AI made ability free. It didn't make judgment free. That's the core claim in Jason Liu's (@jxnlco) essay "If You Want Taste, You're Gonna Have to Eat" — a piece that resurfaced widely on X and is worth reading in full, not just as a headline. It's not a productivity essay or a tool review. It's an argument about what's actually scarce now that generation is cheap, and it lands squarely on explainx.ai's own territory: what should someone actually spend time learning when an AI can already do the mechanical part.
This connects directly to a thread explainx.ai has tracked all year — from AI-driven de-skilling in developers to Lars Faye's argument that AI coding prevents expertise to whether AI is out-thinking mathematicians or just out-remembering them. Liu's essay gives that whole conversation a name: taste.
TL;DR — the essay's core argument
| Claim | What it means |
|---|---|
| Ability used to be the bottleneck | Before AI, most people's taste exceeded their ability to execute it — Liu describes this from art school |
| AI inverted that | Ability is now roughly free; for most people, ability now exceeds taste |
| Taste ≠ consensus | Taste is choosing to deviate from the safe average, not matching what's already popular |
| AI gives you "the chart," not "the ear" | It hands you the output while skipping the attention that would let you judge the output |
| The fix is "eating slowly" | Engage with the whole thing (the album, not the single) and ask why you like what you like |
| Taste is learned through relationships with makers | Know the designer, the drummer, the producer — not just the object |
| Money is a vote, not just a purchase | Liu frames spending as choosing which decisions you want more of in the world |
| Reference point | The Book of Tea's Rikyū vs. Kobori Enshū story — taste is courage, not curation of what's already liked |
The argument, unpacked
Liu opens with something concrete and personal: personal style. Clothing, he argues, is the one form of taste you can't keep private — every other preference stays hidden until you choose to share it, but the moment you step outside, people see your taste before they hear a word you say. He uses that as the entry point for a bigger claim: as AI lets more people create things, the real skill gap is taste itself — the trained ability to sense what will actually resonate, paired with the willingness to deviate from the safe average instead of regressing to it.
That claim inverts his own art-school experience. Back then, his taste exceeded his ability — he could see what good work looked like and couldn't yet make it. AI flips that: ability is now basically free, so for most people, ability now exceeds taste. You can generate almost anything on request. You often don't know what's actually worth generating.
The essay's sharpest passage is a metaphor about transcription. Learning a music solo by ear means rewinding the same four bars repeatedly, forcing deep, sustained listening — the process is how a trained ear gets built. Liu's line for what AI does instead: "it gives you the chart, not the ear." That's not a complaint about AI being bad at music. It's a claim that AI strips the aura from learning itself — the felt distance that used to make a skill feel earned, and that distance is exactly where taste used to get built.
"Eating slowly": the practical method
Liu's prescription is less abstract than it sounds. He calls it "eating" — engaging directly with the thing itself, not a summary or a recommendation of it:
- Consume the whole work, not the algorithmic fragment. Listen to a full album on record instead of the short clip an algorithm has trimmed a song down to. Try on the unfamiliar jacket in the store instead of scrolling product photos.
- Ask why, specifically, you like what you like — until you have real vocabulary for it. Liu describes working out his own hierarchy this way: materials and draping first, structure and form second, color last, and for color, just four choices — gray, blue, black, red.
- Find the human behind the object. Who played drums on the record? Who designed the piece? What was happening in their life when they made it? Liu argues most consumption today has collapsed the maker out of the picture entirely — people recognize a trend without ever learning who made the decisions behind it.
- Treat spending as a vote. Money spent isn't just a transaction — it's a vote for a particular way of making things, a signal you want more of those decisions in the world, not fewer.
This is the part most directly useful for anyone learning to build with AI, not just anyone dressing themselves: the method for developing taste hasn't changed just because the tool changed. You still build it by paying slow, specific attention — to real codebases, real design decisions, real shipped products — not by asking a model to summarize the pattern for you.
The Rikyū story — why taste isn't consensus
Liu closes with a story from the Book of Tea. A student told tea master Kobori Enshū — who came after the legendary Rikyū — that his taste must be even better than Rikyū's, since everyone agreed everything Enshū collected was beautiful, while only Rikyū himself could appreciate his own collection. Enshū's answer: that was exactly why he was the lesser man. He had simply been picking what most people would already like, while Rikyū had the courage to love things only he could see.
That's a direct rebuttal to the instinct to treat "what performs well" or "what the model rated highest" as taste. Curation that only selects consensus favorites is closer to market research than to judgment. In Liu's framing, curiosity has to come first — without it you never broaden what you're paying attention to — and courage comes after: the willingness to actually commit to what resonates with you specifically, even when nobody else is buying it, listening to it, or shipping it that way.
What this means if you build with AI
Strip the essay of its fashion and music examples and the claim generalizes cleanly to anyone using AI agents or generation tools day to day:
- If an AI can generate five working versions of a feature, a landing page, or a business plan, the differentiator is which one you ship — not whether you could produce one at all. That's a taste judgment, and it's trainable the same way Liu describes: by looking closely at a lot of real, shipped examples and building vocabulary for why some work and some don't.
- Prompting well is not the same as having taste. A well-crafted prompt gets you a competent output faster. It does not substitute for having internalized what "good" looks like in the specific domain you're working in — that still comes from slow exposure, not faster generation.
- This is an argument for depth over volume in how you learn. It's a reason to actually read the source material a model summarizes for you sometimes, actually trace through a codebase instead of only reading the AI's explanation of it, actually build something by hand occasionally even when a tool could do it faster — not out of nostalgia, but because that's where the judgment gets built that later lets you evaluate AI output critically instead of accepting whatever it hands you.
What people are still asking
- "Isn't this just 'learn the fundamentals,' rebranded?" Partially, but the essay's sharper point is about attention, not curriculum. "Learn the fundamentals" is a content claim; Liu's claim is about process — that the slow, repeated noticing itself is what builds judgment, independent of which fundamentals you're studying.
- "Doesn't this argue against using AI at all?" No — Liu doesn't call for avoiding AI tools. He calls for not letting AI's ability to skip the noticing process quietly erode the taste that used to come from doing things the slow way, in whatever domain you actually care about developing judgment in.
- "How do I know if I have taste or I'm just being contrarian?" The essay doesn't fully resolve this, and it's a fair challenge to the framing — Liu's answer is closer to "courage to love what resonates with you" than a test you can run, which makes it more of a personal practice than a checklist.
Related on explainx.ai
- AI-Driven De-skilling in Developers — the practitioner-side version of the same worry
- Lars Faye: AI Coding Will Prevent Expertise — the studies behind whether AI erodes hands-on skill
- Is AI Out-Thinking Mathematicians, or Just Out-Remembering Them? — a parallel case for judgment vs. raw recall
- geohot: I Love LLMs, I Hate Hype — another opinionated read on where AI's real value sits
- "Programming Is Low-Intelligence Work" — kache's X Debate Explained — a companion debate about what skill AI actually replaces
- How to Survive the AI Apocalypse: A Practical Guide — what skills hold up regardless of how far generation gets
Primary source: Jason Liu — "If You Want Taste, You're Gonna Have to Eat" (published on X).
This piece summarizes and responds to Jason Liu's essay; quotations are limited excerpts used for commentary. Read the full essay at the source for the complete argument. Follow @explainx_ai for updates.