You Need Taste

AI makes it easier to produce a plausible spec, mockup, or prototype. It does less to settle which one deserves your team’s time.

That is where taste matters: the ability to judge whether an idea fits the problem, the product, and the people who will use it. More options make that judgment more valuable.

From execution to evaluation

Product managers still need to coordinate work and get software shipped. But when a tool can draft ten versions of a brief or interface, producing another option is rarely the next useful step.

Someone has to choose what to investigate, explain why it fits, and reject what adds complexity without helping the user. A coherent draft can conceal a weak idea. Reviewing it takes more than checking whether the usual sections are present.

What I mean by taste

Taste in product work is informed judgment about a specific context. Does this solve the actual problem? Does it fit what we are building? Will the user understand what to do?

I see it in three behaviors:

Recognizing fit. An experienced reviewer can connect a screen or proposal to the constraints behind it, then explain where it works or misses.
Removing unnecessary work. When a generated design adds screens, features, and explanatory copy, the right decision may be subtraction.
Noticing inconsistency. A flow can follow familiar patterns and still feel confused: the language promises help while the interaction pressures the user, or a simple task asks for too many decisions.

These judgments come from exposure, critique, and feedback. They deserve an explanation, not an appeal to personal instinct.

What the metrics miss

Metrics help compare the options a team chooses to test. They do not tell the team which possibilities it never considered.

The 2021 paper Investigating the Homogenization of Web Design examined how websites had grown visually more similar. Shared templates and frameworks were part of the context, well before today’s generative tools became common.

My concern is that teams asking models for familiar designs, then selecting only through short-term metrics, could reinforce that sameness. This is a risk to test, not a conclusion the older study establishes about generative AI.

Sometimes a standard pattern is exactly right. A checkout should not be surprising for the sake of distinction. Elsewhere, a less familiar choice may better express the product’s purpose or serve its users. Taste helps identify that possibility; evidence still has to test it.

Look beyond other software

I love photography. Art Wolfe, Annie Leibovitz, and Chase Jarvis are among the photographers whose work I return to. Studying their images makes me pay attention to composition, timing, and what they leave out.

Jarvis has written about learning through collaboration outside your usual field. That advice applies to product work. If every reference is another SaaS interface, your comparisons will be narrow.

Look at how a hotel handles an arrival, how a physical control communicates its purpose, or how a film introduces information without explaining everything at once. Study Bauhaus design and Swiss typography for their use of hierarchy and restraint.

The point is to notice a decision, understand why it works in that setting, and consider whether the principle transfers. Borrowing a visual style without understanding its purpose is just another kind of template use.

Put the judgment into the review

Use AI to explore options, then spend review time on the decisions that matter:

State the standard before choosing. Name the user problem, constraints, and qualities a good solution must have.
Keep a reference library with notes. Save examples of clear instructions, good pacing, or respectful error handling. Record why each example works.
Review tone and consequence. Ask whether the design reassures, patronizes, pressures, or confuses the user. Check those impressions with users.
Explain unusual choices. If an option sacrifices a short-term metric for a longer-term benefit, make that tradeoff explicit and decide how to evaluate it.

The explanation matters as much as the selection. It lets colleagues challenge the judgment and learn from the result.

Keep taste open to challenge

Models may become better at representing preferences. Some markets may reward consistency more than distinction. And what a team calls “taste” can be personal preference or bias dressed up as expertise.

None of that removes the need for judgment. It means the judgment should be specific, explainable, and open to correction. Broaden your references, make a choice, then look closely at what happens when people use it.