My fellow pro-growth/progress/abundance Up Wingers,
I’ve got something for everyone today. If you’re an AI optimist, this essay shows generative AI is having real-world impacts. So not algorithmic vaporware!
And if you’re an AI worrier, well … not everyone will love the impacts.
For the latter group, you can relax a bit — well, most of you. Artificial intelligence is not gutting the American job market. But it is starting to reshape it in telling ways. In “Canaries in the Coal Mine? Six Facts About the Recent Employment Effects of Artificial Intelligence,” Stanford economists Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen analyze payroll data from ADP, the country’s largest processor, and find that entry-level jobs are where the impact is landing first.
For 22- to 25-year-olds in the most AI-exposed jobs, employment has fallen 13 percent relative to peers since late 2022. The biggest casualties are in roles where generative AI substitutes directly for entry-level tasks: software developers, customer service reps, accountants, and receptionists. According to the Stanford team’s ADP analysis, headcount for 22- to 25-year-old software developers is nearly 20 percent lower than its late-2022 peak.
(In an explanatory Substack essay, Chandar also notes the team tested and controlled for broader factors such as tech-sector cutbacks and pandemic-related disruptions, but still found a significant AI-specific effect.)
The researchers suggest part of the reason here is that young workers mostly bring codified knowledge — the kind of book learning and standardized skills that AI systems can replicate quickly. Older workers, by contrast, draw more on what economcis term “tacit knowledge”: the practical judgment, workplace know-how, and interpersonal skills built up over years, which AI struggles to mimic. Those jobs have held steady or grown, and they tend to slot into the augmentation-heavy category rather than the automation-heavy one.
Augmentation-heavy roles include registered nurses, cooks, welders, executives, and IT managers, according to the paper. In practice, that might mean nurses using AI to help check diagnoses or streamline paperwork, cooks and welders tapping planning tools, or managers brainstorming strategy with an AI assistant. These examples aren’t in the study itself, but I think they illustrate the kind of augmentation the researchers have in mind.
To put it another way, automation cuts entry-level employment, while augmentation does not. That’s why job losses show up in software engineering — where AI systems can now solve 70 percent of benchmark coding problems, up from four percent in 2023 — while employment remains robust in health care or management roles where human judgment still dominates.
As Brynjolfsson told the Wall Street Journal:
Simply automating tasks that people do can save money, but it doesn’t really create anything new, said Brynjolfsson. What’s more valuable is doing new things that extend people’s capabilities, leading to gains that encourage businesses to hire more, rather than fewer, people. “I was delighted to see in the data that indeed, this augmentation approach could benefit people and lead to more employment,” he said.
While the findings here are valuable, they’re maybe not so surprising. Technological revolutions rarely spread their benefits evenly. New tech tools tend to displace some groups of workers even as they enrich others, creating a painful adjustment period before broader gains materialize. The rise of generative AI looks to be no different.
Indeed, the Stanford economists note that earlier upheavals — from the spread of information technology in the 1990s back to the mechanization of agriculture —produced winners and losers in the short run, but ultimately delivered higher employment and rising real wages once firms and workers adapted. That should remain your baseline expectation. Today’s entry-level coders may be suffering, but the longer-term pattern could echo past transitions: disruption first, growth later.
By the way, a new analysis from the St. Louis Fed reaches many of the same broad conclusions at the Stanford study: AI’s impact is showing up first in white-collar jobs such as software development, with evidence of displacement in some roles and augmentation in others. Also: Both stress that these are early signs, and the longer-term balance between disruption and new job creation remains uncertain.
As the St. Louis Fed study concludes:
The data raise important questions about our economic transition into the AI era. If current trends continue, new approaches to workforce development, social safety nets and economic support for displaced workers may merit greater consideration. As we navigate this technological revolution, understanding potential early warning signs could prove crucial for protecting U.S. workers and maintaining economic stability.
