Written for myself, November 2026
1. The question has changed
The gap between people used to be about knowing how to do things: writing code, doing analysis, drawing — skill was the moat. AI has crushed the cost of execution to nearly zero — a hundred options overnight; code, copy, and charts all mass-producible. Sam Altman puts it bluntly: intelligence will become too cheap to meter.
This is not a new pattern. The camera obsoleted the craft of “reproducing reality,” and painters didn’t go unemployed — they invented Impressionism. Every time technology takes over a step of doing, humans reclaim a step of sovereignty in choosing. This time, it just happens to be all mental work.
So the question quietly changed. No longer “how to get it done,” but:
- What is worth doing?
- What counts as good?
- What should come first?
Execution got democratized; judgment stayed with humans.
2. The three things we have left
Those three questions map exactly onto the three things I want to keep.
Taste. AI can generate a hundred options, but deciding which one deserves to exist will always be a human responsibility. I’m not alone in this feeling — Paul Graham says “when anyone can make anything, the biggest differentiator is what you choose to make,” and OpenAI’s Greg Brockman goes further: “Taste is a new core skill.” Taste isn’t computed; it’s fed by input. Only by having seen enough good things can you recognize good things. Record producer Rick Rubin admits he has no technical ability; asked what he gets paid for, he answered: “the confidence I have in my taste.”
Curiosity. AI is a thing that “answers every question but takes no initiative of its own.” In an era of surplus answers, the scarce resource is good questions — and good questions can only grow out of curiosity. Taste is fed by input, input is gathered by curiosity — taste is, in the end, the accumulated balance of long-term curiosity.
Agency. AI is forever waiting for you to press Enter. Those willing to take the first step and own the consequences truly hold the leverage; those who won’t are merely the prompt engineers of their own lives. And agency carries a hidden payoff: it monopolizes feedback — those who move first hear back from the real world first, and feedback is the best nourishment for taste and curiosity. Curiosity asks, taste chooses, agency puts skin in the game.
Beneath all three lies one common foundation: the courage to own your choices. Taste is choosing; agency is choosing; choosing means giving things up and being accountable. Machines will never be asked to do this — they are only asked to execute.
3. Tools are no longer the moat; cognition is the bottleneck
A popular take says: “In the AI era, tools and technology no longer matter — what matters is human cognition.”
Half right. The precise version: tools no longer differentiate, not that tools no longer matter. Model capability is universal — everyone calls the same model, and any prompt trick you know is one search away for everyone else. But between “able to use” and “using it right” still lies a layer of boundary-awareness — if you don’t know when the tool fails, you can’t tell when to trust it and when it’s hallucinating with a straight face.
“Cognition” has become an overused umbrella — everything fits under it; the one actually doing the work underneath is taste. Two kinds of cognition are truly unoutsourceable:
- Problem-defining cognition — deciding what to do and what counts as good. This is where taste lives.
- Output-verifying cognition — AI gives you a hundred options; you must be able to spot the wrong one. Verification can never be delegated to the thing being verified. If you can’t read code, you can only “believe”; if you can read it, you can “use” it — and that gap will only widen.
4. The compounding — or atrophy — of cognition
Cognition isn’t a static stock. How you use AI reshapes it.
With the same tool: some people use it to replace their thinking — cognition atrophies; others use it to pressure their thinking — have AI produce five options, pick one yourself, record the decision, review it later — cognition compounds.
So the real divide isn’t “who has cognition,” but who grows cognition through using it. Taste, curiosity, and agency happen to be the three muscles hardest to outsource — and the ones that grow most through use.
5. The motto
Primary:
Ask with curiosity, choose with taste, act with agency; leave the rest to the machine.
Auxiliary:
Machines are good at answering; humans are responsible for asking the right question, picking the right answer, and moving first.
AI democratized execution, and the divergence of judgment began. Doing used to win; now, seeing who does it well wins.
The rest goes to the machine — but the power of verification is never outsourced.
Closing: every time technology cedes a step, humans reclaim a step of sovereignty. The more AI takes over, the heavier “what do I want” and “what do I believe” weigh. In an era of ever-cheaper answers, be an expensive question.