AI can now write your PRD, run your discovery, and create your roadmap. So what remains purely yours? The answer is Product taste — and most product leaders cannot define it.
The Question the profession is avoiding
A question has been quietly unsettling product craft for eighteen months. It deserves a more honest answer than it has received. If AI can synthesise user research in seconds and create product specs in minutes, what exactly is the product manager’s job? The quick answers sound weighty but dissolve upon close review. “Strategy” can be modelled. “Leadership” is a behaviour, not a output. “Vision” often means the opinion of the most senior person in the room. None of these words describe a skill that AI cannot copy. None explain why a human should remain at the centre of a product organisation.
The real answer is more specific, more hard to hear, and more important than any of these vague words. The core human input to product craft is product taste. Developing it with intent is now the highest-leverage bet a product leader can make.
Defining Product Taste without mysticism
The word carries an unhelpful aura of feeling. In common usage, it suggests personal preference — a trait one either possesses or does not. This framing is just wrong for product craft. It is also the reason most discussions of taste remain trapped in vagueness.
Product taste, in practice, is not a preference. It is trained judgment about what should exist, what should not, and why the gap matters. It is the built-up residue of thousands of product choices, refined into a sharp eye for what works. This edge operates faster than review. However, it is grounded in experience rather than impulse. The designative.info group named this with uncommon precision earlier this year. Taste is not instinct in the mystical sense. Rather, it is instinct that has been trained, tested, and made accountable through practice. The gap is crucial. Untrained instinct produces faster noise. Trained judgment produces faster edge.
In practical terms, taste shows up as the ability to look at ten plausible directions and select the one that will matter — before the data confirms the choice. It is the capacity to see when a product is functionally complete yet in felt terms hollow. And it is the judgment to diagnose the clear constraint that needs to change. These capabilities do not emerge from processing more data. They emerge from the slow buildup of hands-on experience.
The Paradox: Why AI makes Product Taste more valuable
The old constraint has vanished
A tempting belief floats in the product group. AI reduces the need for human judgment by providing better data and faster review. The opposite is true. Understanding why is essential for any product leader who intends to remain relevant.
When building was expensive, taste was a luxury few teams could exercise. Most orgs could ship one version of a feature. They could pursue one market position. They could run one pricing experiment per quarter. The sheer cost of execution imposed a natural discipline. There was no space for bad choices because there was barely space for good ones. Constraints performed the filtering that taste would otherwise need to do.
The new bottleneck
AI has altered this dynamic at its root. The cost of creating options has collapsed. Ten PRDs can appear in an hour. Five prototypes can land before lunch. A rival analysis, a pitch doc, and a go-to-market plan can all be ready by close of business on Tuesday.
The constraint that once forced careful selection has gone. In its absence, the bottleneck has shifted. It is no longer “can we build this?” It is “should we build this?” That second question is a taste question. No amount of computational power resolves it.
The sameness problem
Productboard’s CEO Hubert Palan and Intercom’s president Archana Agrawal surfaced this tension with admirable edge. As the cost of building drops, the duty to choose well increases. Velocity ceases to be the differentiator. Judgment turns into the scarce resource.
LogRocket’s review extends this into more hard to hear territory. When every PM uses the same AI tools on the same data, the outputs converge. Entire product categories drift toward sameness. This happens not because anyone copied anyone else. It happens because the profession handed over its signal detection to the same system. The resulting products are competent, functional, and alike. They are products without taste.
Therefore, taste has become the primary engine of product edge. The orgs whose products feel distinctive and irreplaceable are not the ones with the most advanced AI toolchains. They are the ones whose leaders possess judgment that the tools could never generate.
The Five Dimensions of Product Taste
Taste sounds abstract until you decompose it into clear choice categories. Five dimensions account for the vast majority of taste-dependent choices that product managers face daily. Each represents a judgment call that AI can inform but cannot make.
