Everyone is using it. Almost no one was taught how.

Depending on which survey you read, roughly three in four knowledge workers already use generative AI on the job, and most brought the tools in themselves, without any employer guidance. In the same populations, fewer than half report ever receiving AI training — and majorities say they rely on AI output without checking it, have seen it cause mistakes at work, and hide their use from their employer.

Two cautions before you lean on any of this. First, several of the most-quoted figures come from companies that sell AI (Microsoft and LinkedIn, McKinsey) — useful as adoption signals, not as independent measurement, and labelled that way throughout this page. The single strongest source is academic-led: a University of Melbourne and KPMG study across 47 countries and roughly 48,000 people. Second, none of these are the same thing as capability. Using a tool daily and using it well are different skills, and only one of them shows up in an adoption statistic.

Start with what the big numbers are and aren’t. The World Economic Forum’s employer survey projects 170 million jobs created and 92 million displaced by 2030, with 39% of core skills changing. Those are projections from a survey of employers — informed guesses about the future, not measurements of it. Treat them as the range of what leaders expect, not as facts about what will happen.

The more solid evidence is narrower and more useful: it’s about tasks changing, not jobs vanishing. Analyses like Jobs for the Future’s find most occupations being transformed rather than eliminated, with human skills still central to the large majority of top jobs. The work changes; the worker is still in it.

The first credible employment-levelsignal arrived in 2025, and it’s worth stating precisely. A Stanford working paper on payroll data found that since late 2022, employment for early-career workers — roughly ages 22 to 25 — in the most AI-exposed occupations declined about 13% relative to less-exposed peers, while overall employment held steady. That is the best available early signal, and it is also a working paper, about entry-level roles specifically, whose interpretation is contested. Best available and settled are not the same thing.

Two randomized studies point the same direction. In a Science experiment with 453 professionals, ChatGPT cut writing time about 40% and raised quality about 18% — and the largest gains went to the people who started out weakest. In a field experiment with more than 5,000 customer-support agents, AI raised productivity about 15% on average and roughly 35% for the least-experienced workers, essentially transferring the tacit know-how of top performers to novices.

Put those together and the equity case is straightforward: the workers with the most to gain from AI fluency are the newest, the least credentialed, the ones furthest from the frontier. And here is the problem. The OECD finds the supply of general AI-literacy training is “likely insufficient,” and skewed toward workers who are already skilled — while the low-skilled and automation-exposed are least likely to access it. Left to the market, the tool that could most help the people with the least reaches them last.

A randomized trial of junior engineers learning an unfamiliar library found the AI-assisted group scored 17 points lower on a later mastery quiz — worst of all on debugging — while their speed gain was not even statistically significant. In education, unrestricted AI raised students’ practice performance 48% but left them 17% worse on the unassisted exam; a version redesigned to give hints instead of answers erased the harm. The pattern is consistent: how you use the tool decides whether it builds capability or quietly removes it.

This is the bridge between a workforce strategy and a personal one. The same finding that should shape a training program should shape how you, individually, work with AI tomorrow morning.

Federal.The July 2025 America’s AI Action Plan includes a worker-focused pillar — AI-skills programs and apprenticeships — with Department of Labor guidance following in early 2026. Separately, the new Workforce Pell Grant program (effective July 1, 2026) extends federal aid to short, credential-bearing programs of 8 to 15 weeks. At the same time, the Digital Equity Act’s capacity grants were terminated in May 2025, removing one funding channel some states had planned to use for digital-skills work.

International context.The EU AI Act’s Article 4, in force since February 2025, is the first legal AI-literacy mandate, requiring organizations to ensure staff have a sufficient level of AI literacy.

North Carolina.The state’s AI Strategic Roadmap (July 2026) commits to deploying foundational AI-literacy training across all 100 counties through NCWorks, community colleges, and public libraries, and to credentialing more than 50,000 residents in AI skills by 2028. These are commitments with deadlines, not programs already operating — and the roadmap itself notes the state cannot yet attribute job losses to AI, because its layoff-reporting data was not built to capture it.