The skills AI is making more valuable may be the same skills it makes easier for us to stop practicing.

That tension runs through four pieces I read over the past couple of weeks.

Employers are placing a growing premium on judgment, curiosity, critical thinking, communication, emotional intelligence, and relationship-building. At the same time, AI is increasingly doing the everyday questioning, drafting, interpreting, deciding, and communicating through which many of those capabilities were developed.

The immediate work gets easier. But where do the human skills come from if we remove the experiences that used to build them?

The human premium has a practice problem.

There is a comforting argument about the future of work that goes something like this: AI will take over routine tasks, freeing humans to focus on higher-order capabilities like judgment, creativity, empathy, critical thinking, and relationships. There is plenty of evidence pointing in that direction.

A recent Forbes piece drawing on LinkedIn’s 2026 Skills on the Rise report and an analysis of more than 150,000 job postings found AI skills increasingly appearing alongside so-called soft or meta skills. Judgment, communication, curiosity, critical thinking, and interpersonal intelligence are moving closer to the center of employability.

Another piece in Inc. makes a similar argument around five “durable skills”: emotional intelligence, relationship-building, critical thinking, ethical judgment, and navigating ambiguity.

The message seems straightforward:

As AI becomes more capable, human skills become more valuable.

I think that is true, but leaves out an uncomfortable second half.

How do people develop those skills if AI increasingly removes the experiences through which they were built?

Nobody became good at judgment by reading the definition of judgment. You developed it by making decisions with incomplete information, getting some of them wrong, seeing consequences, adjusting, recognizing patterns, and doing it again.

You developed critical thinking by wrestling with ambiguous evidence.

You developed communication by trying to explain something, watching it fail to land, and trying differently.

You developed empathy through interactions with actual people whose perspectives did not neatly match your own.

You developed curiosity because you encountered something you did not understand and had to stay with the uncertainty long enough to explore it.

Those were reps.

And increasingly, AI can remove them.

Ask it to identify the questions.

Ask it to summarize the evidence.

Ask it which option is best.

Ask it to draft the difficult message.

Ask it to turn a vague thought into a polished recommendation.

All of those uses can be productive. I use AI for some of them myself. But productivity and capability are not the same thing. You can produce better work while getting fewer opportunities to develop the capability underneath it. That creates what I think of as the experience-ladder problem.

Junior work has historically contained a lot of repetition. Some of it was tedious and absolutely deserves to disappear. But hidden inside that work were opportunities to notice, compare, struggle, make mistakes, receive feedback, and slowly develop the judgment expected at more senior levels.

If AI removes the task, we need to ask what else disappeared with it.

Not every lost skill deserves saving. I have no desire to keep my long-division abilities sharp in case Excel goes down.

The useful question is not: What should humans never delegate?, but: What capability do we still need to possess even after we delegate much of the work?

A Chief Learning Officer article on AI and deskilling makes this distinction particularly well. Deskilling itself is not inherently bad. Technology has been removing the need for human skills for centuries. The sharper question is how much deskilling is acceptable, in which contexts, and with what safeguards. The article uses a useful analogy: pilots still practice manual landings. Surgeons should retain the ability to operate when robotics fail.

We could apply the same test far beyond aviation or medicine. If the AI disappeared tomorrow, which things would you simply do less efficiently? And which things would you discover you no longer knew how to do well? That is the manual landing test.

But one small study on empathy suggests another possibility. Ten adults practiced structured interactions with AI personas designed to exercise perspective-taking, emotional labeling, validation, and repair-oriented communication. Nine of the ten improved their empathy scores. With ten participants and no control group, this is nowhere near proof that talking to AI makes people more empathetic. The interesting part is how AI was used.

The machine did not perform the empathy for them. It created situations in which they had to practice empathy themselves.

That is a fundamentally different model of AI use. AI can remove reps. Or AI can create reps.

Instead of:

Write the difficult response for me.

Try:

Play the person I need to speak with. Make me respond. Then tell me what I missed.

Instead of:

Tell me which option is best.

Try:

Give me a case with imperfect alternatives. Make me decide. Then challenge my reasoning.

Instead of:

Give me five good questions.

Try:

Here is what I think I need to understand. Ask me questions that expose where my curiosity stopped too early.

The difference looks small, but it's not. In one case, the AI performs the capability; in the other, AI becomes a practice environment for yours. That may be one of the more important choices we make as AI becomes woven into everyday work. The human premium is not automatically protected because we happen to be human. It has to be practiced.

THE TELL

Here is how to recognize when AI may be improving your output while quietly removing the reps underneath it:

• AI increasingly produces the first attempt at work you once had to struggle through yourself.
• You are getting better at choosing among generated answers but less confident producing a strong answer without them.
• You can recognize that an AI recommendation is weak but struggle to explain what a substantially better one would look like.
• Your first response to ambiguity is to ask AI to resolve it rather than spending time inside the uncertainty.
• Difficult conversations increasingly begin with AI writing what you should say.
• Junior people reach polished outputs without encountering the mistakes and messy intermediate work that built judgment in more experienced colleagues.
• You know how to improve an AI-generated artifact but have difficulty reconstructing its underlying reasoning without returning to the tool.
• When AI is unavailable, the loss feels less like inconvenience and more like incapacity.

