Stop Defining Human Skills by What AI Cannot Do

Every list of essential human skills I have read in the last three years was assembled the same way. Someone looked at what the current AI models could not do, wrote those things down, and called it a framework.

In 2023 the list included long-form writing. Then it did not.

In 2024 it included coding. That one is going fast.

This year the list includes taste, judgment, and "asking good questions," which feels appropriate right up until you watch a model interrogate its own output more rigorously than most graduate students do.

I am not saying those things do not matter, but the method seems broken. When you define human value as what is “left over” after the machines take a pass, you have handed the definition of your work to a product roadmap you do not control. Every capability release forces a revision. The list gets shorter. The teacher who built a semester around last year's version has to rebuild it.

Let’s also be real…the technology is not narrowing. It is widening by a lot.

We now have AI agents that take multi-step actions without checking in. There are wearables that capture context all day. The soon-to-be-future of robotics is finally leaving the demo video. If your framework was already wobbling under a chatbot, it will not survive the next four things.

So here is a better question (maybe a different question).

"What does a person need in order to make good decisions about anything, including the machine?"

That question produces a list that holds still long enough to teach.

Knowledge, and yes, still knowledge

The most confident claim of the past two years is that AI has made knowing things optional. Folks will say, “Why memorize when you can retrieve?”

The research does not support this, and neither does anyone's actual experience using these tools well.

Daniel Willingham has spent two decades making the case that critical thinking is not a portable skill you install once and apply anywhere. It is domain specific. You can’t reason well about the causes of the Civil War without knowing a great deal about the Civil War. Your thinking runs on the content and context. If you remove the content, what remains is the sensation of thinking rather than the thing itself.

AI raises the stakes on this rather than lowering them. An AI model might hand a student a fluent, confident, beautifully organized, and entirely wrong paragraph about photosynthesis. The only defense is knowing enough about photosynthesis to catch it. Fluency must include accuracy, and a student who knows nothing cannot tell the difference, and worse, has no reason to suspect there is one.

This part of the conversation keeps getting skipped because it sounds weird in a post-google world to say kids still need to know things. But it is the precondition for every other item on the list.

Judgment about what is worth making

Here is what actually shifted. The cost of producing something dropped close to zero, which means the value moved upstream into deciding what to produce. I’ve been calling this discernment, but that is actually the pause before using, this is really good ol judgement.

Anyone can generate 40 campaign slogans in 11 seconds. Choosing the right one is a different act, and it is not a matter of preference. It requires knowing the audience, the history of what has been tried, the constraint you are actually working against, and the difference between clever and true.

As a teacher, I failed to teach this well because school rarely asks students to choose. I would hand them the prompt, the format, the length, the due date, and the rubric, then grade the execution. Execution is precisely the part that just got commoditized. Think of a student who has never selected a problem, defended the selection, and lived with the consequences of a bad selection has no practice at the only stage of the work that still belongs to them.

This is the agency argument that I will continue to share leading up to the release of my new book on the topic, and one that is not going away any time soon.

The willingness to stay inside difficulty

Robert Bjork's research on desirable difficulties points at something uncomfortable for anyone building efficient learning experiences. The conditions that make learning feel smooth are frequently the conditions that make it fail to stick. Struggle, spacing, retrieval feel worse and weirdly work better.

AI is the most effective friction removal tool ever built. We can point it at a hard problem and the hard part disappears like poof! That is genuinely useful when the friction is pointless, like formatting citations or cleaning a data set. But it destroys real learning that needs desirable difficulty.

So the human skill that matters is not grit as a personality trait. It is the ability to recognize which difficulty is the learning and which difficulty is just tedium, and to protect the first one on purpose. That is a judgment call students (and adults) have to make dozens of times a day now, usually alone, usually at 11 at night with an assignment due.

Almost nobody is teaching them how to make it. Maybe, at best, we are just monitoring whether they made it correctly.

What this looks like today

None of this requires a new initiative, but it does require a shift in what we assign value to.

Build knowledge deliberately and stop apologizing for it. Content and context is the fuel for learning.

Give students agency earlier in the process. Let them pick the problem, not just solve the assigned one, and make them defend the pick on why.

Ask students to evaluate AI output in a domain they actually know. This is the cheapest and most honest assessment available right now, because it is very hard to fake understanding while critiquing something.

Describe the friction we are looking for out loud. Tell students which part of this task is the point and which part is not, and then let them decide where the tool goes. They will get it wrong and that is the practice.

The part that does not change

Every version of the human skills list I have seen tries to answer a defensive question: What is left for us?

The reality is that a person who knows things, who can tell what is worth doing, and who can sit inside a hard problem without immediately outsourcing it, has leverage no matter what AI changes or improves next year. That was true before any of this and it will be true after whatever comes after agents.

The technology will keep moving, but the list should not have to.

We should spend more time looking at what skills and knowledge will continue to matter, instead of trying to keep up with what is changing at all times.