Why We Know Less Than We Think About AI and Jobs
Fewer than 20% of firms even use AI. Youth jobs in exposed roles are falling — but unemployment is not rising. Brookings says AI labor research is "still in the first inning."
The Confidence Gap
Open any news site and you will find confident predictions about AI and jobs. Millions of roles will be automated. Entire industries will be reshaped. The transformation is already underway.
But according to Jed Kolko, writing for the Brookings Institution in March 2026, the honest truth is far more humble: we do not really know what is happening yet. His analysis, "Research on AI and the Labor Market Is Still in the First Inning", argues that the gap between public certainty and actual evidence is dangerously wide.
Start with a basic fact that rarely makes headlines: according to the U.S. Census Bureau's Business Trends and Outlook Survey (BTOS), fewer than one in five firms are using AI in any capacity. Not "using AI to replace workers" — using AI at all, for anything. And among those that do, even fewer are deploying it directly in the production of goods and services. The breathless narrative of an AI-transformed economy is running far ahead of what most businesses have actually done.
This matters because the labor market data we do have is shaped by this early, uneven adoption. Drawing sweeping conclusions from what is essentially a pilot phase is exactly the kind of mistake researchers — and workers — should be cautious about.
Contradictory Signals Everywhere
Kolko highlights a troubling pattern in the research: studies keep contradicting each other, even when using similar data.
Take youth employment. A study by Brynjolfsson and colleagues (2025) found that employment fell more for young workers in occupations with high AI exposure compared to those in lower-exposure roles. That sounds alarming — and for young people considering careers as customer service representatives or administrative assistants, it could feel personal. But the same data shows minimal differences for older workers across the AI-exposure spectrum. Why would AI selectively affect younger workers but leave older ones untouched? One possibility: younger workers are simply more mobile and responsive to early signals, leaving exposed roles before they are forced out. Another: the data is too noisy and the timeframe too short to distinguish real AI effects from normal labor market churn.
Meanwhile, a separate analysis by Eckhardt and Goldschlag (2025) found the opposite trend for unemployment: workers in higher AI-exposure occupations actually saw unemployment rise less than those in lower-exposure roles. If AI were already displacing workers at scale, you would expect the reverse. Brookings' own analysis of recent employment data also found no evidence of an AI jobs apocalypse — at least not yet.
For software developers — one of the most-discussed AI-exposed professions — the picture is similarly muddled. Coding assistants like GitHub Copilot and Claude are widely adopted, yet developer unemployment has not spiked. Hiring has cooled, but that is entangled with interest rates, tech sector corrections, and post-pandemic normalization. Isolating an "AI effect" is genuinely difficult. Research from ADP and Stanford HAI underscores how challenging it is to separate automation effects from broader economic shifts.
History Offers a Humbling Perspective
One of Kolko's most striking arguments involves historical comparison. Many analysts point to the current moment as unprecedented — a technological disruption unlike anything we have seen. But the data tells a different story.
According to Kolko's Brookings analysis, occupational shifts between 2019 and 2024 — the period when generative AI emerged — occurred at roughly the same pace as shifts after 1984 (the personal computer era) and after 1996 (the internet era). In other words, the rate at which people are moving between occupations has not noticeably accelerated since AI tools became mainstream.
And here is the real humbler: the occupational shifts of the 1910s through 1950s — when agriculture mechanized, manufacturing surged, and millions of workers moved from farms to factories — were far more dramatic than anything we are seeing today. The current AI moment, at least so far, looks more like a continuation of slow, steady technological evolution than a sudden rupture.
This does not mean AI will not eventually cause dramatic shifts. It means we may be in the earliest phase of a long transition — and the most important changes may still be years away.
Why "We Don't Know" Should Worry You More Than Certainty
Kolko identifies what he calls a potential "narrator's bias" among researchers. Academics and analysts who use large language models daily may be more inclined to assume these tools are transformative — because they feel transformative in their own work. But the experience of a researcher at a think tank using ChatGPT to draft memos is very different from the experience of a factory floor worker, a nurse, or a truck driver.
For workers trying to plan their careers, the uncertainty is actually more important than any specific prediction. If we knew AI would automate customer service within three years, you could plan accordingly. But the reality is messier: it might happen in three years, or ten, or it might evolve in ways nobody expects — transforming the job without eliminating it.
The practical takeaway is not complacency. It is preparation without panic. Understand your own role's AI exposure — our data on occupations like customer service representatives, software developers, and administrative assistants can help with that. But treat any confident prediction about timelines with healthy skepticism. The research, as Kolko puts it, is still in the first inning.
Sources
- Kolko, J. (2026). "Research on AI and the Labor Market Is Still in the First Inning." Brookings Institution. brookings.edu
- Brookings Institution (2026). "New Data Show No AI Jobs Apocalypse — For Now." brookings.edu
- ADP Research / Stanford HAI (2025). "Assessing the Real Impact of Automation on Jobs." hai.stanford.edu
- U.S. Census Bureau. "Business Trends and Outlook Survey (BTOS)." census.gov
Update History
- 2026-03-19: Added source links and ## Sources section
- 2026-03-15: Initial publication
This article was researched and written with AI assistance using Claude (Anthropic). Key findings are drawn from Jed Kolko's March 2026 analysis at the Brookings Institution. The interpretation reflects AI-generated analysis of public research and should not be taken as professional career advice. We encourage readers to consult the original sources linked above.