The danger is not merely that AI might occasionally be wrong. It is that AI can be right often enough that we stop practicing what we will still need when it isn't.

YOUR PRACTICE THIS FORTNIGHT

🛠 The Rep Audit

Pick three things you now routinely use AI to help you do. For each one, ask: What human repetition did AI replace?

Then put the lost repetition into one of three categories:

Safe to delegate The underlying capability has little continuing value. Let the machine have it.

Worth retaining You do not need to perform it constantly, but you still need enough capability to recognize failure, intervene, and recover.

Worth developing The repetition helps build something that becomes more valuable as AI handles routine work: judgment, empathy, curiosity, critical thinking, communication, advocacy, orchestration, shaping, or another capability central to your work.

Now choose one item from the last two categories.

Redesign how you use AI so it creates the rep rather than completing it for you.

Examples:

Instead of asking AI to make the decision, make your decision first and have it challenge you.

Instead of asking AI to write the message, rehearse the conversation with it.

Instead of asking AI what the research means, form your interpretation and ask it to find weaknesses.

Instead of asking AI for the finished concept, create a rough direction and use it to force genuinely different alternatives. You still get the leverage. But you keep the rep.

🤖 THE SPARRING PROMPT: MAKE ME PRACTICE

Use this with any human capability you want to strengthen:

I want to practice [SKILL]. Do not perform the skill for me. Create a realistic situation in which I have to exercise it myself. Give me enough information to begin, but leave genuine ambiguity or tension that I need to work through. Let me respond before you evaluate anything.

After I respond: 1. Identify what I did well. 2. Point out what I overlooked, assumed, or handled weakly. 3. Ask one question that forces me to reconsider my reasoning. 4. Give me another round that is slightly more difficult.

Do not give me the ideal answer unless I explicitly ask for it after I have completed the exercise.

You can use this for:

• Critical Judgment: imperfect decisions with competing evidence.
• Curiosity: situations where the obvious questions miss something important.
• Empathy: difficult conversations from another person's perspective.
• Advocacy: competing interests where someone's needs risk disappearing.
• Orchestration: multiple contributors producing individually reasonable but collectively incoherent work.
• Shaping: an emerging intention that needs to become something tangible without jumping directly to a polished answer.

The important instruction is the first one:

Do not perform the skill for me.

That is PAIR in practice: Prime by engaging your own capability. Assess your performance with AI. Interpret the feedback rather than merely accepting it. Retain the learning so the next rep begins from a stronger place.

THE ARTICLES

📄 AI Is Creating A ‘Human Premium’ — 5 Career Skills Suddenly Worth More | Forbes

Drawing on LinkedIn’s Skills on the Rise report and other employment data, Bryan Robinson argues that judgment, communication, curiosity, critical thinking, and interpersonal intelligence are becoming more central to employability alongside AI skills. The useful implication is not that humans should compete with AI at tasks AI performs well. It is that widespread access to similar models makes the capabilities surrounding the tool more differentiating. When everyone has access to the answer machine, knowing which answers deserve confidence becomes more valuable.

📄 AI Is Making 5 Human Skills More Valuable Than Ever. Most Companies Are Investing in the Wrong Thing | Inc.

Soren Kaplan focuses on emotional intelligence, relationship-building, critical thinking, ethical judgment, and navigating ambiguity as durable capabilities across technological changes. The part that deserves more attention is the development problem. If AI absorbs increasingly large portions of the routine work through which less-experienced people once learned, organizations cannot simply assume that senior judgment will materialize later. Removing work is easy. Replacing the experience it created is harder.

📄 Deskilling, AI and the Questions We Should Be Asking | Chief Learning Officer

Treca Bourne offers a useful correction to the idea that all deskilling is inherently harmful. The question is not whether technology should ever replace human capability. Of course it should. The question is what capabilities must remain available for resilience, oversight, recovery, or high-stakes situations. Think of pilots practicing manual landings. For each capability you delegate, decide whether losing it is an acceptable trade rather than discovering the answer accidentally later.

📄 AI Persona Practice Boosts Empathy Scores in Small Pilot Study | Phys.org / University of Phoenix

Ten adults completed two structured interactions with AI personas designed to exercise perspective-taking, emotional labeling, validation, and repair-oriented language. Nine improved their empathy scores. The sample is tiny and there was no control group, so the study should be treated as an exploratory signal, not evidence that AI causes durable empathy gains. But the mechanism is intriguing. AI did not substitute for human empathy. It became a rehearsal environment in which people could exercise it. That points toward a much more productive question than whether AI strengthens or weakens human skills: How are we choosing to use it?

ONE THING TO TRY

The next time you are about to ask AI to do something that requires judgment, empathy, curiosity, or critical thinking, change one sentence in your prompt.

Instead of: Do this for me., try: Make me do this.

Have AI create the situation, provide the counterargument, play the other person, reveal the blind spot, or critique your attempt.

But keep the rep. Then notice whether the interaction feels slightly harder. That friction may be the point.

The human edge is not automatic. It has to be built.