AI is set to dominate the 2028 election, and the forecasts behind that claim are more specific than the politics suggests

Dario Amodei, Anthropic’s chief executive, expects AI to be “maybe the most important issue in the presidential election in 2028.” That is a short runway. It is also a more specific claim than the ambient predictions about AI’s political importance that have circulated for years, and it deserves to be taken on its own terms.

The timeline Amodei sketches does not stop at 2028. He places a cash dividend to all Americans in 2033, and in 2037 what he describes as “the apocalyptic arrival of truth on Earth.” Those later entries carry a speculative freight that the 2028 prediction does not, but the sequence is worth noting: Amodei is not describing a slow drift toward political salience. He is describing a compressed series of discrete ruptures.

Ryan Greenblatt, a researcher at Anthropic, fills in the technical scaffolding beneath those political predictions. He expects full automation of AI research and development somewhere around 2030 or 2031. After that point, his median expectation is that four or five years of AI progress will collapse into a single year. His median for AI systems that beat all humans on any job sits at roughly 2033. He assigns a 35 to 40 percent probability to AI takeover by 2040 and states plainly that, by his default timeline, misalignment concerns become “really, really crazy and concerning” about three years from now. These are not fringe estimates from an outsider. They come from someone working inside one of the two firms that, as Brad Gerstner observes, have pulled decisively ahead of the field in the past year.

Maybe my median expectation is something like four or five years of AI progress in a single year. Ryan Greenblatt

The labor dimension is where the political consequences become most legible. Jason Calacanis relays a forecast he attributes to Amodei: 50 percent of entry-level knowledge worker jobs disappear within one to five years. Tobi Lütke, Shopify’s chief executive, offers a historical counterpoint worth holding alongside that number: eight of today’s top ten best-paid jobs did not exist 20 years ago. The juxtaposition is not reassuring so much as clarifying. Displacement and creation have coexisted before, but the speed Greenblatt and others describe compresses the adjustment period that historical transitions allowed. Sarah Guo, co-founder of Conviction, goes further: coding will be a fully solved problem by the end of this year, she argues, with a light form of recursive self-improvement arriving by end of next year.

Public sentiment is already shifting to reflect that anxiety. Nathan Labenz notes that, for the first time, a majority of adults under 30 say they are more concerned than excited about AI, putting them on par with those in their 30s and 40s and those 65 and up. The only cohort still majority-excited is the 50-to-64 age group. Melisa Tokmak, observing hiring conversations, reports seeing something consistent with that data from the inside: a “permanent underclass mentality” among younger candidates who fear that if they do not secure financial stability within 18 months, AI will subsume their capabilities entirely. Whether that fear is calibrated correctly is beside the point for the political question. Widespread economic anxiety, grounded or not, produces political pressure.

The geopolitical layer adds further instability. Harry Stebbings reports that the Chinese administration is discussing preventing Chinese companies from selling open-weight models to the United States, a mirror of existing US restrictions on Nvidia chip exports to China. Stebbings also notes that a Chinese open-source model he refers to as Kimmy K3 recently beat the best closed-source American models on an important subset of tasks, calling into question how much runway the US retains at the frontier. Eric Vishria adds that China is bringing on 10 times as much energy next year as the United States, which matters because energy is the primary constraint on compute at scale.

Bernt Bornich states his position without qualification: hard takeoff, defined as robots autonomously building robots, data centers, and chip fabrication plants, is less than a decade away. That claim sits at the outer edge of what the other evidence supports, but it is not disconnected from it. Greenblatt’s timeline, Amodei’s political predictions, and Guo’s near-term coding milestone all point in the same direction. The disagreements among these speakers are about pace and probability, not about whether transformation is coming. Institutions built on the assumption that this kind of change arrives gradually, and can be addressed through ordinary policy cycles, are operating on a premise the evidence does not support.