Entry-Level Hiring Didn't Slow Down at AI-Adopting Companies -- It Fell 80%

If the job search has felt disproportionately brutal since you graduated, there's a number behind that feeling now, and it's a startling one.

A Harvard working paper published this year found that at companies that have adopted generative AI, entry-level hiring has dropped roughly 80% per quarter since 2023 -- not 8%, eighty. Senior hiring at those same companies kept growing over the same period. Researchers gave the pattern a name, seniority-biased technological change, which is a dry way of describing something that, lived through, feels like the ladder got pulled up right as you reached for the first rung. Recent graduate unemployment has climbed to 5.7%, now higher than unemployment for the overall workforce, for the first time in decades.

The mechanism isn't that AI targets "easy" jobs broadly. Generative models are unusually good at codified, checkable tasks -- the kind learned from a training set, done the same way a thousand times before and written down somewhere. That description is close to a job spec for entry-level work, where the whole point has traditionally been doing the structured version of a task while learning the judgment calls on the job. The part that should unsettle more than the headline number: this pullback happened before AI had actually replaced those workers in most functions. Companies cut entry-level hiring in anticipation, almost immediately after ChatGPT's public release, on the expectation automation was coming -- whether or not it had fully arrived yet.

This isn't "no jobs for anyone." Overall postings are flat to declining, but that average hides a narrower, pickier market: employers are still hiring, just concentrating it around roles and skills explicitly tied to AI. That's a structural shift, not a cyclical dip that ends when the economy turns -- it ends, if it ends, when the tasks that define entry-level work get redefined, which rewards whoever moves first.

The pitch that stops working is "I can do the structured, checkable version of this job" -- that's a losing argument against a tool that does the same version faster and never asks for a raise. The pitch that still works, and that companies keep hiring for even while cutting the junior layer: catching the exceptions a process misses, turning a messy ask into the right question, being trusted with a relationship, owning an outcome instead of a task.

Harvard working paper on generative AI and entry-level hiring, as reported by Forbes (May 2026); Indeed Hiring Lab's January 2026 US Labor Market Update; recent-graduate unemployment figures cited in EPIC's March 2026 jobs report.

Also worth reading: AI vulnerability score by job title · how we score AI resilience