Skip to content
AI-Native PM
Field Notes

Field Note

Taste Is a Muscle, Not a Gift

For twenty years, making one good option was the hard part. Now ten arrive in a minute and the whole job is choosing, which means taste, the muscle that does the choosing, is the one you most need to build.

· 8 min read

You describe a screen to a tool, a settings page, say, or an onboarding flow, and a minute later ten finished-looking versions sit in front of you. They are all plausible, they all run, and each one looks like something you could ship to a real user this afternoon. But the abundance barely helps, because the question on the table is no longer how to make one good version but which of the ten is actually right, and that question does not get easier when the options arrive for free; it often gets harder.

For most of the history of software, production was the expensive part. Making one good option took time, money, and a person who could build it, so whoever could produce was scarce and valuable, and selecting from what they made was the easy part at the end. That economy has flipped. Production is close to free and options are abundant, so the hard part is now the part that used to be easy: choosing the one that ships and knowing why. The bottleneck moved from making to choosing, and choosing is the act taste governs.

Where the bottleneck moved, from making to choosingTwo panels side by side. On the left, before, effort concentrates at the top of a funnel because the scarce act was making one good option, which then ships. On the right, now, ten faint plausible options arrive for free and the scarce act, marked in the accent color, is choosing the one that ships. Caption: the work moved from production to selection.BEFOREmake one good optionthe scarce actshipNOWten plausible options, for freechoose the one that shipsthe scarce act nowshipthe work moved from production to selection.

This is the same pattern as the gap our note The Missing Layer describes, but it is a different gap. There, the thing between you and a shipped product was technical literacy, the vocabulary the tools assume you have. Here, the literacy is judgment. The model hands you ten options and leaves the choice to you, and the person who freezes in front of them is not missing a gift they were born without; they are missing reps.

The gift framing is wrong, and convenient

The most common thing said about taste is that you either have it or you do not. It is the explanation people reach for when someone makes consistently good calls, and it is the one the person making them tends to accept, because it is flattering. It is also wrong in a way that matters, because it ends the conversation exactly where the useful part begins. If taste is a gift, there is nothing to teach and nothing to practice, and the people who have it are excused from explaining how. That is convenient for them and useless for everyone else.

The essay this note extends, The Discipline, puts the principle in a single line: taste is the editorial muscle that decides what a product becomes by deciding what it refuses to be, and a model can produce ten options in a minute but cannot tell you which one to ship. This note is the longer argument behind that line.

Taste is a muscle, not a gift, and in a market where anyone can generate ten options in a minute, it is the muscle that decides which one ships.

A muscle has a training method, and this one has three parts that compound: exposure, reps, and deliberate exclusion. None of them is mysterious, and all of them are available to a product person who decides to do the work.

Exposure: see a lot, see it closely

The first part is study, and it is more demanding than it sounds. Julie Zhuo, writing in June 2025, describes excellent taste as pattern recognition built on a vast base of cultural knowledge, and the people who have it as voracious students of their discipline, obsessing over details that others might find tedious. The phrase that matters there is voracious students. Exposure is not osmosis. It is looking at a great deal of work and asking, in front of each piece, why this one holds together and that one does not.

The clearest place to train this eye right now is the generated default. Ask a tool for a landing page or a dashboard with no further direction and you tend to get a recognizable look: purple gradients, pill-shaped cards, a certain stock-friendly sameness. A community of builders has started naming these tells and cataloguing them under headings like anti-slop, and design-focused teams now treat rejecting that default on sight as a basic competence. The interesting thing is what the rejection requires. To know the generated option is generic, you have to have seen enough non-generic work to feel the difference, which is exposure doing its job. The trained eye is not born knowing the default is wrong; it learns by looking.

The everyday version of this for a product person is a product diary, a running record of interfaces and flows you use, with a sentence on each about why it works or grates. It costs a few minutes and it builds the only thing exposure can build, which is a larger library of examples to recognize against.

Reps and exclusion: choose, cut, track

Exposure without practice produces a good critic, not a good builder. The second and third parts of the muscle are reps, making real choices and watching how they land, and deliberate exclusion, deciding what the thing refuses to be.

