AI 시대, '좋은 취향'만이 유일한 해자다

Good Taste the Only Real Moat Left Apr 3, 2026 # ai # analysis # engineering # strategy AI and LLMs have changed one thing very quickly: competent output is now cheap. A landing page can be generated in minutes. A product memo can appear in a single prompt. A pitch deck can look polished before anyone has done the hard work of deciding what the company actually believes. That is why taste has become a serious topic in tech. When everyone can produce something that looks decent, the advantage shifts to judgment. The people who stand out are no longer just the ones who can produce. They are the ones who can tell what is generic, what is true, and what is worth pushing further. But there is a second point that matters just as much: taste is not the final answer. If humans reduce themselves to selecting from AI outputs, they risk becoming reviewers of a machine-led process instead of builders with real stakes in the outcome. The real opportunity in the age of AI and LLMs is not to become a better selector. It is to combine taste with context, constraints, and the willingness to build something that could not have emerged from the average alone. What taste actually means In this context, taste is not about luxury, status, or personal aesthetic branding. It is about distinction under uncertainty. Most meaningful work does not come with perfect data. You do not get a spreadsheet that tells you which sentence will make a customer care, which feature is worth a month of engineering time, or which design crosses the line from polished to forgettable. You still have to decide. Taste shows up in three places: What you notice What you reject How precisely you can explain what feels wrong That last part matters more than it first appears. Many people can say, "this feels off." Far fewer can say, "this fails because it sounds like every other SaaS product," or "this explanation collapses a regulatory constraint into marketing language and will confuse the customer." Taste becomes useful when it moves from vibe to diagnosis. Why AI and LLMs flatten the middle LLMs are extraordinary pattern-compression engines. They absorb huge volumes of language, design patterns, and interfaces, then recombine them at speed. That is their strength. It is also their default bias. By design, these systems are much better at producing statistically plausible output than at originating something deeply specific to your exact context. Left alone, they tend toward the safe center of the distribution. That is why so much AI-generated work feels familiar: Landing pages with different logos but the same structure Product copy that could describe almost any app Essays with clean headings and little lived judgment Visual design that looks modern, but not memorable This is not a failure in the catastrophic sense. It is a success at average. The problem is that average used to be hard enough that it still created some separation. Now it is abundant. The result is a crowded 7 out of 10 world. The middle is full. The new bottleneck is judgment Before AI, mediocre work often reflected a lack of time, resources, or execution skill. Today mediocre work often means something else: the person stopped at the first acceptable draft. That is the economic shift AI introduces. It compresses the cost of first drafts, which means the value moves downstream. The scarce part is now the ability to say: This looks fine, but it is too generic This sounds impressive, but it hides the real trade-off This interface is polished, but it does not fit how the user actually thinks This plan is ambitious, but the operating constraints make it unrealistic In other words, the scarce skill is not generation. It is refusal. AI as a mirror for your own taste One of the most useful things about AI is also one of the most humbling: it reveals how clear your own judgment actually is. Ask an LLM to produce ten versions of a homepage hero, onboarding flow, support email, or product pitch. You will usually see a pattern: A few clearly weak versions A large cluster of acceptable versions One or two that seem closer to what you want The interesting question is not, "Which one should I pick?" It is, "Why are most of these still wrong?" Your answer to that question is the quality of your taste. If your critique stays vague, your taste is still underdeveloped. If your critique becomes precise, your judgment is stronger than the model output. You can then use the model well instead of being led by it. A practical way to think about it is this: Layer AI and LLMs do well Humans still need to do Generation Produce many plausible variations quickly Decide which direction matters Pattern matching Recombine common structures and phrasing Spot what is too generic for this situation Optimization Improve toward a stated target Decide whether the target itself is right Scaling Turn one idea into many assets Carry the real context, stakes, and consequences The system can generate options. It cannot supply ownership. A practical loop for training taste Taste improves through repeated exposure, critique, and shipping. AI can accelerate that loop if you use it correctly. A simple method looks like this: Pick one high-leverage artifact from your week. A paragraph, a pricing explanation, a dashboard label, a customer email, or a key slide. Generate 10 to 20 versions with an AI model. For each version, write one sentence that starts with "fails because..." Rewrite the strongest version with a hard constraint such as: No buzzwords One idea per sentence Must acknowledge a real trade-off Must make sense to a first-time user Ship the final version somewhere real and observe what happens. The goal is not to let AI choose for you. The goal is to build a sharper rejection vocabulary. Over time, this changes how you work. You stop admiring polish for its own sake. You get faster at spotting empty specificity, borrowed tone, and fake confidence. Why taste alone is not enough This is where the conversation gets more interesting. There is a strong version of the "taste matters" argument that quietly pushes humans into a narrow role. In that version, AI generates many outputs and the human stands at the end of the pipeline selecting the best one. That is a useful role, but it is also too small. Historically, important work did not emerge from detached selection alone. It emerged from co-creation under constraint. Builders argued with reality, with collaborators, with budgets, with materials, with timelines, and with the consequences of getting things wrong. That friction matters. It is where depth comes from. Once you see that, the risk becomes clearer: if human value is reduced to curation, the human becomes a discriminator in a mostly machine-driven loop. The analogy to machine learning is imperfect but useful. In generative adversarial setups, the discriminator exists to help the generator improve. Once the generator is good enough, the discriminator is not the part that ships. The warning is not that taste has no value. It does. The warning is that taste without authorship, stake, or construction can become a narrow and eventually fragile role. What humans still do that models cannot own AI can generate. It can recombine. It can optimize against prompts. What it cannot own in the human sense are the parts of work that carry real consequence. Three examples matter: 1. Holding the stake Real products operate under consequences that do not fit neatly inside a prompt. Trust, regulatory exposure, outage risk, team capacity, customer confusion, brand damage, and on-call pain all live here. A model can suggest copy for a payments feature. It cannot hold responsibility when that copy obscures a regulatory limitation and support tickets spike. 2. Working with the truly new Genuinely new ideas often look wrong at first because they do not resemble the training set. They feel awkward, incomplete, or suspiciously non-standard. Humans can sit with that discomfort. They can