Dario Amodei, who runs Anthropic, has been issuing an unusual warning about his own product since 2025. Half of entry-level white-collar jobs could disappear within one to five years, he says. Asked whether his boss is right, Peter McCrory, who leads economic research there, drew a distinction instead of giving a number. He separated what AI is already doing to the economy from what it might do next.
His team can at least measure the first part. A March 2026 report by McCrory and Maxim Massenkoff tracked whether AI was showing up as job losses in the occupations where it is used most. Employment in those occupations had not fallen in a way that stands out from before AI. The one exception was a barely detectable slowdown in hiring the youngest workers into those roles.
What comes next is where that confidence runs out. In 2026, the most capable AI models could work through roughly fourteen hours of a task on their own. That figure had been doubling every four to seven months. Anthropic’s own researchers argue this could speed up how new ideas get discovered.
McCrory’s team tried to put a number on that possibility. A November 2025 study had Claude, Anthropic’s AI assistant, estimate how long tasks from real conversations would have taken a skilled person without it. The typical task took about ninety minutes and was finished roughly 80 percent faster with help. Over a decade, that would add 1.8 percentage points a year to productivity growth. That roughly matches the burst of growth seen in the late 1990s.
Not all of that use points the same way. Automated use, where the AI finishes a task alone, has grown fastest in jobs such as computer programming and data entry. The government’s ten-year projections show weaker job growth for them. Yet close to half of 81,000 Claude users said AI let them take on a wider range of work. Those seeing the biggest speed gains were also the most anxious about their jobs.
Asked in an interview to resolve that tension, McCrory points to history. Some past waves of automation mostly moved work around rather than wiping it out. Much of that reshuffling stayed hidden under stable-looking headline numbers. AI is moving through the economy faster than the tools before it, and the team paid to watch it closely cannot say which pattern it will follow.
