---
title: "Is AI Eating Entry-Level Jobs? Stripe Checked"
url: https://nativefirst.ai/blog/wfh-not-ai-young-grads/
description: "Stripe's economist tested why young-graduate employment fell. Work from home explains 56% to 88% of it. AI exposure came out indistinguishable from zero."
author: John Tan
category: "AI Economics"
published: 2026-11-12
updated: 2026-11-12
publisher: nativefirst.ai
format: Markdown copy of the web page at the url above, same text.
---

# Is AI Eating Entry-Level Jobs? Stripe Checked.

Stripe's economist tested why young-graduate employment fell. Work from home explains 56% to 88% of it. AI exposure came out indistinguishable from zero.

John Tan · November 12, 2026 · 5 min read

The story everyone has settled on is that AI is eating entry-level work. It is tidy, it fits the headlines, and a number of companies are using it as the reason they are not hiring juniors this year.

Stripe's economist went and tested it. Ernie Tedeschi published the analysis today, and in his model the thing that best explains the collapse in young-graduate employment is not AI exposure at all. It is **working from home**.

## What happened to young graduates

**Young graduates · Sep 2026** Employment down, unemployment up, and the search a third longer than it was.

Charts by Stripe Economics from Current Population Survey data, 12 November 2026.

Take Americans aged 22 to 26 with exactly a bachelor's degree. Their employment-population ratio was 88% in September 2026, a point below three years earlier and nearly 1.5 points below where it sat before the pandemic. Unemployment in that group ran close to 6% at the end of 2025, against about 4% pre-pandemic. The typical unemployed young graduate has now been looking for **12 weeks**, about a third longer than a few years ago.

The detail that makes this a real phenomenon rather than a soft labour market: workers aged 30 to 49 have fully recovered to pre-pandemic levels. Whatever this is, it is happening to the people at the start.

**Who recovered · Sep 2026** Workers aged 30 to 49 are back above where they started. The 22 to 26 cohort is not.

Each series is indexed to its own December 2019 level. Stripe Economics, 12 November 2026.

## The result

Tedeschi compares 2023, when recent-graduate employment peaked, with the twelve months to September 2026, using young graduates against mid-career workers as the control. Then he asks which explains the gap: how teleworkable the job is, or how exposed it is to AI.

**By commuting zone · 2023–24** Where remote-work spending grew fastest, young-worker employment fell furthest. Correlation minus 0.353.

Stripe Economics, 12 November 2026. Spending is office supplies, furniture and electronics as a share of GDP.

**Fig. 1 Three explanations, one survives · Teleworkability tracks the decline. AI exposure does not.**

*Chart: Three results from Stripe's model. Occupational work-from-home share has a negative and statistically significant relationship with young graduate employment growth. Occupational AI exposure is negative but shallow and not statistically significant. Industry AI adoption is slightly positive and not significant.*

*Chart: The same three results, stacked.*

Figure by nativefirst, from the three panels in Stripe Economics, 12 November 2026. Shapes are illustrative of the reported direction and significance, not a redraw of the underlying scatter.

Every additional 10 percentage points of work-from-home adoption in an occupation is associated with young-graduate employment growing **3.9 to 6.6 points more slowly** than mid-career employment in the same occupation.

Run that as an accounting exercise and work from home accounts for **56% to 88%** of the 1.23-point fall in full-time-equivalent employment. Measured as simply having a job or not, it accounts for all of it.

Occupational AI exposure, over the same period and the same people, came out negative but much smaller, and statistically indistinguishable from zero. Industry AI adoption came out slightly *positive*, also not significant. Tedeschi's own summary is that telework adoption **"explains the decline in their employment better than AI exposure"** in the model.

## Why an empty office would hurt juniors specifically

The mechanism is the part worth holding onto, because it is the part you can act on.

Emanuel, Harrington and Pallais studied software engineers and found that sitting near colleagues increases the feedback a worker receives, and that the effect is strongest for the youngest ones. Lambert and Schindler found junior hiring shares fell as work-from-home exposure rose, after controlling for AI exposure.

Put plainly: juniors learn by proximity. They learn from the overheard call, the screen someone turns round, the correction that nobody would schedule a meeting to deliver. Remove the room and you have not just changed where the work happens. You have removed the apprenticeship, and with it much of the reason to hire someone who needs one.

## The part that should stop you getting comfortable

This is not a finding that AI is harmless, and the piece does not claim it is.

The central problem is stated outright: AI exposure and teleworkability are **"highly correlated with each other, making it hard to distinguish definitively"** between them. The jobs you can do from a kitchen table are, very largely, the jobs a model can do. Two explanations that move together are extremely hard to separate, and the model is **"suggestive rather than truly causal"**.

The literature is genuinely split rather than settled. Brynjolfsson and colleagues found slower employment growth for young workers in AI-exposed occupations even after controlling for work from home. Orr, Tucker and Warren, using linked Census records, found that graduates from the most AI-exposed tenth of majors saw initial employment fall about 5 percentage points and initial earnings about 13% against less-exposed majors. Against that, Fairlie and Wu found no unusual rise in graduate unemployment, a Finnish study found no systematic displacement through September 2025, and an IMF paper found no meaningful junior-versus-senior difference in AI's effect on job openings. Frank and colleagues found the deterioration started before ChatGPT existed.

And the line that should sit uncomfortably with anyone declaring this settled: the full extent of AI's effects **"might not manifest themselves until the next recession"**. Firms rarely fire people because a tool arrived. They decline to rehire when demand falls, and discover the tool covers it. That is how the internet took travel agents, and it did not look like a disruption while it was happening.

## What to do with this

**Stop using AI as the reason you are not hiring juniors.** It is the least supported of the available explanations in the best analysis published so far, and it is the only one that lets you off the hook. It is also not the first time the data has pointed the other way: across 21,559 companies, [the heaviest AI adopters hired more people, not fewer](https://nativefirst.ai/blog/heaviest-ai-adopters-hired-most/). If your graduate pipeline has thinned, the evidence points first at a change you made to where people sit.

**If you are remote and you want juniors to work out, you have to manufacture the feedback.** The thing proximity was quietly providing was correction at high frequency and low ceremony. Remote does not supply it by accident, and a weekly one-to-one is not a substitute for twenty small corrections a week. Someone has to own that deliberately or the junior does not develop, and then the decision not to hire the next one looks justified.

**Watch the next downturn, not this quarter.** If AI is going to show up in the entry-level numbers, the evidence says it shows up as a failure to rehire rather than a wave of redundancies. The measurement that matters is not whether you cut, it is whether the roles come back.

The honest summary is not that AI is innocent. It is that the two suspects were in the same room, one of them is a decision you made, and you can test that one yourself.

**Check the room first.**

[John Tan ↗](https://www.linkedin.com/in/johntans/)

Founder of nativefirst.ai. Runs everything on agents and writes down what works.

Sources

1 Ernie Tedeschi, "AI, work from home, and the young-graduate job market", Stripe Economics, 12 November 2026: young graduates aged 22 to 26 had an employment-population ratio of 88% in September 2026; each additional 10 percentage points of work-from-home adoption is associated with young-graduate employment growing 3.9 to 6.6 points more slowly than mid-career; work from home accounts for 56% to 88% of the 1.23-point fall in full-time-equivalent employment, and AI exposure is statistically indistinguishable from zero. The author states the model is suggestive rather than causal. [stripeeconomics.com](https://www.stripeeconomics.com/p/ai-work-from-home-and-the-young-graduate)
