TL;DR: Eight human capabilities matter more in 2026 than they did five years ago, and they’ll matter more in 2031 than they do now. The augmented humans who win the next decade are the ones who get sharper at them, whatever AI tools ship in between, because nobody can take those skills from you when the next tool arrives.
Almost every “AI skills you need” article you’ll read this year is a list of tools. ChatGPT, Claude, Perplexity, Midjourney, whatever shipped this quarter. The lists are accurate in the moment and obsolete six months later. The tools change. The model behind them changes. The vendor changes. The interface changes.
What doesn’t change, what hasn’t changed across forty years of watching technology arrive and reshape work, are the human capabilities the tools amplify.
The skills below are the ones that matter more every time a new tool ships. Get sharper at these and the next AI tool that arrives is a productivity gain. Don’t get sharper at these and the next tool exposes the gap.
The tool lists are a waste of a reader’s attention. They go stale in a season, and they train people to think knowing where a button lives counts as expertise. These eight skills are the ones I’d want anybody building, because they make every tool better and they’ll still be worth something in 2031. Pick one this week.
1. Asking better questions
AI is a question-answering machine. The quality of the question limits the quality of the answer. The professionals getting real productivity gains from AI can frame a question precisely. With the right context. The right constraints. The right specification of what a useful answer looks like.
Almost every ‘AI skills you need’ article you’ll read this year is just a list of tools.Share on X
The professionals getting nothing out of AI ask vague questions, get vague answers, and decide the tool is useless. I don’t buy that verdict. Hand the same question to a junior colleague and you’d get the same mush back, and nobody would blame the colleague.
Sharpen this. Notice when your prompts are vague. Notice when you give the AI a question you would have rejected if a junior colleague asked it that way. It’s the same skill that makes you a good manager, a good consultant, or a good researcher. AI has now made it the central skill of professional knowledge work.
2. Verifying with appropriate paranoia
From the podcast
Ben “The Automator” Christensen put the verification standard in one line on Leaders and Their Stories: How would you treat that stuff if it came from a person? He QAs every call on the voice assistants he builds, because a system with nobody checking it is a system waiting to embarrass you.
AI invents things. Citations that don’t exist. Statistics that came from nowhere. Quotes attributed to the wrong person. Legal cases that aren’t real. The mechanism is in AI Hallucination: A Survival Guide for People Who Publish Under Their Own Name, but the skill is independent of the mechanism.
Get sharper at noticing when you’re trusting an AI claim that you shouldn’t. Get sharper at the small voice in the back of your head that says “I should check this.” Get sharper at checking it instead of moving on because the work is more comfortable when you assume the AI is right.
The augmented humans publish work that holds up because they verified the specifics. The replaced ones publish work that falls apart because they didn’t.
3. Writing prose that sounds like a person
This is the skill AI is least equipped to do, and therefore the one that’s most valuable to maintain.
AI prose is the average of everything written on a subject. It hits the common arguments, names the common facts, lands the common conclusion. It almost never includes the one specific risky detail only someone who lived the work would know to include.
The professionals whose writing has a voice, who include the specific that nobody else would have thought to include, who take the risk of saying the thing that costs them something to say, are going to be valuable in 2031 in a way they aren’t yet. AI and the Curse of Shallowness walks through the mechanism.
The skill is the inverse of the mechanism. Practice having a voice. Practice including the specific. Practice saying what you think instead of what an average thinker on the topic would say.
4. Judging what’s worth doing yourself
Call it the skill behind the skill behind every effective use of AI. The question isn’t “should I use AI for this,” it’s “which part of this should I keep and which part should I hand off.” That decision requires judgment about what the part requires.
If the part is routine, bounded, low-stakes, and easy to review, give it to AI. If the part is judgment, voice, lived experience, or a binding commitment, keep it. The augmented humans I work with have an internal sense of which part of their work is which, and they don’t have to think hard about the routing.
The unaugmented either keep everything, which is slow, or hand everything to AI and end up with work that has nothing of them in it. The skill is the line, and the people who’ve taken the time to find where it falls in their own work draw it sharper.
5. Reading critically across long documents
AI generates long documents that drift. The drift is in AI Drift: The Failure Mode Nobody Talks About, but the response to drift is a human skill. Somebody has to read the long document end to end and notice when the conclusion isn’t supported by the body, when section 22 contradicts section 4, when the document started in one place and ended in another.
I’d take a colleague with one sharp, well-defended opinion over one who hands me a tidy summary of every side.
