AI and Entry-Level Jobs in 2026: What the Research Shows, and Why Licensed Work Holds Up
What Stanford and New York Fed research says about AI and early-career employment, where the evidence disagrees, and why licensed, accountable work like IRS representation by Enrolled Agents is harder to automate away.
Last reviewed: November 10, 2026. This topic is contested; both sides are linked so you can judge the evidence yourself.
The worry is reasonable: if AI can write a first draft, what happens to the people whose first job was writing first drafts? The research is still young and it disagrees with itself. Here is what the strongest studies say, where they conflict, and what tends to stay human.
The evidence that AI is hitting early careers
Stanford's Digital Economy Lab used payroll data to track workers by age and AI exposure. In its August 2026 revision, employment for 22- to 25-year-olds in the most AI-exposed jobs was about 19% below what the trend predicted, while experienced workers in the same jobs showed no such gap (Stanford Digital Economy Lab, "Canaries in the Coal Mine"). That is a gap against a counterfactual, not a measured 19% drop in jobs.
The evidence that it's not (mostly) AI yet
- New York Fed researchers found that job postings in AI-exposed occupations started falling before ChatGPT, and junior and senior postings moved together (Liberty Street Economics, May 2026).
- A separate New York Fed analysis estimates remote work explains about 64% of the rise in young-graduate unemployment, a rise that also began before generative AI (Liberty Street Economics, June 2026).
- Recent-graduate unemployment was 5.2% in Q3 2026 (New York Fed).
Honest summary: something is happening to early-career white-collar hiring, and AI is a plausible part of it, but the size of AI's role is not settled.
Why licensed work is harder to automate away
| Feature | Why it resists automation |
|---|---|
| Legal accountability | Someone has to sign, be responsible and answer to a regulator |
| Representation | An Enrolled Agent can represent a taxpayer before the IRS; a model can't hold that license |
| Judgment on messy facts | Real clients bring incomplete records and changing stories |
| Trust | Clients and the IRS rely on a named, credentialed person |
The IRS grants EA status to people who pass its Special Enrollment Examination and a suitability check, and lists no education requirement (IRS: Enrolled Agent information). AI tools will make tax professionals faster. They don't remove the need for the licensed person who is accountable for the work.
What to do with this
- Aim for accountable roles. Prefer work where a license, a signature or a client relationship is part of the job.
- Use AI as a tool, not a plan. Learn the rules well enough to check what a tool tells you.
- Build proof. A passed licensing exam is evidence of what you know. Start with the free EA diagnostic.