AI is becoming a political force faster than political systems are built to handle it

Dario Amodei’s timeline is worth sitting with before moving past it. The Anthropic CEO predicts AI will be the most important issue in the 2028 US presidential election. He sketches further milestones beyond that: a universal cash dividend to all Americans by 2033, and what he describes as “the apocalyptic arrival of truth on Earth” by 2037, driven by the invention of something like perfect lie detectors. These are not vague gestures at disruption. They are specific enough to be tested, and specific enough to be wrong. What makes them worth taking seriously is not Amodei’s confidence but the degree to which present-day operational and geopolitical evidence is already moving in the direction he describes.

Fred Turner, who runs the healthcare company Curative, offers a ground-level view of what the political pressure will actually be rooted in. Turner notes that current-generation models can already handle every back-office task at his company, with deployment speed the only remaining constraint. That framing matters: the bottleneck is no longer capability, it is rollout. What was a research question has become a logistics question, and logistics questions tend to resolve faster than the institutions they disrupt expect.

The economic dimension of that disruption is sharpening. Turner projects that within three years, companies will spend somewhere between two and five times an engineer’s salary on AI API costs, compared to roughly 3.8 percent of salary today. That trajectory, if it holds, rewrites the cost structure of software development faster than most workforce planning cycles can accommodate. The political pressure that generates, across engineering labor markets and the tax base that depends on them, is the kind of pressure that ends up on debate stages.

We think that it'll be maybe the most important issue in the presidential election in 2028. Dario Amodei

Turner also describes the organizational model that is beginning to emerge at Curative. Rather than keeping contract volume flat and simply replacing human negotiators with an AI agent, the company is using the agent to do ten times as many contracts per year as it could the year before. One senior engineer now manages multiple downstream agents that implement features, with engineers reviewing outputs rather than writing code. That is not a productivity story in the conventional sense. It is a story about what the unit of professional work becomes when the agent layer is capable enough to absorb the volume. The political implications, for how labor markets are organized, taxed, and protected, follow directly from that structural shift.

The geopolitical dimension is arriving on a parallel track. Harry Stebbings reports that the Chinese administration is in discussions about preventing Chinese companies from selling open-weight models to the US, mirroring existing US restrictions on Nvidia chip exports to China. If that account is accurate, AI model access is already generating the kind of state-level friction typically reserved for the most consequential strategic technologies. Semiconductors took decades to become a site of explicit export control. The trajectory here appears to be compressing that timeline considerably.

None of this settles whether Amodei’s specific milestones arrive on schedule. A cash dividend to all Americans by 2033 is a political outcome, not a technical one, and political outcomes depend on coalitions and contingencies that no single trajectory can capture. But the underlying pressure that would produce that kind of proposal, the decoupling of output from headcount, the compression of back-office labor, the restructuring of software engineering into a supervisory function, is not hypothetical. It is already operating at companies like Curative. The question Amodei is really asking is whether political systems will respond to that pressure through redistribution, or whether they will simply absorb the disruption in ways that make redistribution harder to avoid. By 2028, that question will be unavoidable.