Essay ·
AI lowers the bar; the ceiling stays ours
The most useful sentence in my operating vocabulary this year has been a short one: AI lowers the bar; the ceiling stays ours. The bar is the cost of producing a plausible version of design work. A product manager with an AI tool now reaches an 80% mock-up in an afternoon, and I think that is genuinely good: it is the paved path working as intended. But watch what happens to the remaining 20%. It does not get cheaper. It gets more valuable, because it is now the only part that differentiates anything, and it is defended by the one faculty AI has not commoditized: knowing what is wrong with the output.
That faculty has an old, slightly embarrassing name. Taste. We are embarrassed by it because it sounds aristocratic and unmeasurable, a mystique we rightly distrust. So let me try to rescue the word by making it boring.
Taste is a judgment, not an aura
Taste, operationally, is fast and accurate defect detection against a standard the org has agreed on. The designer who looks at the fluent AI mock-up and says "the hierarchy is lying about what matters, the empty state is an afterthought, and this flow will humiliate a first-time user" is not performing sensibility. They are running a trained discriminator. The generator just got cheap; discriminators did not. Every economics of the next decade of design work follows from that asymmetry.
And crucially, prompting is not the skill. A prompt can only ask for what you can already specify, and the whole value of taste is noticing the defect you did not know to forbid. The industry's rush to teach designers prompting has it backwards: the scarce skill is not asking better, it is judging better.
Taste is an organizational property
Here is where I part company with the mystique completely: at organizational scale, taste is not a person, it is infrastructure. A quality bar nobody wrote down is a rumor. In my org the ceiling is held by mechanisms, not heroes: a design system that encodes the thousand small decisions so they cannot regress, a validation gate that puts evidence between fluency and shipping, crit culture that makes articulating "why is this wrong" a daily rep rather than a performance, and exemplars, actual named examples of the standard, because taste transfers through examples faster than through principles. If your ceiling lives in one senior's head, you do not have a ceiling. You have a single point of failure with good opinions.
This is also why "hold the taste" is a leadership verb. Choosing the exemplars, funding the crit time, refusing the tenth mediocre thing the roadmap would happily ship: those are allocation decisions, and only leaders can make them stick.
The uncomfortable part
Everything above is the comfortable half of the argument. The uncomfortable half is where taste comes from. Mine was built on thousands of production reps: the years of doing the 80% by hand, badly, then less badly, with seniors correcting me, until the discriminator trained. Those reps are exactly what we just automated. The junior designer of 2026 starts at the review stage, judging outputs without ever having made the ten thousand small mistakes that teach you why the output is wrong. We have kept the ladder's top and melted its bottom rungs.
I do not have a complete answer, and I distrust anyone who claims one. I have the beginnings of one: treat taste development as deliberate practice rather than a byproduct of production. Crit as curriculum, not ceremony. Reps on judging, with feedback, the way we used to get reps on making. Rotations through the parts of the craft AI has not eaten, research contact hours above all, because taste about users decays fastest of all. What I am certain of is the stakes: a discipline that automates its bar and fails to apprentice its ceiling will, within one generation, have neither.
The belief this expands: vision, belief three · where the bar gets paved: Helix