Product Taste as a Competitive Advantage
In an age of infinite generation, choosing well is the scarce resource
Every creative tool in history has shifted the bottleneck. The printing press moved it from copying to writing. The camera moved it from painting to seeing. Digital audio moved it from performing to producing. Each time, the skill that mattered most before the tool arrived - manual reproduction, realistic rendering, flawless live performance - depreciated, and a different skill, one that had always existed but never been the binding constraint, moved to the centre.
AI is doing this again, and the bottleneck it is revealing is taste.
When generating a logo costs six weeks and ten thousand dollars, you get three options and choose the best one. The cost of generation forces economy, and that economy obscures the selection problem. You never have to be particularly good at choosing because you never have enough options for choosing to be hard. When generating a logo takes fifteen seconds and costs nothing, you can have three hundred options before lunch. The generation problem is solved. The selection problem - which was always there, hidden behind the cost of creation - is now the only problem. And most people, it turns out, are not very good at it.
The ability to choose well - to look at a set of options and identify the one that is right, or to reject all of them and articulate precisely why - is what I mean by taste. It is the scarcest resource on any team using AI to create, and it is becoming more scarce relative to demand with every improvement in generation capability.
What Taste Actually Is
I want to be precise about the word, because it carries baggage. When most people hear "taste," they think of subjective preference. I like blue, you like red. I prefer minimalism, you prefer ornamentation. If that were all taste meant, it would be uninteresting - a personality trait, not a skill. You could not build a competitive advantage on it any more than you could build one on preferring coffee to tea.
Taste, as I am using the word, is something different. It is pattern recognition developed through massive, deliberate exposure to both good and bad work. The designer who has studied a thousand typefaces does not prefer Garamond to Arial the way you prefer chocolate to vanilla. She recognises that Garamond communicates something specific in a specific context - authority, warmth, tradition - and that Arial communicates something else - neutrality, modernity, efficiency - and she can match the typeface to the intent without conscious deliberation. This is not preference. It is expertise that presents as instinct.
The same faculty operates in product development, in strategy, in hiring, in writing. The product leader who has shipped twelve products can feel when a feature is dilutive before the usage data confirms it. The investor who has evaluated a thousand pitches can sense when the founder's story does not cohere, before identifying the specific gap. The writer who has read widely can hear when a sentence is wrong - not grammatically wrong but tonally, rhythmically, structurally wrong - and fix it without being able to cite a rule.
In each case, taste is the residue of accumulated experience. It is what remains after you have seen enough examples to internalise the patterns. It looks like intuition. It is actually expertise that has been compressed below the threshold of conscious articulation.
The Convergence Problem
Practising the violin does not make people converge on a favourite key. But exposing people to enough good and bad work makes them converge on what good looks like. That convergence is the evidence that taste is knowledge, not opinion.
There is a common objection to taking taste seriously as a professional skill: "It's subjective." If taste is just personal preference, then one person's taste is as valid as another's, and the whole concept dissolves into relativism.
But there is an inconvenient fact that the relativists have to contend with. Expert judgments converge.
Show a set of website designs to a hundred people with no design training and you will get widely scattered evaluations. Show the same designs to twenty experienced designers and the evaluations cluster. Not perfectly - experts disagree. But they disagree far less than novices, and they disagree about different things: about the subtle questions (is this particular shade too warm?) rather than the fundamental ones (is this layout working?). The fundamental assessments converge to a degree that is difficult to explain if taste is purely subjective.
The convergence pattern appears in every domain where taste operates. Sommeliers agree with each other far more than untrained tasters do, despite the popular narrative that wine expertise is fraudulent. Experienced editors, reading the same manuscript, identify the same structural weaknesses. Senior engineers, reviewing the same codebase, flag the same architectural problems.
If taste were merely preference, expertise would not produce convergence. Exposing people to enough examples of good and bad work, across enough contexts, makes them converge on what good looks like. This convergence is the evidence that taste is a form of knowledge, not a form of opinion. It can be developed. It can be evaluated. And - the point of this essay - it can be a competitive advantage.
Taste in Product
Let me bring this down from the abstract.
The defining act of product taste is the decision to remove something. Not because it does not work - it works fine - but because it is not right. It is the feature that tested well in isolation but makes the product feel cluttered. The onboarding step that is individually logical but collectively exhausting. The setting that gives users control they did not ask for and will never exercise. Cutting these things requires a specific kind of courage, because every one of them has a defender, a rationale, and data to support it. Taste is what tells you the data is answering the wrong question.
I have watched this play out dozens of times. A product team adds a feature because users requested it. Usage is modest. They add another. Usage is modest. After a year, the product has forty features, each individually defensible, none individually compelling, and the whole thing feels like a committee designed it - because a committee did, with the committee being the aggregate of every user request the team lacked the confidence to refuse.