1. The habit of problem selection
AI excels at pattern insight in user data. It clusters feedback into themes and surfaces often mentioned pain points. However, it cannot tell which problem is worth solving. Frequency is not a proxy for importance. Volume does not indicate urgency.
Consider a common scenario. The most mentioned problem in your feedback corpus might be the least valuable to solve. It might affect a segment you do not intend to serve. The solution might would degrade the experience for core users. The problem might be is a symptom of a deeper architectural issue that demands a different intervention.
Taste in problem selection means choosing the problem that unlocks a cascade of downstream value — even when the data does not rank it first. It means distinguishing between what users name and what they actually need. This gap needs not just empathy but a theory of the product’s role in the user’s world. This skill separates senior product leaders from competent operators. Junior PMs solve the problem the loudest customer describes. Directors solve the one that removes the most friction. VPs solve the one that changes the market’s expectations.
2. The courage of saying no
AI creates options with striking skill. Feed it your product context and it will produce twenty feature ideas. Each will be plausible. Each will cite some strand of evidence. All will survive a stakeholder meeting. The challenge is not creating these options. It is killing the nineteen that should not exist.
The finest products in any category are defined by what they with intent omit. Every feature left unbuilt is edge gifted to the features that ship. Every skill on the cutting room floor is complexity the user never navigates and the team never maintains. AI cannot make this call. It does not absorb the weight of built up complexity. It does not feel how each additional surface increases cognitive burden on the user and maintenance burden on the team.
Saying no at scale needs belief. Conviction is the working form of taste. It is the deep, experience-built sense that this product should be this and not that. And that the habit of hold that line is more valuable than the flexibility to allow for every request.
3. The standard of quality calibration
There is a recurring moment in every product cycle. The work is technically complete. Every specification is met. QA gives the green light. And yet something is wrong. Not wrong enough to file a bug. Not wrong enough to delay the release. Wrong in a way that shows up as a persistent, low-grade dissatisfaction among users who cannot name the source.
Taste is the ability to spot what is lacking when there is no spec to point to. A workflow might need two steps fewer. A tone might be correct but cold. Or a feature might work fine yet feel out of place — right on paper, wrong in practice. This is the quality bar that no automated system can establish. It needs holding a mental model of not just what the product does but what it should feel like at its best. AI can verify whether a product meets requirements. Only taste can tell whether it meets its standard.
4. The instinct of timing
When to launch matters as much as what to launch. Release too early and the market lacks the vocabulary to understand the offering. Release too late and the window has closed. Your input arrives as a footnote rather than a headline.
AI can present market data and rival timelines with impressive breadth. However, timing instinct operates on a different substrate. It feeds on weak signals that do not surface in dashboards. Watch for a subtle shift in how customers frame a problem during discovery calls. Notice a rival’s minor change in messaging that suggests a strategic pivot. Catch a regulatory development that has not yet reached the news cycle but will reshape buyer priorities within months.
Taste in timing means acting on pattern insight that no model can replicate. The patterns were constructed from years of observing how markets absorb new ideas. They encode the cadence of adoption, the rhythm of resistance, and the inflection points where momentum turns into irreversible.
5. The perception of the experience gap
This is the most elusive dimension. It is also the one that most reliably distinguishes rare caseal products from merely working ones. The experience gap is the distance between what a product does and how it feels. Two products can possess identical feature sets and produce entirely different emotional responses. One feels effortless and anticipated. The other feels like work. Each interaction carries a faint but unmistakable friction that builds up into the quiet choice to seek an alternative.
The difference lies in micro-choices that individually seem trivial. Look at the sequence of fields in a form. Examine the cadence and tone of notifications. Notice the words chosen for an error message. Think about the moment the system requests permission versus the moment it acts on its own. These choices as a group define the product’s character.
AI creates working products with considerable skill. Taste creates products that feel intentional. That gap between working and intentional is where product taste lives. It tells whether users tolerate a product or love it.
