The NY Fed Just Reframed the CS Bench: 64% Remote, Not AI
The NY Fed's June 2026 paper blames remote work, not AI, for 64% of the new-grad unemployment spike. Here is how to source the juniors other teams skip.
On June 1, 2026, the New York Fed published "Remote Work Leaves Younger Workers Sidelined" and quietly torched the story most of your hiring managers have been telling themselves for two years. The CS new-grad bench is not, primarily, an AI displacement event. It is a mentorship-availability event. If you can put a junior engineer next to a senior one, two days a week or five, you have access to a credentialed talent pool that distributed-team competitors have self-selected out of touching.
That is the arbitrage. Here is how to read the data and how to actually source it.
What the NY Fed paper actually says
Natalia Emanuel (NY Fed), Emma Harrington (UVA), and Amanda Pallais (Harvard) ran the numbers on the 2022 to 2025 increase in unemployment among young college graduates and concluded that remote work alone can explain 64 percent of it. Not AI. Not vibes. Remote work.
The mechanism is unglamorous and correct: employers do not want to hire inexperienced people into roles where mentorship cannot be delivered over a desk. In jobs that require physical presence (the paper's example is nursing), the age gap spiked briefly in 2020 and then normalized. In remote-eligible work, it never did. The gap is entirely concentrated in remotable fields, with software engineering and financial analysis named explicitly.
The headline numbers underneath:
- Unemployment for young college graduates rose to 5.6% in March 2026, up from 3.6% in March 2019.
- Average unemployment for under-29 college grads was 3.1% in 2017 to 2019. It rose to 3.7% in 2022 to 2025.
- For more experienced college grads, the rate actually declined over the same window, from 1.9% to 1.8%.
The paper is also explicit on one point worth printing out and taping to your monitor: the uptick in youth unemployment predates the rapid diffusion of AI. The story did not start with ChatGPT. It started with Zoom standups and a generation of managers who decided they could not afford to train someone they could not see.
The AI counter-evidence, and why both can be true
Stanford's Digital Economy Lab (Brynjolfsson, Chandar, Chen, "Canaries in the Coal Mine?") found that workers aged 22 to 25 in AI-exposed occupations saw a roughly 16% relative decline in employment after generative AI went mainstream. For software developers specifically in that age band, the FindSkill.ai summary of the Stanford AI Index 2026 puts the drop at nearly 20% since 2022.
Both papers can be true. The NY Fed identifies the bigger effect (remote-mentorship aversion) and Stanford identifies a real but smaller AI-exposure effect that survives even after you strip out remote-friendly roles. The operational implication is sharper than the academic debate suggests: an in-office or structured hybrid hiring strategy does not eliminate the AI headwind. It eliminates the much larger remote-mentorship headwind sitting on top of it.
That is why ServiceNow CEO Bill McDermott's March prediction that new-grad unemployment could "reach the mid-30s within a couple of years" reads, after the NY Fed paper, as a category error. He is naming an AI problem. The Fed found a real-estate problem.
The supply side: why the bench is this deep
CS degree production roughly doubled in a decade, from 51,696 in 2013 to 2014 to 112,720 in 2022 to 2023. The pipeline ramp landed exactly when remote-friendly junior roles collapsed. Indeed's 2026 Tech Hiring Outlook reports that remote-friendly junior roles have declined 71% since 2022. The safety valve closed at the same moment the cohort doubled.
number: 71%
label: Decline in remote-friendly junior roles since 2022, per Indeed's 2026 Tech Hiring Outlook
note: The remote junior market did not soften. It vanished.