I watched Jason Liu talk about a line he tweeted a few days before the video. If you want taste, you have to eat.
AI keeps making more websites, apps, and images. Most people generate or scroll. Almost nobody curates.
His idea of taste has a few parts.
Can you tell when something is actually good. That is aesthetic judgment. Can you guess what other people will like. That is theory of mind. And the harder part: can you ship something that is not the safe average and still feel it is right.
People say they want taste. Then they "research" a competitor by watching the launch video and looking at screenshots. That is reading the menu. That is not eating.
Want to get good at building AI agents. Try a lot of agents. Looking only at your own work will not train you.
He tells a story from The Book of Tea.
Kobori Enshū came after Rikyū. Students told Enshū that everything he collected was so beautiful that anyone would admire it. Enshū said that was exactly why he was the lesser man. Rikyū dared to love things only he loved. Enshū was picking what most people already liked.
Guessing what the crowd likes is one skill. Having courage to love something strange on purpose is another.
The art school part is the one I kept thinking about.
His taste was ahead of his skill. He could make a piece, dislike it, and have no words for why. School gave him vocabulary. Bad composition. Flat depth. Wrong color. Weak form. When he could name the problem, he knew what to practice.
AI changes the order. Skill becomes cheap. You can make almost anything. Without the words, you still do bad work.
That is why senior engineers often get more from these tools than juniors. They talk about interfaces, layers of abstraction, API contracts, how the system is shaped. Those words are judgment. That is taste with another name.
Animation is the same. "Make it better" does almost nothing. "Think about sound. Think about easing curves." That is a different request. Food is the same. If you can name acid, sear, not enough salt, marinade time, you can fix dinner.
Eating is not enough. Close your eyes and name what is missing.
Walter Benjamin wrote about aura in The Work of Art in the Age of Mechanical Reproduction. Mechanical copies remove the here-and-now from a piece. A camera has less friction than a painting.
AI goes one step further. It gives you the end result without the hours of attention that used to take you there.
He talks about learning drums. Ask AI to write down a drum part, then try to play from that. Your ears still do the real work, or they do not. The finished notes on paper were never the point. Listening hard enough to make those notes was. Call and response. Tempo changes. Did I hear that right. Copying used to be the hard part. AI makes copying cheap, and you lose the friction that made the skill feel earned.
So where do I put the friction back, if I still want to learn with my body.
Video games do a quieter version. Mini-maps. Waypoints. Pathfinders. Point A, point B. Discovery gets smaller. You start caring about the destination and stop wandering.
People with good taste wander. They put clothes on in the store even when they are not buying. They feel shape, fall, form on their body. They look through books and keep what stays. Words help. Silhouette. Color blocking. Draping. Words without putting the jacket on are like knowing salt science without cooking.
When copying is easy, the hard part moves. Making gets cheaper. Noticing gets rarer. Curation too. Taking time to consume too.
Want better AI art. Look at more art. Want better AI music. Listen to more music. The model was not what was missing.
AI made the path from you to the thing you want to make much shorter. In that shortcut, paying attention became optional. Wandering too.
If you want taste, you have to eat.