Why Taste Matters More in the AI Age

Knowledge used to be the bottleneck. If you didn't know how databases worked, or how to structure a frontend, you couldn't build a good website. AI has mostly dissolved that bottleneck — facts, patterns, and even working code are now cheap and available to anyone who asks.

But cheap knowledge has exposed a different scarcity: judgment.

Building anything complex means breaking a large decision into smaller ones — frontend, backend, data layer, performance, polish — and making a "good enough" choice at each one, then trusting that these choices combine into something coherent. Herbert Simon described this as bounded rationality: we satisfice our way through sub-decisions because no one can optimize a complex system all at once. Knowing how to break a problem into the right sub-decisions, and seeing how they interact, is a real skill. Call it decomposition skill. AI is increasingly good at this part, because it's mostly about facts — what causes latency, what scales, what breaks.

But Simon made a second distinction that matters more here: every decision rests on both factual premises and value premises. Facts can be checked against the world. Values cannot — they're statements about what's desirable, and two people can agree on every fact and still disagree on the right answer because they're optimizing for different things.

This is where taste lives. Taste isn't knowing what's true about a system — it's knowing what's good, at every one of the hundreds of small decisions a build requires. Spacing that feels right. An API shape that will age well. The difference between a website that works and one that feels considered. None of this is verifiable the way a fact is. It's a value premise, applied consistently, thousands of times.

That's also why taste resists being prompted into existence. You can hand someone a fact. You can't as easily hand them a value premise — it's absorbed through practice, through making the call wrong enough times to recognize when it's right. It shows up as a feel before it can be stated as a rule.

So the shift AI has caused isn't that skill stopped mattering. It's that the kind of skill that's scarce has changed. When execution was expensive, just getting something to work was the achievement. Now that execution is nearly free, anyone can generate a hundred plausible versions of anything. What's scarce is the discriminating eye that knows which one is actually good — and the only way to grow that eye is the slow way: making the sub-decisions yourself, watching what improves, and letting your sense of "good" get sharper over time.

AI has made knowledge abundant. It has made taste the moat.