In the AI Era, Taste Is an Amplifier

🌏 中文版

Once tools drive execution costs extremely low, differences do not shrink. They expand.

Give AI to someone with taste and they can realize ideas that previously exceeded their capacity. Give the same tools to someone without taste and they merely produce mediocrity faster and at greater volume. AI is an amplifier, not a generator: it magnifies what you already have. It does not conjure what is missing.

Seeing a good design and immediately thinking, "I'll use AI to make something like that," reveals your position. A person with taste asks: Why is this good? What problem does it solve? What can I learn from it? Can I make something better?

The question, then, is whether taste can be trained.

Taste Is Not a Gift

Japanese creative director Manabu Mizuno, who designed Kumamon, defines taste as knowing what is ordinary. Broad exposure establishes a baseline; taste is the ability to depart from it. It is accumulated knowledge, not talent.

Paul Graham's 2002 essay Taste for Makers listed fourteen properties of good design: simplicity, solving the right problem, and looking easy despite being difficult. In 2021, Is There Such a Thing as Good Taste? argued by contradiction: if taste does not exist, good art does not exist and artists cannot be good at their work—an obviously false conclusion.

Manager Today makes the idea more operational: taste combines pattern recognition—exposure that builds an ordinary-versus-excellent baseline—with creating and defending criteria. Explain why, accept objections, and update the standard. Taste is not preference itself. It is sensitivity to quality differences and a willingness to revise judgment.

If you are willing to be contradicted, your taste improves. If you are not, you are performing taste.

31 Dimensions

Taste is not one faculty but a combination of judgments. I compared eleven frameworks from Paul Graham, Shreyas Doshi, Julie Zhuo, Logan Lincoln, NN/g, Intercom, and PostHog; design psychology from Don Norman and Christopher Alexander; YC's startup judgment from Kevin Hale, Michael Seibel, Dalton Caldwell, Garry Tan, and Peter Thiel; Steve Jobs and Jony Ive's design philosophy; and marketing theory from Rory Sutherland, Byron Sharp, and April Dunford. The result is 31 independently trainable dimensions in eight layers.

No essay uses all of them. Each selects the two or three most relevant dimensions. This table is a scanning checklist, not an exam.

Fundamentals: Understanding the Thing

Craft examines construction, empathy the experience, and domain knowledge the prerequisite. Intercom's Paul Adams states it plainly: taste is domain-specific; there is no universal good judgment.

Instinct differs from craft: polished but lifeless and rough but magnetic are different cases. Care also differs from craft: Jobs would reject a beautiful surface hiding chaos beneath it. Ive argued that what people do when nobody sees proves what they truly care about. Delight unfolds over time: the iMac's color could make entering the office pleasant every day, not only during unboxing.

People and Markets: Understanding Humans

Someone who sees only craft can be a critic. Someone who also reads human nature and markets can build products. Ben Thompson's Aggregation Theory is the cleanest framework for internet market structure. Signal perception comes from the third principle in Apple's 1977 Marketing Philosophy, Impute: people judge by appearance, so presentation shapes perception.

Consequences: Looking Farther

These tools point in six directions: systems outward, second order forward, inversion backward from the end, first principles downward, superlinearity toward curve shape, and time experience across unfolding use. In Farnam Street's collection of Munger models, inversion is among the best ways to kill bad bets. Gary Klein's HBR premortem—assume failure and work backward—improved the ability to identify future causes of failure by 30%. Paul Graham's Superlinear Returns notes that people think linearly although the most important results are superlinear.

Superstructure: Taste About Taste

Restraint asks whether you can choose not to build; culture whether you can avoid copying; creativity whether you can do more than judge; coherence whether the whole exceeds its parts; calibration whether you can face the quality of your own judgment. Julie Zhuo's The Tax of New says every new feature carries a hidden tax, and removing a failed feature is harder than launching it. Coherence follows Christopher Alexander's Quality Without a Name: if removing a part does not weaken the whole, that part did not belong.

The Founder's View: YC Heuristics

These come from Paul Graham's essays and YC partners' frameworks. Schlep Blindness: Stripe succeeded because others avoided payments infrastructure. Black Swan Farming: the best startup ideas initially look bad. Do Things That Don't Scale: Airbnb knocked on doors. Bus Ticket Theory: great work comes from unstoppable obsession. How to Think for Yourself: are you independent or conventional?

Marketing and Perception: How It Gets Seen

Rory Sutherland's central argument in Alchemy is that 10% of the cost of changing reality can sometimes achieve 100% of the perceptual effect. Uber's map does not shorten a wait; it shortens the anxiety of waiting. Byron Sharp's How Brands Grow argues that growth comes from mental availability—the quality and quantity of memory structures—rather than differentiation. April Dunford's Obviously Awesome shows that the same product placed in a different category triggers different expectations. Positioning is the first marketing step, not the last.

The AI Era: A New Blind Spot

The model proposes, frames, recommends, and drafts; the human approves without contributing independent thought. If you cannot defend the decision without the model's language, you have not finished thinking.

How This Series Works

Each essay selects one object—a product, design decision, technical choice, or market phenomenon—and examines it through two or three dimensions. The structure stays fixed:

1. What it is: the object and the problem it solves.
2. My judgment: thesis first; state the conclusion.
3. Analysis: support the judgment through two or three named dimensions and concrete scenes or decisions.
4. Extract the principle: one sentence to carry away.
5. Self-calibration: return after three months and ask whether the judgment was right.

The fifth step matters most. Taste grows not from making judgments but from verifying them later. Annie Duke explains that a good outcome is not the same as a good decision. Separating result quality from decision quality requires honest calibration, not more confidence.

This is therefore not a review series. It is a public, falsifiable training log for judgment.

Every essay practices the same thing: see an object, form a clear judgment, state it, and wait for reality to calibrate it.

References

Taste methodology

• Paul Graham — Taste for Makers (2002)
• Paul Graham — Is There Such a Thing as Good Taste? (2021)
• Dieter Rams — Ten Principles for Good Design
• Logan Lincoln — Product Taste Handbook
• Logan Lincoln — Taste Is the Last Skill AI Can't Commoditise
• Frank Harris — The Ingredients of Taste
• Kristen Berman — Product Teardowns
• PostHog — Good Taste Makes Great Products

Markets and consequences

• Ben Thompson — Aggregation Theory (2015)
• Farnam Street — Inversion
• Farnam Street — Second-Order Thinking
• Gary Klein — Performing a Project Premortem (HBR, 2007)
• Julie Zhuo — The Tax of New
• Annie Duke — Don't Be a Resulter
• Paul Graham — Superlinear Returns

YC startup judgment

• Paul Graham — Schlep Blindness
• Paul Graham — Black Swan Farming
• Paul Graham — Do Things That Don't Scale
• Paul Graham — The Bus Ticket Theory of Genius
• Paul Graham — Earnestness
• Paul Graham — How to Think for Yourself
• Kevin Hale — How to Evaluate Startup Ideas (YC)
• Michael Seibel — The Real Product Market Fit
• Dalton Caldwell — Tarpit Ideas: The Sequel
• Garry Tan — YC in the Age of AI
• Peter Thiel — CS183: Startup
• Sam Altman — How to Start a Startup
• Gustaf Alströmer — How to Get Users and Grow
• Patrick Collison — Craft & The Future of Payments

Marketing and perception

• Rory Sutherland — Alchemy
• Byron Sharp — How Brands Grow
• April Dunford — Obviously Awesome

Design psychology

• Don Norman — The Design of Everyday Things
• Christopher Alexander — A Pattern Language