The products that win - the ones people describe with words like "clean" or "obvious" or "it just works" - are products where someone had the taste to say no. Not arbitrarily. Not out of aesthetic snobbery. Out of a clear, experience-driven understanding of what the product is for and the discipline to refuse anything that does not serve that purpose, regardless of how reasonable it sounds in isolation.
This is why taste functions as a competitive advantage and not merely a nice quality to have. Two teams with identical technology and identical market access will build different products. The team with taste will build something coherent. The team without it will build something comprehensive. Coherent wins, almost every time, because coherence is what users experience as quality. They cannot always name it, but they can feel it. It is the reason they choose one product over another that does "more."
The AI Amplification
The taste gap: AI is lowering the floor (anyone can generate competent work) while doing nothing to raise the ceiling (which still depends entirely on human judgment). In a world where the floor is rising and the ceiling is fixed, the distance between adequate and excellent becomes the most valuable territory in the market.
AI changes the economics of taste in a way that is not yet widely understood.
When creation was expensive, taste was a tiebreaker. You had two options (because producing more was prohibitively slow), and taste helped you pick the better one. The stakes of the choice were moderate because both options were usually acceptable - you had invested enough time in each to ensure a minimum standard. Taste saved you from "good enough" but the floor was already high.
When creation is nearly free, taste becomes the primary determinant of quality. You now have two hundred options, generated in minutes, and most of them are competent. They are grammatically correct, visually balanced, structurally sound. AI is extremely good at producing work that clears the minimum bar. What AI is not good at - and may never be good at, for reasons that go beyond the purely technical - is distinguishing between competent and excellent. That distinction is taste, and it is now the only thing standing between a team and an avalanche of mediocrity.
I see this in practice constantly. Two founders using the same AI tools, with similar technical ability, producing work of wildly different quality. The difference is not in how they prompt the AI. It is in how they evaluate the output. One founder looks at the AI's work and says "that's good enough." The other looks at it and says "the structure is right but the tone is wrong in the third paragraph, and the opening buries the lead." The second founder's product, over time, diverges sharply from the first's - not because of better technology but because of better judgment applied to the same technology.
The gap between these two outcomes is taste.
Developing Taste
Taste is learnable. It is not teachable. The distinction matters and is frequently collapsed.
You cannot attend a workshop and leave with better taste. You cannot read a book about taste and acquire it. Taste develops through a specific process: massive exposure to work (both good and bad), repeated attempts to make things yourself, honest evaluation of why your work falls short, and enough repetitions of this cycle for the patterns to sink below conscious thought. The photographer who has taken fifty thousand photographs and ruthlessly culled them has taste the weekend hobbyist does not. Not because of innate talent. Because of volume, combined with self-criticism.
This is why taste correlates with experience but is not guaranteed by it. Some people work for twenty years without developing taste because they never close the feedback loop. They produce work, it is received, they produce more work. Without the step of honest self-evaluation - looking at your own output and asking "is this actually good, or is it merely finished?" - the exposure does not convert into pattern recognition. It just becomes repetition.
The implication for teams and organisations is practical. Taste cannot be hired the way other skills can. You cannot write a job description for it. You cannot test for it with a coding challenge or a case study. You can only recognise it in the work - by looking at what someone has built, what they have chosen, what they have refused, and whether those decisions, taken together, produce something coherent. This makes taste difficult to evaluate, which is part of why it is undervalued. But difficulty of evaluation does not reduce its value. It increases its scarcity.
The New Bottleneck
For most of the history of technology, the bottleneck was capability. Could you build it? Did you have the engineering talent, the infrastructure, the time? Companies that could build things faster and more reliably than their competitors won, because building was the hard part.
AI is dissolving that bottleneck. Building is becoming cheaper and faster every quarter. The things that were difficult to create eighteen months ago - a working prototype, a polished interface, a first draft of copy, a data analysis pipeline - are now generated in hours. The companies that won on build speed are discovering that their advantage is eroding, because everyone can build quickly now.
The new bottleneck is judgment. Not "what can we build?" but "what should we build?" Not "can we generate this?" but "is this the right version of this?" These are taste questions. They require the accumulated pattern recognition that comes from years of making things and evaluating what works, the confidence to reject competent work in favour of excellent work, and the ability to articulate a standard that an entire team can internalise and follow.
The leaders who will matter most in this era are not the ones who can generate the most. They are the ones who can choose the best. The skill that looks like magic is actually expertise, developed slowly, through deliberate exposure and honest self-assessment. And it is the one thing you cannot outsource to the machine - not because the machine lacks processing power, but because choosing well requires knowing what you want. Knowing what you want requires having a developed point of view about what good looks like. And developing that point of view is work that no tool, however powerful, can do on your behalf.