Taste is not snobbery. It's the engineering ability to look at a thing — your thing, AI's thing, anyone's thing — and say this is good and that is not, and here's why. When AI can make anything in seconds, this is the only moat left.
People hear "taste" and think it means having opinions about wine, or fonts, or which Hayao Miyazaki film is the best. That's a small piece of it.
Taste is the trained ability to tell good from bad — and to tell why. It's the muscle that says: "this paragraph is fine but the third sentence is a cliché. Cut it." It's the muscle that says: "my agent's output is technically correct but it sounds like every other agent — fix that."
It is, at its core, judgment with reasons. Not "I like it" but "here's the criterion I'm holding it against, and here's how it scores."
In a world where AI can generate a polished anything in two seconds, the limiting factor is no longer making things. It's knowing which version is the one you'd put your name on. That decision is taste. There is no automatic version of it.
Eric De Castro wrote a sentence in 2026 that is going to age very well: "When the cost of creation drops to zero, taste is the only moat."
Think about what's happening to the internet. Anyone can generate a passable essay. Anyone can produce a slick image. Anyone can write code that runs. The volume of technically fine output is now infinite. So what gets attention? What gets remembered? What's worth shipping?
Whatever someone with taste decided was worth keeping.
Here's the trap, though: AI output is just good enough to fool kids without a trained eye. It's polished. It's grammatically clean. It cites things. It would have gotten an A- in 2023. To a sharp 13-year-old in 2026, it's gray slop — generic, voiceless, often subtly wrong, almost always boring. The difference between those two reactions is the difference between someone whose taste atrophied and someone whose taste grew.
This is why Harness Studio (the fourth academy on this site) exists at all. Taste isn't just a personal trait. It can be turned into a system — a rubric, a judge, a process — that catches AI when AI is being mediocre. That's the most leveraged thing any builder your age can learn.
Ren is 14. She asks Claude for a 200-word reflection on a poem. Claude returns a competent paragraph. She reads it and says:
By draft 4 the piece sounds like Ren. AI was a fast first pass; Ren was the editor with the actual point of view.
Same age, same first draft from Claude. The kid reads it, thinks "yeah, sounds smart", ships it.
The piece is fine. It's also indistinguishable from the 4 million other AI-generated reflections produced this year. Nothing of the writer is in it. Worse — week by week, the writer's own ear gets duller, because they keep accepting AI-cadence as their own.
What taste does: it turns AI from an end into a beginning. The first draft becomes raw material. The shipped version is yours.
Taste is the slowest of the three to grow, because it requires exposure plus reflection plus rejection. There is no shortcut. But there are three things that genuinely move the needle:
1. Become a curator, not a consumer. Algorithms reward what's popular. Taste rewards what's good. The best move at age 10 or 11 is to start a personal collection — a notes file, a folder, a Notion — of things you've decided are good, with one sentence on each saying why. That "why" is where taste lives. After 100 entries you'll have a real point of view.
2. Throw something away. Once a month, take something you made that you were proud of, look at it again, and notice three things you'd change. If you can't, your eye hasn't grown yet — that's information too. The act of rejecting your own past work is the single most reliable taste-builder.
3. Write the rubric. When AI gives you something, before you read it, write down three criteria you're going to grade it against. Then grade it. The act of articulating the criteria — out loud, in writing — is the move that converts vague intuition into actual judgment. This is exactly what Harness Studio teaches you to do.