I used to think the smart people in any room were the ones who knew the most. I was wrong about which way the bet was running.
For most of my career, expertise was the asset. You spent a decade getting good at something, and that something compounded: the frameworks you carried, the patterns you recognized, the instincts that fired before you could articulate why. These were the things that made you valuable, and quietly, the things that made you you.
Then a class of machines arrived that can do a respectable version of all of that, in seconds, at the marginal cost of a cup of coffee (at least for now, until the VC funding burns through).
The conversation immediately became about which jobs go away. That is the wrong question. The more interesting one is what kind of human you have to become to work alongside something that can produce in minutes what used to take you weeks. The honest answer is that most people have not figured this out yet, including the ones loudly claiming they have.
This is a piece about the skill that quietly determines who is going to be useful in the next ten years. It is the skill of unlearning, and you cannot talk about it seriously without talking about machines.
The skill is not learning, it is unlearning
Learning is comfortable. It is additive. You collect frameworks, mental models, instincts, and you stack them like proof that you belong in the room. Every promotion you ever got was probably tied to something you had learned.
Unlearning is the opposite shape. It is subtractive. It asks you to look at the thing that made you successful and consider that it might be the thing slowing you down. Most people cannot do that, because somewhere along the way they stopped having opinions and started being them. That is the trap, and when your knowledge becomes your identity, every update to the world feels like a personal attack. The defensiveness is not a character flaw, it is structural. You built a self on top of a body of knowledge, and now the knowledge is moving or commoditized and the self does not want to.
This was always a problem, and AI just made it THE problem.
Why this matters more than ever
For most of human working history, the half life of an expertise was longer than a career. You learned a craft, you practiced it, you got better, you retired. The world moved, but it moved slowly enough that what you knew at thirty was still mostly useful at sixty.
That bargain has quietly broken. The thing you spent a decade mastering can be commodity in eighteen months. The framework that made your career can quietly turn into the framework that makes you legacy. Not because the underlying domain stopped mattering, but because a machine learned to do the mechanical parts of it cheaper and faster than you can.
More often than not, if you cannot let go of the part that has been automated, you go down with it.
This is what makes unlearning the skill of this decade and not just a nice idea. It used to be optional because the world was patient. The world is no longer patient.
Coexistence is not the same as adoption
There are three reflexes most people have when AI lands in their workflow, and I think all three are wrong.
The first is fear. The machine is here to take my job, so I will protect my territory by refusing to use it, by mocking it, by listing the things it cannot do yet. This buys you time and costs you the future. Every month you spend defending your old workflow is a month a peer is learning to use the new one against you.
The second is dismissal. It is just autocomplete, it hallucinates, it cannot really think. All true, and all beside the point. The argument is not whether the machine is intelligent, it is whether the machine is useful enough to reorganize how the work gets done. It already is.
The third, and the one I see more often in ambitious people, is uncritical adoption. They paste everything into the model, they publish what comes back, they mistake fluency for thinking. A few months later they look up and realize they have lost a skill they did not notice they were using.
None of these are coexistence. Coexistence is a specific stance, and it is harder than any of the three.
Coexistence means treating the machine as a collaborator with a strange shape of competence. It is excellent at the things you used to grind through, and unreliable at the things you used to do well without thinking. The shape of your judgment has to change to match. You have to know, in real time, when to lean in and when to pull back. That is not a setting, it is a practice.
The two failure modes
There is a way to use AI that quietly destroys you, and a way to refuse it that quietly destroys you, and they look like opposites but they end the same way.
The first is over delegation. You hand the machine more and more of the cognitive work, and at first this feels like a free upgrade. You ship faster, you write more, you synthesize across domains you barely understand. Then one day someone asks you a question that does not have a model on the other end, and you notice the muscle is gone. You have been a manager of outputs you cannot evaluate, performing competence you no longer possess. The cost shows up late, all at once, and it is hard to recover from.
The second is under delegation. You refuse to let the machine into the parts of your work where it would actually help, because those parts feel like your craft, and craft feels like identity. So you keep grinding through the mechanical eighty percent the machine could handle, and you have less time and less energy for the twenty percent only you can do. Your peers who delegated correctly are spending their time on judgment and taste, and the gap between you widens quietly.
Both failure modes feel virtuous from the inside: the first feels like progress, the second feels like principle, and neither is.
The actual practice is harder. You have to be willing to hand over the parts of your work that defined you, while staying sharp on the parts that are now the only thing you bring.
What actually compounds
If knowledge is no longer the moat, what is?
The honest list is shorter than people want it to be: judgment, taste, the ability to ask a precise question, the ability to recognize when an answer is plausible but wrong, reading the room and knowing what to say to whom, and the capacity to hold a problem in your head long enough to know what good looks like before the machine renders ten versions of it.
These are not skills you learn from a course. They come from doing the work, and they only stay sharp if you keep doing the parts of the work that demand them, which is exactly why over delegation is dangerous. The machine will happily take the reps that were keeping your judgment alive.
There is also a quieter thing that compounds, which nobody quite knows what to call. It is something like: knowing what you actually want. The machine is a wish granting engine with a flat affect. It will give you what you ask for, faithfully and instantly, and if you do not know what you want, it will produce a high resolution version of nothing. The people who get the most out of these tools are the ones who have done the slow, unglamorous work of figuring out what they are actually trying to make. That work is more valuable now than it has ever been, and almost no one is teaching it.
The practice
The practice is not complicated, it is just uncomfortable.
Once a quarter, sit with one belief that has shaped how you operate, and ask whether you still believe it because it is true, or because you have not bothered to check in a while. Notice which ideas you defend hardest, because those are usually the ones closest to dying.
Once a month, take a piece of work the machine could now do for you, and let it. Watch closely. Notice where it is better than you, and accept that. Notice where it is worse, and get curious about why. The gap between the two is your real job description for the next five years.
Once a week, do something the machine cannot help you with: a conversation you have to actually have, a decision you have to actually make, a piece of writing where you sit with the discomfort of not knowing what you think before reaching for a model. This is how you keep your judgment from atrophying while the surface area of your work shrinks.
When you find a belief or a workflow that no longer holds, do not perform the update, just quietly change. Let the new behavior run for months before you announce a new philosophy. Most public unlearning is theatre, and real unlearning is quiet.
You are not trying to know more. You are trying to be less attached to what you currently know, and more honest about what you actually bring to a workflow that now includes a machine.
The skill of the decade
The next ten years will not reward the most informed people, because there is no such category anymore. A teenager with a good prompt has access to more synthesized information than a senior partner had a decade ago.
What it will reward is something rarer: people who can hold conviction loosely, people who can drop a framework cleanly when it stops working, people who can stand next to a machine that is faster and more fluent than they are and still know what to do with their own attention, people who have done enough internal work to know what they actually want, so that the speed of these tools amplifies them instead of diluting them.
That is the work. Learn, yes, but also unlearn, and learn to coexist with something that is going to keep getting better at the easy parts of your job, so that you can keep getting better at the parts that were always going to be hard.
The people who refuse this work will not lose suddenly. They will lose quietly, over years, the way most people lose. They will find smaller and smaller rooms where their old expertise still gets respect, and they will stay there for as long as the room remains open.
You do not have to be one of them. Be honest about what you know that is no longer worth knowing. Be honest about what you have been avoiding because the machine has made you face how much of your craft was actually mechanical. Then protect what only you can do, because increasingly that is the whole job.
The rest, more and more, is the machine’s.