It’s a reading skill, and it’s harder than it sounds. Most professional reading right now is section-by-section, point-by-point, looking for individual errors. The skill that catches drift is whole-document reading, comparing the bookends, tracking the through-line, asking whether the document still answers the question it was supposed to. Sharpen this and you become the person who catches the problems with AI-generated long-form work before they ship. That’s a valuable role and almost nobody is doing it well.
6. Having strong opinions, weakly held
AI produces averages. The average opinion. The average framing. The average set of considerations. The average conclusion. If you bring an average opinion to the conversation, AI is going to outproduce you on the average, and your contribution becomes redundant. What AI can’t do is have a non-average opinion based on lived experience and the specific things you’ve seen that other people haven’t. The skill is having those opinions, defending them when they’re under-tested, updating them when they’re wrong, and bringing them to every conversation where you’re the human in the room.
Without strong opinions, you’re a redundant input to AI’s output. With them, you’re the differentiator the AI can’t replace. The skill is to have them and to hold them well. That means changing them when the evidence demands it and defending them when it doesn’t. I’d take a colleague with one sharp, well-defended opinion over one who hands me a tidy summary of every side. A machine can write the summary now. The opinion is the reason anybody keeps you in the room.
7. Knowing when AI is bullshitting you
AI sounds confident on everything. The confidence is uniform whether the model is right or wrong, whether the topic is well-trained or not, whether the question is answerable or not. The skill is reading the confidence signal at the right discount.
That’s harder than it sounds because confidence is one of the human signals we’re calibrated against. When a real expert sounds confident, they usually have reason to. When an AI sounds confident, the confidence is a stylistic artifact. The two feel the same on the page.
See the full Technology of Writing Hub for related guides. The skill is feeling the difference and not granting confidence the deference it would deserve from a human source.
Smooth AI output stops fooling people who develop this skill. People who don’t develop it ship work that was wrong and didn’t sound wrong, and they don’t find out until somebody else catches it.
This one scares me most for people who publish. Smooth prose lowers your guard, and the errors that slip through tend to be the ones that sounded surest. When a reader catches one, the name on the cover takes the blame.
8. Doing the work the machine can’t do
The meta-skill is this. The point of using AI well is to spend the freed time on the work that only a human can do. The skill is knowing what that work is and getting better at it.
For a writer, the freed time goes into the voice, the lived specifics, the moments that cost something to put on the page. For a doctor, it goes into the part of the patient relationship that no AI can have. For an engineer, it goes into the architectural decisions and the trade-offs that require taste. For every profession, there’s a part of the job that’s irreducibly human, and the augmented human is the one who has gotten dramatically better at that part because the routine has been offloaded.
Most people use AI to save time and then spend the saved time on more routine work or on nothing, instead of using it to get sharper. The augmented human is the one who put the freed hours into deepening. The replaced human is the one who put them into more shallow work or into vacancy.
I’d push people hardest on where the saved hours go, because saving time is worth nothing if they go into more busywork. The point of handing off the routine is to come back to the part of your job only you can do and get better at it, and too many people skip that step.
The 71-year-old memoir client I work with is the model. The 71-Year-Old Memoirist Who Uses AI Better Than You Do walks through what this looks like in practice. The saved Saturdays went into the writing. The writing got better. The skill is doing the same thing in your own work.
My take
Most of the AI skills people chase will be obsolete by the next release. The eight in this piece won’t, because none of them belong to a tool. If I had to pick one to start this week, it’d be where your saved hours go. Time saved and spent on more busywork is worth nothing. Hand off the routine, then put the hours into the part of your job only you can do, and get better at it every month.
Which AI skill should you start building this week?
Pick one of the eight. Spend half an hour identifying where you’re weakest at it. Spend the next two weeks practicing.
In six months, you’ll have noticeably leveled up on one of the human capabilities AI can’t reproduce. The tools will have changed twice in that time. The skill you built will still be there. The Birth of the Augmented Human covers the wider picture of what this looks like at scale, but the start is one skill, this week, your weakest one.
It frustrates me to think of people spending the next five years collecting tool tutorials while the skills that pay stay exactly where they were. The tools are the easy part, and they’ll keep changing on you. Pick your weakest skill of the eight, work on it for two weeks, and quit waiting for somebody to hand you a syllabus.
Those eight capabilities are what the augmented human sharpens. My book The Birth of the Augmented Human covers what goes right when you use AI properly.
