All notes

What taste tools miss about taste

Taste tools are having a moment. Half a dozen startups shipped taste infrastructure in the last month — profiles that tell your AI agent what good looks like. I have written about why this category matters and what it means for solo founders. The momentum is real.

But here is what nobody is saying: encoding your preferences is the easy part. The hard part is applying those preferences consistently across every surface of your product, every edge case, every empty state, every day. Taste tools capture what you like. They cannot capture your willingness to care about the parts nobody sees.

On this page

Preferences are not taste

Taste tools capture your preferences — the colors you like, the spacing you prefer, the typography you gravitate toward. They turn these into a profile your agent reads before generating. That is genuinely useful. It saves you from repeating yourself and it prevents the worst generic defaults.

But preferences are not taste. Taste is what happens when your preferences collide with reality. It is knowing that a beautiful color palette is wrong for this specific page because the context demands something quieter. It is overriding your favorite font because the product needs to feel utilitarian, not expressive. Paul Graham made this case better than anyone — true taste is about knowing what to leave out, not knowing what to put in.

Taste tools capture the putting in. The leaving out is still on you.

The last mile is where products win

Every product has surfaces that AI will never generate well. The empty state when a user has no data. The error message when something breaks. The loading state that should feel like progress, not waiting. The hover interaction that should communicate without demanding attention.

These are taste decisions. They happen at the edges. And they are where users form their real impression of quality — not from the hero section that looks beautiful in the screenshot, but from the moment something goes wrong and the product still feels intentional.

The Nielsen Norman Group has been making the argument about consistency for decades: users trust products when they behave predictably. Predictability at the edges is what separates a product that feels crafted from one that feels assembled. Taste tools cannot generate that predictability because they only see the surfaces you tell them about. They do not know what you have not shown them yet.

Why taste tools struggle with edges

The problem is structural. Taste profiles are trained on examples — screenshots, reference UIs, design systems. They learn patterns from what exists. The empty state you need to design? It exists nowhere in the training data because most products treat it as an afterthought. The error message that should feel like the product is on your side? Most error messages read like blame. The taste profile has no reference for "good" here because there is so little of it to learn from.

This is why the last mile cannot be automated. It requires the kind of judgment that only comes from having used your own product, felt its friction, and decided that the edges matter as much as the center.

The real practice

The solo founders who ship products that feel intentional do not have better taste tools than everyone else. They have a practice. They review empty states before they ship. They write error copy that sounds like a person wrote it. They test loading sequences and ask themselves: does this feel fast or does it feel rushed?

This practice is not glamorous. It does not scale. It takes time that feels like it should be spent on something more important. But it is the difference between a product people respect and a product people tolerate.

If you are a solo founder building with AI, by all means use the taste tools. They will save you from terrible defaults and give you a consistent baseline. Just do not confuse having a taste profile with having taste. The profile encodes your preferences. Taste is what you do when those preferences are tested — by complexity, by ambiguity, by edge cases. And nobody can generate that but you.

What the tools are actually good for

To be fair: taste tools are genuinely useful for specific things. They eliminate decision overhead on routine styling choices. They keep color and spacing consistent across screens built days apart. They provide a reference your agent can consult instead of guessing. For solo founders stretched across a dozen responsibilities, that baseline consistency is valuable.

The mistake is treating taste infrastructure as a substitute for taste judgment. The baseline matters. The last mile matters more. Use the tools for what they are good at. Then do the work they cannot.

Frequently asked questions

  • No. Taste tools are great at encoding visual preferences — colors, spacing, typography — but they cannot judge whether a feature is necessary, whether an interaction feels right, or whether the product communicates its purpose clearly. Those judgments require context, empathy, and the willingness to say "this is not good enough" even when the tool thinks it is.

About the author

mosh

mosh is a product designer and design engineer working with design systems, LLM-powered prototypes, agent-safe interfaces, production UI, and automated workflows.

Keep reading