The entry ramp is narrowing before the unemployment numbers move

Anthropic's Economic Index, updated in March 2026, reports two things worth reading together rather than separately. First: hiring of 22–25 year-olds into the occupations most exposed to AI has slowed by roughly 14% since AI tools became widely embedded in those roles. Second: there is no clear broad unemployment signal yet in those same occupations.

Read in isolation, each finding supports a different, comfortable story. The hiring slowdown supports "AI is coming for junior jobs." The absent unemployment signal supports "the AI jobs crisis is mostly hype for now," a framing several outlets ran with in early 2026. Both stories are too simple, and reading the two findings together explains why.

Employers can defer a hiring decision long before they take the more visible, more reportable step of reducing headcount. A slowdown in new graduate hiring shows up in recruiting data months or years before it would ever show up as a layoff number or a rise in the unemployment rate for that occupation. The absence of a headline signal is not evidence that nothing is happening; it is what the early stage of this particular shift looks like.

This matters for what kind of response is adequate. If the crisis were simply "jobs disappearing," income support and retraining would be the right tools. But narrower hiring at the entry level is really a narrower apprenticeship pipeline — fewer people getting the junior-role repetitions that turn novices into practitioners who can catch an expert-level mistake. That is a supervisory-capacity problem, not (yet) an unemployment problem, and it needs an answer that works even when the traditional pipeline for producing verified experts is thinner than it used to be. That is the gap the Cupel framework is built to address.