Why Product Taste Becomes Critical in the Agentic AI Era
Everything above applies to traditional product craft. However, agentic AI introduces a new category of taste choices. These are, if anything, more consequential and less amenable to data-driven resolution.
When you design an AI agent, you face choices that no framework resolves. Where should the autonomy boundary sit? Should the agent approve purchases below ten thousand dollars on its own, or below twenty-five thousand? No dataset contains the answer for your clear users, your clear risk profile, and your clear regulatory environment. That is a taste call.
When should the agent ask for help? Set the confidence threshold too low and the agent escalates everything. It turns into an expensive routing system that adds latency without value. Set it too high and the agent acts when it should defer. Trust erodes. The right threshold needs domain expertise and user empathy. Data alone cannot set it.
How much should the agent remember? Retaining all things risks an experience that feels surveilled. Retaining nothing feels frustrating and amnesiac. The right memory depth is a design choice that demands judgment, not optimisation. Every one of these choices will tell whether an agentic AI product succeeds or fails. And every one is purely a taste choice — a judgment call that AI cannot make about its own behaviour.
Cultivating Product Taste: A Practical Programme
If taste is trained judgment, then it can be developed with intent. The following practices represent the most reliable pathways to strengthening it.
Interrogate what you admire
Do not merely use excellent products. Dissect them with the analytical intensity of a close breakdown. Ask what makes a given onboarding sequence feel welcoming rather than burdensome. Examine what makes a notification feel helpful rather than intrusive. Explore what makes a complex product feel simple without being simplistic.
The act of articulating why something works is the core exercise through which taste develops. Each act of naming adds to your bank of judgment. Over time, that vocabulary turns into the lens through which you evaluate your own work.
Ship and absorb consequences
Taste does not emerge from review alone. It takes shape through the experience of shipping and observing the reception with honest attention. The product manager who has shipped ten products possesses a judgment vocabulary that no amount of reading can approximate.
Every launch that underperformed taught something that success never could. Taste is the residue of hands-on experience. The painful kind compounds fastest.
Seek disconfirming perspectives
AI gives you what you ask for. It confirms your framing and renders your ideas in polished prose. Taste develops in the opposite direction. It grows through exposure to perspectives that challenge your assumptions and force you to defend or abandon positions held with confidence.
Find the person on your team who disagrees with you most often. Cultivate the habit of listening to them before you listen to the model.
Practice the final pass
Before you release any significant decision, perform one additional review. Not to find defects. To locate the element that is merely adequate when it could be rare caseal. Taste lives in that final pass. It is the difference between a product that works and a product that sings.
Cross-pollinate widely
The richest product taste is nourished by exposure to distant domains. Study how a great restaurant anticipates a guest’s needs before they are expressed. Observe how a film editor controls pacing through the duration of a cut. Examine how an architect designs a building that guides movement without signage.
The PM whose references are only other software products will develop software taste. Competent, but bounded. The PM who draws from architecture, hospitality, film, and game design will develop experience taste. That is a richer, more useful form of judgment.
The Conviction This Requires
One final dimension deserves explicit note. Taste needs the willingness to act on judgment before evidence arrives — and to accept ownership for the outcome.
AI provides cover. It creates analyses that justify choices and forecasts that distribute responsibility across models. Taste offers no such comfort. It asks you to stand in front of a room and say, “this is what we should build, and this is why.” Your belief is grounded not in a model’s output but in hard-won pattern insight. That insight comes only from years of making consequential choices and living with their results.
In the feature era, the product manager’s value was knowing how to ship. In the agent era, where AI handles the mechanics of execution, the value migrates. It moves to knowing what should exist. That question — what should exist? — has no modelic answer. It needs taste. Trained, practised, consequence-informed judgment that distinguishes products people tolerate from products that reshape expectations.
That is taste. And in an era where all things else can be generated, it is the one skill worth cultivating above all others.
Check out Cat Wu’s Podcast on Lenny’s podcast
