by Kara Kothmann
Nearly every workplace tech conversation I’ve been in this year eventually lands in the same place: we need to upskill people on AI. The technology is moving quickly, the skills employers need are changing with it, and companies are trying to make sure their people can keep up.
But I keep coming back to a different question: Where are people supposed to learn these skills in the first place? Because right now, we’re asking employers to solve a skills gap that didn’t start when someone entered the workforce. And the data suggests we need to look much earlier.
A few weeks ago, the largest school district in the country gave us an interesting example of what that could look like.
The demand for AI skills is already here
For the first time, AI capabilities top the global list of hard-to-fill skills. ManpowerGroup surveyed more than 39,000 employers across 41 countries and found AI Model & Application Development leading the list at 20%, with AI Literacy right behind at 19%, ahead of engineering, sales and marketing, and manufacturing. At the same time, 72% of employers said they’re struggling to fill roles overall, and their most common response is to upskill and reskill the people they already have (ManpowerGroup, February 2026).
The second-place skill is the one I find particularly interesting. Not AI engineering. AI literacy.
We’re talking about the baseline ability to understand these tools, use them appropriately, question their outputs, and work alongside them. That’s becoming less of a specialized technical skill and more of an expectation for participating in the workforce at all.
And the baseline is moving quickly. According to Lightcast’s analysis of the 2026 Stanford AI Index, 2.5% of U.S. job postings now mention AI skills, up 55% in a single year and nearly 300% over the decade (Lightcast analysis of the Stanford AI Index 2026).
So yes, employers need to upskill their people. But I’m convinced that’s only part of the conversation.
We’re asking employers to close a gap that starts much earlier
This is the part I keep getting stuck on. Gallup and the Lumina Foundation surveyed 2,000 employers and thousands of students for their State of Higher Education 2026 report. Ninety-three percent of undergraduates felt their education had prepared them for work. Only 54% of employers agreed that colleges are producing graduates with the competencies their businesses need. And 70% said recent graduates require moderate or significant additional training (Gallup/Lumina, via The Washington Times).
That 39-point gap tells us something is getting lost in the handoff between education and employment. When you look specifically at AI, it gets even more interesting. CarringtonCrisp’s See the Future 2026 study found that 77% of employers expect new hires to arrive with some experience using AI tools, while 58% say universities aren’t doing enough to build those capabilities. So where are graduates learning AI? Mostly on their own. Seventy percent said they taught themselves by downloading tools and experimenting. Sixty-four percent learned through online videos. Just 3% completed a university course focused on AI (CarringtonCrisp via Poets&Quants).
You read that right. Just three percent.
The Digital Education Council’s 2026 global survey tells a similar story. Among more than 45,000 students and faculty across 35 countries, 43% of students said they hadn’t encountered AI integrated into any of their courses. Only 29% believed their instructors were well equipped to guide them on AI use. In the U.S. and Canada, that number dropped to 17% despite 64% of faculty reporting that they had completed AI literacy training (Digital Education Council, July 2026).
We’re training faculty, but students aren’t necessarily feeling the impact yet. And then they graduate and the same pattern continues. iCIMS surveyed 1,000 U.S. job seekers and found the share teaching themselves AI skills increased from 22% to 30% year over year, while the share receiving AI training from an employer remained at roughly 16%. Nearly half had built AI skills in the previous six months, but 61% said their proficiency stopped at general-purpose chatbots (iCIMS Insights, September 2026).
Different stage of life, same behavior: people are largely figuring this out themselves.
Eventually employers inherit that inconsistency and have to sort it out on the back end.
K-12 is moving, but access matters
The encouraging part is that education isn’t standing still. FutureEd is tracking 77 bills across 27 states this legislative session addressing AI in classroom instruction, including AI literacy requirements, governance frameworks, and state-funded professional development for teachers.
At the federal level, the 2025 executive order on Advancing Artificial Intelligence Education for American Youth directed Education Department priorities toward AI teacher training and Labor Department efforts toward AI apprenticeships for high school students.
But as a former educator, the part I’m watching closely is whether schools actually have the resources to make any of this real. RAND’s American School District Panel found that the share of districts offering AI training to teachers jumped from 23% to 48% in a year. That’s meaningful progress. But 67% of low-poverty districts offered training compared with just 39% of high-poverty districts (RAND, American School District Panel).
That gap matters. If AI literacy becomes a baseline workforce expectation – and the labor data suggests it’s moving in that direction – then who gets meaningful exposure to it in school becomes a workforce question too.
Then New York drew a very public line
This month, New York City gave us one of the clearest examples yet of how complicated this is going to be. On September 2, Mayor Mamdani and Chancellor Kamar Samuels announced a one-year moratorium on student-facing generative AI from 2-K through eighth grade. Companion chatbots are prohibited across grade levels, and the district also introduced recommended daily screen-time limits. Teachers can still use approved AI tools for things like planning, translation, and drafting communications, while certain higher-stakes uses, including grading, behavior monitoring, and special education plans, are restricted. At the high school level, the district is allowing five vetted AI pilots reaching up to 50,000 students (NYC Mayor’s Office, September 2, 2026; Chalkbeat New York).
A lot of the coverage called it an AI ban. I think the more interesting part is the distinction New York made between using AI and learning about AI. Alongside those restrictions, the district is requiring every high school student to complete two 45-minute AI literacy modules each year covering how the technology works, bias, and misinformation.
So the largest school district in the country is simultaneously saying: younger students don’t necessarily need unrestricted access to generative AI tools, but older students do need to understand the technology they’re going to encounter outside of school.
Is 90 minutes a year enough to meaningfully build AI literacy? Probably not on its own. But I’m less interested in whether New York got the exact number of minutes right than I am in the framework underneath it.
AI in the classroom and AI literacy in the classroom are not the same thing. We’ve spent a lot of time debating whether students should use AI. We need to spend just as much time figuring out what they should understand about it, at what age, and who is responsible for teaching it.
This is one skills conversation, not three
We tend to talk about K-12 AI literacy, higher education workforce readiness, and corporate upskilling as separate issues. I’m not sure they are. They’re different points along the same continuum.
For those of us working in education and workplace technology, I think that matters for how we talk about what we’re building, too. It’s easy to focus on the newest AI feature or capability. The bigger question is whether the people expected to use these tools actually understand how to use them well.
A few things I’d like to see more of:
- Employers getting much more specific about what “AI literacy” actually means. If education is supposed to prepare students for an AI-enabled workplace, educators need something more concrete than “AI skills” to build toward.
- Higher education treating AI fluency as a general competency. If AI is going to touch finance, marketing, healthcare, communications, education, and nearly every other field, it can’t live only in computer science departments.
- K-12 teacher training treated as infrastructure. The gap between high- and low-poverty districts should concern anyone thinking about the future workforce.
- More discussion about what meaningful AI literacy actually requires. New York’s 90 minutes is a starting point. What should a student know by the time they graduate?
- A clearer distinction between using AI and understanding AI. Schools can decide that a particular tool isn’t appropriate for a particular age while still deciding that understanding the technology is essential.
I don’t think the corporate upskilling push is misguided. Employers have a real skills gap in front of them, and they have to respond to it. But we’re trying to solve the problem pretty late in the chain. If AI literacy is becoming a baseline workforce skill, the conversation can’t begin when someone gets their first corporate login. It has to start years earlier with what we teach, when we teach it, and how we make sure students actually have access to it.
