk of being "left behind." The optimism is real; so is the asterisk.

Altman's abundance, and the metric that breaks

Where Amodei sees an adaptation crisis, Sam Altman sees a measurement problem. His IIT Delhi framing is pure abundance: society built "instincts and institutions, policies, structure to deal with a world of scarcity. But almost none of that applies well to a world of abundance." His sharpest claim is technical and easy to miss: "GDP is going to turn out to be a terrible metric because AI is so deflationary." A "proper metric" of quality of life will more than double, he says, but "GDP measured in nominal dollars... gets kind of squirrelly." By 2035, "a hugely deflationary economy that the world isn't thinking about."

This is the genuine fault line. Amodei's centaur and Altman's deflation are not the same story dressed differently. One says the displacement comes too fast for the social fabric; the other says the abundance comes so fast our accounting can't see it and money becomes "a whole different thing." One is a warning about people. The other is a thesis about prices.

Greg Brockman, Altman's co-founder, is the bluntest and the least varnished: "AI is going to change a lot of jobs... I think we are going to change a lot of fundamentals of the social contract" (Matthew Berman, Oct 2025). But Brockman's real obsession is upstream of labor entirely... compute. "We are heading to a world of absolute compute scarcity... energy, certainly in the US, is going to be a massive bottleneck." For OpenAI the binding constraint on the AI economy is not jobs. It is gigawatts.

The skeptic, the verifier, and the bureaucratic amber

Two voices refuse the whole frame. Ilya Sutskever, in his first deep interview since leaving OpenAI (Dwarkesh, Dec 2025), names the elephant: "The models seem smarter than their economic impact would imply." They crush hard evals, then reintroduce a bug they just fixed, then reintroduce the first one. His theory: RL training over-optimizes for the evals themselves... "the real reward hacking is human researchers who are too focused on the evals." If he is right, the entire timeline debate is built on benchmarks that don't generalize to the economy.

Jack Clark, Anthropic's policy head, supplies the most original economic idea in the set. Summarizing the "Some Simple Economics of AGI" paper (Import AI, Mar 2026), he frames the AGI economy as two racing cost curves... an "exponentially decaying Cost to Automate and a biologically bottlenecked Cost to Verify." The scarce resource stops being intelligence and becomes human verification bandwidth. The failure mode is a "Hollow Economy" of "high nominal output but collapsing realized utility," where agents "generate counterfeit utility." (Clark, being Clark, also flags that the paper's theory sections read like "theory slop.") And on Conversations with Tyler (Feb 2026), he predicts the political endgame with a phrase that should outlive him: a movement to "freeze a load of human jobs in bureaucratic amber," not done "in a reasoned way" but "driven by the chaotic winds of political forces."

The honest summary: they all expect a phase change, and they are improvising the economics in public, in real time, with billions in market cap moving on a single Anthropic blog post. The most candid line belongs to Sutskever, watching 1% of GDP pour into data centers: "how normal the slow takeoff feels."