When Intelligence Becomes Cheap, Taste Gets Expensive

I've argued elsewhere on this site that answers are being commoditised, and that good questions are the scarce resource of this era. I want to push that further, because I don't think it goes far enough. It isn't only answers that are getting cheap. It's competence itself.

Consider what a capable model now does before breakfast. It writes a serviceable strategy memo, drafts clean code, produces prose that clears the bar of "professionally acceptable," summarises the research, generates the deck, proposes the plan. None of it is genius. All of it is competent - the exact tier of output that, until about five minutes ago, took a trained human a career to reach and a salary to retain. We built entire professions on the scarcity of competence. That scarcity is evaporating, and it's not coming back.

Here's the thing nobody quite says out loud: when competence becomes abundant, competence stops being valuable. This is just supply and demand wearing a lab coat. The market never pays for what's abundant, however hard it once was to acquire - it pays for what's scarce. So the interesting question isn't "what can the machine do?" It's "what got more scarce the instant the machine could do everything?" And the answer, I've become convinced, is taste.

By taste I don't mean aesthetics, or not only. I mean the whole cluster of judgement that the machine conspicuously lacks: knowing which of ten competent options is the right one for this moment, this audience, this constraint. Knowing what to leave out. Knowing that the technically correct answer is the wrong answer here because it misreads the room. Knowing which question was worth asking in the first place, and which brilliant analysis is brilliantly answering something that doesn't matter. A model can generate a hundred competent strategies. It cannot tell you which one your specific, weird, path-dependent situation actually needs - because that requires a point of view, and a point of view is precisely the thing that can't be averaged out of a training set.

Watch how the value migrates. When any founder can generate a plausible business plan on demand, the plan is worth nothing and the judgement to know which plan is alive is worth everything. When anyone can produce competent prose, the market for competent prose collapses and the premium on a genuine voice - a way of seeing that's recognisably one person's - goes vertical. When code writes itself, the scarce thing isn't the code; it's the architectural taste to know what to build and the ruthlessness to know what not to. In every domain the machine touches, value doesn't disappear. It evacuates the commodity and pools in the judgement layer above it - in selection, framing, curation, and the nerve to commit to one answer among many equally competent ones.

This is why the "collector of interesting questions" idea matters more than it might first appear. In a world of infinite competent answers, the person who can frame - who can look at a churning mess of machine-generated possibility and say this one, because of this, for them, now - is doing the only work left that pays a premium. Framing is taste applied to abundance. It's the human standing at the output end of an infinite competence machine, doing the one thing the machine can't: caring, specifically, about the right things.

I'll be honest about the uncomfortable edge of this, because pretending otherwise would be exactly the kind of tasteless move I'm warning against. Taste is far harder to teach than competence, and far harder to fake. You could always grind your way to competence - reps, study, time. Taste resists the grind; it comes from exposure, judgement, a thousand small acts of caring what's good and noticing why, and it stubbornly refuses to scale the way skill does. Which means this shift may be less democratising than the optimists hope. When competence was the moat, effort could cross it. When taste is the moat, the crossing gets stranger and less fair, and we should say so plainly rather than sell a fairy tale.

But here's where I land, and it's not where I expected to. For a long time we told people the safe path was to be smart, capable, competent - to be a reliable producer of correct answers. The machines just repriced that advice to roughly zero. What's left, what's actually scarce, what's suddenly worth a fortune, is the least mechanical thing about us: the capacity to look at everything that's possible and have a point of view about what's good. We spent a century optimising humans to think like machines. The machines arrived, did it better, and handed us back the one assignment we'd been neglecting the whole time - to have taste, and the nerve to trust it.

That, it turns out, was the human part all along. We just couldn't afford to notice until the competent part got cheap.