AI Reskilling Gap 2026: Only 6% of Companies Are Actually Training Workers

Eighty-nine percent of business leaders say AI skills are now critical for their workforce. Only 6% have actually started training anyone. That gap — between what companies say about preparing their people for AI and what they're actually doing — has become one of the defining disconnects of 2026.

The numbers come from a Metaintro analysis of current workforce data, and they line up with a pattern visible across multiple reports this month. DataCamp found that 59% of enterprise leaders admit their organizations have an AI skills gap, even though most claim to be investing in some form of training. The World Economic Forum estimates 59% of the global workforce will need reskilling by 2030. And the IMF's managing director, Kristalina Georgieva, has compared the AI-driven shift to a "tsunami" that will hit 40% of global jobs — rising to 60% in advanced economies.

So everyone agrees the wave is coming. Almost nobody is teaching people how to swim.

The $300 Billion Contradiction

Companies are spending $300 billion globally on AI systems in 2026. They're buying the tools, building the infrastructure, deploying the models. What they're not doing, in most cases, is preparing the humans who have to work alongside those systems.

A 2026 workforce-trends analysis found that 95% of AI pilot projects fail to reach scale. Not because the technology doesn't work, but because organizations treat AI as something bolted onto existing workflows rather than something that requires redesigning how people actually do their jobs.

The typical corporate response has been a webinar. Maybe an optional online course. Perhaps a lunch-and-learn with a slide deck about "prompt engineering." This is not reskilling. This is checkbox compliance dressed up as transformation.

The OECD's 2025 Skills Outlook spelled this out clearly: only 19% of adults without upper-secondary education participate in any form of non-formal learning, compared with 61% of university-educated workers. The people most at risk from AI disruption are the least likely to receive training for it. The gap isn't closing. It's widening.

Who Actually Gets Trained — and Who Gets Left Behind

The World Economic Forum's Future of Jobs report surveyed more than 1,000 major employers. About 80% said they planned to upskill workers with AI training. Two-thirds intended to hire people with specific AI skills. Only 40% expected to reduce headcount because of automation.

These sound like encouraging numbers until you look at what "upskilling" means in practice. When the WEF identifies the most in-demand skills — analytical thinking, creativity, collaboration, digital literacy — it's describing capabilities that take months or years to develop. A four-hour "Introduction to AI" course doesn't get anyone there.

The Fast Company analysis published this week went further, asking a question most corporate training plans avoid: can you actually retrain someone from a routine-task role into the kind of high-judgment, creative, AI-augmented position that's growing? Their answer was blunt. If filling those "top 20%" positions is already hard with experienced candidates, it doesn't get easier by trying to upskill hundreds of workers who've never done that kind of work before.

This isn't about intelligence or potential. It's about the gap between a two-day training program and the reality of what these new roles demand.

The DoorDash Paradox

This week, DoorDash launched a new "Tasks" app that pays its 8 million delivery couriers to film themselves doing household chores — loading dishwashers, folding laundry, opening doors — to generate training data for AI and robotics companies.

Read that again. Gig workers are being paid to teach machines how to do physical tasks. The same workers are not being taught new skills themselves. They're generating the data that trains the systems that may eventually replace the need for their labor.

DoorDash told Bloomberg the footage goes to "in-house AI models and those built by partners across retail, insurance, hospitality and technology sectors." The workers get paid per clip. The AI gets permanently smarter. The workers don't.

This is the reskilling gap distilled into a single product. The investment goes to the machine. The human gets a per-task payment. Nobody pretends it's training.

What the Numbers Mean for Education

The pressure is flowing downstream into classrooms and universities. The IMF has called for redesigning education so students gain cognitive, creative, and technical skills that complement AI rather than compete with it. Online learning platforms report explosive growth in AI-related courses. Universities are experimenting with AI tutors and adaptive learning systems.

But the OECD warns that without deliberate policy, the benefits will flow to people who are already well-educated. A university student with a laptop and curiosity can learn prompt engineering in a weekend. A factory worker whose job just got automated needs something more structured, more supported, and more expensive than what most companies or governments are currently offering.

The WEF identified the skills gap as the biggest barrier to business transformation, with nearly 40% of job skills expected to change by 2030 and 63% of employers citing it as their primary challenge. This isn't a problem that resolves itself. It requires investment that matches the scale companies are pouring into AI itself.

The Real Question

There's a version of this story where AI creates more jobs than it destroys — the WEF projects a net gain of 78 million roles by 2030, with 170 million created and 92 million displaced. That's possible. But it requires an assumption that workers displaced from vanishing jobs can actually move into new ones. And right now, 94% of companies aren't even trying to make that happen.

The executives know the tsunami is coming. The budgets for AI hardware and software are enormous. The budgets for teaching humans how to work in an AI-transformed economy are, for the vast majority of organizations, functionally zero.

That 6% figure isn't just a statistic. It's a policy failure in real time.