Jeff Raikes: The talent debt I warned about is now showing up in the data

Jeff Raikes is the co-founder of the Raikes Foundation, which works to ensure every young person in America has an opportunity to thrive. He is the former CEO of the Bill & Melinda Gates Foundation and former President of Microsoft's Business Division.

In April, I warned in these pages that companies letting AI absorb entry-level work were quietly accumulating a talent debt they didn’t yet see. Since then, new evidence is putting the problem in plainer sight.

Microsoft’s 2026 Work Trend Index surveyed 20,000 workers across ten countries and found that the most valuable human skills in an AI workplace aren’t technical. They’re the ability to think critically and judge AI’s outputs. Only 16 percent of workers have developed the judgment to move fluidly between directing AI and doing the work themselves. Microsoft calls them “Frontier Professionals.” The detail that stuck with me is that those workers deliberately do some tasks without AI, specifically to keep their own thinking sharp.

They are onto something. Gartner predicts that through 2026, the atrophy of critical-thinking skills will push half of all global organizations to require “AI-free” skills assessments. A RAND Corporation study published this spring found that most students using AI for homework are worried it is hurting their ability to think independently. Employers and students are arriving at the same conclusion from opposite ends: the more capable AI becomes, the more valuable human judgment becomes.

AI is collapsing a distinction we’ve long taken for granted. The skills that make someone valuable at work are increasingly the same ones that make someone an effective citizen.So here is the question underneath all of it. Are we building an educational system that produces critical-thinking skills at scale and develops the best possible future workforce?

I don’t think so. But here’s how to do it.

Two visions, one uncomfortable truth

Most colleges and universities are racing toward what I’d call the efficiency model of AI: cheaper credentials, faster output, and more targeted........

© Fortune