Taste: The PM Skill AI Cannot Commoditise

TL;DR

• AI commoditises execution: writing code, generating options, drafting copy. What remains scarce is the judgment to decide which option is right, for whom, and whether it should exist at all.
• Taste is three things: judgment under ambiguity, empathy at scale, and restraint. None of them can be prompted.
• Taste develops by shipping, watching things fail, and forming opinions. There is no shortcut.

---

Every AI-generated product looks the same. The same layout. The same copy cadence. The same feature set, built in the same order. When thousands of PMs use the same tools to build the same category of product, the output converges. What differentiates the good ones is human curation applied at every decision point: selecting one option from the set AI produces, and knowing why.

That capacity is taste. As building costs approach zero, taste becomes the primary source of product differentiation.

What taste actually is

Taste is not aesthetic preference. It's not knowing which font looks good. It's a compound skill with three components.

Judgment under ambiguity. Most product decisions don't have a right answer. They have options with different tradeoffs, incomplete information, and consequences that only reveal themselves after shipping. Judgment is the capacity to make a call anyway, with a reasoned position, and to be more right than wrong over time.

AI can surface options. It can list tradeoffs. It cannot weigh them. It has no stake in the outcome, no experience of what it felt like when the last similar call was wrong, no intuition built from watching users struggle with the consequence. That's what judgment requires.

Empathy at scale. Knowing what a user actually wants, not what they say they want, and not what the data suggests they click on. Users can't articulate their mental models. They can't tell you why something feels off. They can't anticipate what they'd prefer if the product worked differently.

Empathy comes from time spent watching users work, not from user research summaries. It's the capacity to hold someone else's mental model alongside your own and feel the friction they feel. AI can process user research. It cannot feel the friction.

Restraint. Knowing what to cut. When AI makes every feature cheap to build, the natural limit on scope disappears. The PM who can look at a finished product and remove half of it (knowing the remaining half is stronger) has a skill that is almost entirely human.

A product with four features used daily outperforms a product with forty features explored once. Every product team knows this. Very few act on it. Restraint is the capacity to act on it, especially when the cut features took real effort to build.

Why AI makes taste more valuable

The intuition runs backwards. Shouldn't AI make taste less valuable by helping anyone produce quality output?

No. AI raises the floor. It does not raise the ceiling.

Before AI, execution skill created differentiation. The team that wrote better code faster had an advantage. That advantage is gone. A working v1 now takes an afternoon. The new constraint is the quality of the decisions made before, during, and after building: what to build, for whom, in what form, and what to cut. Those decisions require taste.

There's a second effect. When AI generates the first draft of everything, the quality of curation becomes visible. A PM with taste produces products that feel considered. A PM without it produces products that feel generated. Users feel the difference even when they can't articulate it. This is what "AI products all feel the same" actually means: no human curation was applied at the critical decision points.

How taste develops

There is no shortcut. Taste is not a framework. It's not a course. It develops through accumulation.

Ship things and watch them fail. The feedback loop between decision and consequence is where judgment builds. A PM who has shipped ten features and watched four of them fail has developed intuitions that no amount of research can substitute. Those failures leave marks. The marks become the filter.

Study great products obsessively. Not just products in your category. Any product with a strong point of view: physical products, software, service experiences. Ask why specific decisions were made. What was cut. What tradeoffs were accepted. Reconstruct the reasoning behind choices that aren't immediately obvious.

Form opinions and defend them. Taste requires commitment. A PM who can see the argument for every position but won't commit to one has developed analysis paralysis, not taste. State a position. Be challenged. Update or hold. The discipline of defending and revising builds the muscle.

Absorb adjacent disciplines. Product taste draws from design, writing, psychology, and economics. A PM who understands typography has different instincts about visual hierarchy. A PM who has written seriously has different instincts about clarity. Adjacent disciplines add texture to judgment.

Taste in practice

Taste is operationalised through one question: is this the right thing, for these people, at this time, in this form? Not "does this work" and not "did we build what was specced."

The product builder competency describes the shift toward PMs who build alongside engineers. The competency model maps the skills that matter in the AI era. Taste is what makes those capabilities useful rather than merely fast. A builder without taste ships quickly to the wrong place. A builder with it ships to exactly where users need to be.