The AI Jobs Canary: Young Workers Are Already Paying the Price

A sweeping new labor market dashboard built on payroll data from 4.6 million workers across 730+ occupations is offering the clearest real-time picture yet of how AI is reshaping employment — and the signal is sharpest at the bottom of the career ladder. Stanford economist Erik Brynjolfsson, partnering with ADP Research, launched the Canaries Dashboard to continuously track AI's labor market effects using data that covers roughly one in six American workers. The headline numbers look calm. Dig deeper, and they're not.

For workers ages 22 to 25 in the most AI-exposed occupations, employment is now shrinking at 3.8% per year as of April 2026 — a trend that has accelerated since it was first detected. In 2024, that decline stood at 2.8%; it has since surpassed 4% annually. Meanwhile, the least AI-exposed jobs for that same age group are growing at 2% per year. Mid-career workers ages 31 to 34 are also contracting, down 1.7% year-over-year, while workers 35 to 40 are actually growing at 2%. The pattern is consistent: AI is not eliminating work broadly — it's eliminating the entry point into careers.

The mechanism is straightforward. AI absorbs tasks before it absorbs entire jobs, and the tasks it reaches first — retrieving, summarizing, scheduling, formatting, assembling information — are precisely those handed to people early in their careers. Senior workers carry hard-to-codify, experience-based judgment that still buffers them from displacement. Junior workers haven't accumulated that yet. ADP chief economist Nela Richardson frames the dividing line as augmentation versus automation: where AI amplifies human work, employment grows; where it automates tasks outright, it contracts — and early-career workers sit squarely in that second category.

Brynjolfsson has stress-tested the finding against every major counter-argument raised since his original paper last August. Critics blamed interest rates, tech-sector overhiring, remote work distortions, and pandemic noise. He removed the entire tech sector from the dataset, isolated remote-work effects, and checked rate-sensitive industries like construction — which have the lowest AI exposure. The pattern held every time. The new dashboard extends the data to April 2026, nearly four years post-ChatGPT, and the effect hasn't mean-reverted. It has grown by roughly half a percentage point per month, consistently.

The broader debate at the top of the economics profession is shifting. Brynjolfsson and MIT Nobel laureate Daron Acemoglu remain publicly at odds — Acemoglu produces far lower AI productivity estimates and has called much of the AI productivity discourse speculative to the point of fiction. But the argument is no longer about whether AI is transformative; it's about magnitude and timeline. Brynjolfsson compares the disruption not to the internet or globalization but to the Industrial Revolution — and predicts it will be bigger and ten times faster. He holds a public bet on longbets.com with Northwestern economist Bob Gordon that productivity will be significantly higher by decade's end, and says he's already ahead.