The Skill AI Can’t Fake

Ask a room of final year students what will matter most at work in an AI heavy world and most will give the same answer: learn the tools. It’s a sensible answer. Employers just don’t seem to agree.

What Employers Actually Want: The Survey Data

Drexel University’s LeBow College of Business asked 647 employers what they need from early career hires, with fieldwork between August and October 2025. The authors are upfront that the sample leans toward small and mid size firms, so treat the results as directional. On a one to five scale, reliability came out on top at 4.87. Professionalism followed at 4.80. Critical thinking and judgment landed third at 4.61, ahead of teamwork at 4.52 and communication at 4.51.

Now look at the bottom of the list. Automation tools scored 2.65. Prompt engineering and AI tools scored 2.75. Interpersonal and cognitive skills averaged 4.38, while technical skills averaged 3.11. Bigger organisations do put more weight on advanced technical skills, so it isn’t uniform. Still, for the typical employer here, someone who turns up, thinks clearly and can be trusted beats someone with a long list of tools on their CV.

The Judgment Deficit in the Workplace

Microsoft’s 2026 Work Trend Index adds a sharper number. It polled 20,000 workers in ten countries and found that only 16% have built the judgment to switch between directing AI and doing the work themselves. Most people can ask a system for an answer. Far fewer know when to trust it, when to fix it and when to ignore it and do the job by hand.

Productivity research points the same way. Morgan Stanley analysis, cited in a Fortune commentary in July, found that output per worker rose sharply in heavily AI exposed industries while employment stayed relatively steady, and that the biggest gains went to people who know how to direct AI. Knowing how to direct something is a judgment skill. You can’t direct what you can’t evaluate.

The Looming Threat of Talent Debt

That same commentary argued that the more capable AI becomes, the more valuable human judgment becomes. It also raised a less obvious worry. If companies let AI swallow the entry level tasks, they build up a talent debt. Think about how you learned your own first job. Small tasks, someone senior correcting your mistakes, a bit more responsibility each month. If that ladder disappears, where do tomorrow’s experts come from?

The piece adds that a system built only for output will produce workers who can operate AI without really steering it, and that AI literacy without critical thinking isn’t a workforce strategy. That’s a handy test for any training programme. Can your people explain when the tool is wrong? If they can’t, you’ve taught them which buttons to press.

Teaching Judgment on Purpose

So what should universities do? Put judgment in the curriculum on purpose. It won’t turn up by accident, and it’s harder to measure than a checklist of software skills, which is probably why it gets squeezed. The things that build it are well known. Projects with fuzzy edges. Evidence that disagrees with itself. Ethical dilemmas with no tidy answer. Permission to fail and try again. A student who has only ever done neatly defined tasks will freeze the first time a real client hands them something messy.

One technique deserves more use: judge the route, not just the destination. Drafts, reflections and spoken explanations show how a student reached a conclusion, and that’s the part employers care about. Anyone can produce a good looking answer now. Fewer people can say why it’s right, who it might affect and what they’d do if it turned out wrong.

Employers Cannot Sit Out

Employers can’t sit this one out either. Structured, mentored learning on the job is one of the few dependable ways to build judgment at scale. Pair juniors with experienced colleagues. Ask them to explain their reasoning, not just deliver results. Give them real responsibility early enough that they feel the consequences. Cutting those roles to save money may also cut the supply of future leaders.

There’s an equity angle that doesn’t get enough attention. Community colleges enrol roughly 40% of US undergraduates, and many more future workers come through regional universities and historically Black colleges. They teach a large share of first generation students and working adults, and they’re often missing when people talk about how AI skills should be taught and funded.

The Part You Can't Hand Off

For students, the advice is almost old fashioned. Get properly good at something. Learn to write and speak clearly. Stay curious. Then add enough AI fluency to understand how it can stretch what you can do, and where you need to stay in charge. Someone who can make something impressive with AI is useful. Someone who can decide whether it should exist, whether it’s right and what to do when it fails is worth a lot more.

Employers aren’t asking for another tool on the CV. They want a person who can take a problem, work out what matters, explain it plainly and own the result. That was always the hard part of any job. AI has simply made it the part you can’t hand off.