A quiet but consequential shift in the entry-level labour market has drawn renewed attention from economists. Employment among Americans aged 22 to 26 with a bachelor's degree stood at just 88% in September 2026, roughly 1.5 points below pre-pandemic levels and still not recovered. Unemployment for this cohort approached 6% at the end of 2025. The typical unemployed young graduate had searched for 12 weeks, about a third longer than a few years earlier. The question animating a new analysis from Stripe Economics is pointed: is artificial intelligence or the persistence of remote work responsible? In a carefully constructed examination1 published November 12, Stripe Economics' Ernie Tedeschi argues the answer is more complicated than the prevailing narrative suggests.
The Case Against Blaming AI Alone
Tedeschi is direct on the limits of AI-first explanations. "A narrative that blames AI entirely for this slowdown has complications," he writes. Among them: non-graduate youth employment has also fallen since 2023, and has dropped slightly more than graduate employment since December 2019. Mid-career workers aged 30 to 49 have fully recovered. If AI displacement were the dominant mechanism, one would expect the burden to fall disproportionately on credentialed workers in high-exposure occupations, not across the youth labour market broadly.
The complicating factors extend further. Inflation, rising interest rates, reduced job-switching after the Great Resignation, and immigration swings all coincide with AI's commercial rise since late 2022. Attribution, Tedeschi notes, is genuinely difficult.
Work from Home as the Stronger Signal
The centrepiece of the Stripe Economics analysis is a Current Population Survey model comparing early 2023 with the 12 months ending September 2026. The methodology is rigorous: a treatment cohort of non-disabled, non-enrolled 22 to 26-year-olds with exactly a bachelor's degree, measured against mid-career workers aged 30 to 49. The findings are substantial.
Each additional 10 percentage points of WFH adoption in an occupation was associated with young-graduate employment growing 3.9 to 6.6 points more slowly than mid-career employment. Applied to the actual 1.23 percentage-point decline in full-time-equivalent employment, WFH explained 56% to 88% of that gap. For the simpler employment-population ratio, WFH explained 97% to 153% of the decline. AI exposure, by contrast, registered a negative but statistically insignificant association. When both measures were included in the same model, WFH remained significant while AI did not.
The regional data from Stripe's proprietary spend adds a corroborating dimension. Commuting zones where spending on office supplies, furniture, and electronics grew faster relative to GDP saw larger declines in employment for workers aged 22 to 29 from 2023 to 2024. The correlation is moderate but directionally consistent with the CPS findings.
Tedeschi is candid about the limits of the exercise. "Our analysis suggests that WFH is an important factor moving the needle on young-graduate employment," he writes. "Telework adoption explains the decline in their employment better than AI exposure in our model using public data." Yet he is equally clear that "our model is suggestive rather than truly causal." AI and teleworkability are highly correlated across occupations, and other potential drivers were not tested.
A Mechanism Rooted in Learning and Proximity
The existing research literature supports a plausible mechanism. Emanuel, Harrington, and Pallais (2026) find that office proximity improves job feedback for junior software engineers, and that relative unemployment has risen for young graduates in remotable occupations. Lambert and Schindler (2026) document declining junior hiring shares with WFH exposure, even after controlling for AI. The logic is structural: early-career workers depend on in-person observation, informal mentorship, and agglomeration effects that remote work erodes. Senior workers, whose skills and networks are established, face no comparable penalty.
As Tedeschi notes, "WFH is a mixed bag when it comes to the labour market." For disabled workers and those aged 50 to 64, the evidence from Bloom, Dahl, and Rooth (2026) and Burn, Gentry, and Lahey (2026) points to meaningful employment gains. The same arrangement that expands opportunity for one cohort may systematically disadvantage another.
AI Risk Remains an Open Question
Studies by Tucker (2026) and Orr, Tucker, and Warren (2026) find sharp hiring declines among the most AI-exposed young workers and graduates in the most AI-exposed majors. Frank et al. (2026) identify deterioration in entry-level hiring in AI-exposed jobs beginning before ChatGPT's release. Tedeschi acknowledges these signals and urges that "researchers should keep an open mind at this early stage in AI's deployment." He flags the possibility that AI effects could emerge more sharply in a future recession, when firms under margin pressure may move more aggressively to substitute.
5 Key Takeaways for Advisors and Investors
- Labour market softness among young graduates is structural, not cyclical. Employment gaps for this cohort persist 1.5 points below pre-pandemic levels while mid-career workers have fully recovered, suggesting this is not a temporary overhang.
- Remote work adoption carries a measurable cost to entry-level hiring. The analysis links higher WFH penetration in an occupation directly to weaker young-graduate employment relative to mid-career peers, with WFH explaining the majority of the decline.
- AI displacement, while plausible, remains statistically unproven at the aggregate level. Current public data do not support a significant AI-driven employment effect on young graduates, though individual studies point to concentrated impacts in highly exposed majors and industries.
- The two forces are difficult to disentangle. AI exposure and teleworkability are highly correlated across occupations, complicating causal inference and warranting caution when interpreting labour market narratives built on either factor alone.
- Recession risk may be the stress test for AI's true labour market impact. Tedeschi flags that AI displacement effects may become more visible under economic pressure, making the next downturn a critical inflection point for assessing the technology's structural role.
Footnote:
1 Tedeschi, Ernie. "AI, Work from Home, and the Young-Graduate Job Market." Stripe Economics, 12 Nov. 2026, https://www.stripeeconomics.com/p/ai-work-from-home-and-the-young-graduate.