The relationship between practice and time is the part most people get backwards. Time in the chair is not the same as deliberate practice, and the two come apart cleanly. Deliberate practice without much experience produces junior PMs with sharp taste, and experience without deliberate practice produces seasoned PMs with mediocre taste. The reps are what build the muscle, not the years, which is good news, because reps are something you can choose to do starting now rather than something you have to wait out.

The taste muscle, three repsThree bands left to right, each tagged with one rep that builds taste: exposure, reps, and deliberate exclusion. A connector on the right collects the three into one output, judgment you can defend. Caption: taste is a muscle, not a gift.THREE REPSExposuresee a lot, closelyRepschoose, track outcomesExclusiondecide what it refuses to bejudgment you can defendtaste is a muscle, not a gift: the reps compound.

Exclusion is the rep that gets skipped, and it is the one that shows. What you leave out shapes a product as much as what you put in, and the person with taste cuts at least as often as they add. Airbnb's 2024 Summer Release, its Icons program, is a clean public example of exclusion as a deliberate choice. Brian Chesky described it as primarily a brand positioning and a brand investment, and said it obviously was not a business. The executions made the restraint visible: an exact replica of the house from Pixar's Up, built down to the smallest detail, and a re-creation of the X-Mansion where artists hand-painted black outlines onto real furniture to make a three-dimensional room read as a flat cartoon. Each one was defined by what it refused to optimize for, which was revenue, in favor of attention. That is taste as exclusion at the scale of a company decision, and it maps directly onto the line from The Discipline about deciding what a product refuses to be.

The smaller-scale version is a habit. Cut one thing every week from your own work, whether a feature from a plan, a sentence from a document, or a slide from a deck, and notice that the work usually gets better. Cutting is the most underused move a product person has, and it is pure exclusion practice.

Why taste became the moat

The reason this matters more now than it did three years ago is straightforward: when features commoditize, taste is what is left to compete on. If any team can generate a competent version of your feature within a quarter, the competent version stops being a differentiator, and the question becomes who chooses better.

Working operators are saying this directly across very different product categories. David Okuniev, who co-founded the form builder Typeform, put it bluntly in early 2026: in a world where it is so easy to put things together and the language models are doing a lot of the heavy lifting, taste and design and being able to direct it is going to be the big differentiator. The project tracker Linear has built its whole position on the same bet, competing on opinionated design and quality rather than feature count, with its leadership arguing that the era of growth hacking is over and that product teams should compete on quality of work instead. A form builder and a project tracker have almost nothing in common as products, and they have landed in the same place on what now separates winners from the field.

This is also why taste is the highest-leverage skill to build for an AI-native PM in particular. As the tools commoditize execution, the people who pull ahead are not the ones who ship more, they are the ones whose ships are noticeably better, and better is a taste judgment before it is anything else.

The honest counter: maybe the model chooses better

There is a real objection here, and it became a public argument in February 2026. Paul Graham predicted on the fourteenth that taste would become even more important in the AI age, because when anyone can make anything, the big differentiator is what you choose to make, and he linked it back to his 2002 essay Taste for Makers. Two days later, on the sixteenth, OpenAI president Greg Brockman distilled the idea to five words, taste is a new core skill, and the debate took off. Linear's head of product, Nan Yu, pushed back hard with the uncomfortable version of the question: you probably do not have better taste than the model.

That objection deserves a real answer rather than a dodge, and the answer is to be precise about what a model does and does not do. A model can produce ten options, score them against patterns it has seen, and predict what a discerning audience tends to prefer, and on the narrow task of ranking against known preferences it may well outperform a given person. Granting all of that changes less than it seems to, because ranking is not the whole act. Someone still has to own the choice, stand behind it when the room disagrees, and accept being wrong in public when it does not work. Zhuo's framework is useful here precisely because she does not flinch from the hard case: she expects even the cultural, movement-making part of taste to be partly reproducible over time, and locates the last durable human advantage one level deeper, in agency, the will of the hand to act according to our values. Whatever a model ranks, the accountability for the decision does not move. It stays with the person whose name is on it.

What the model ranks, versus what stays yoursTwo columns side by side. The left column lists what a model does well: produce ten options, score them against patterns, predict what an audience tends to prefer. The right column lists what a person owns: choose the one that ships, defend it to the room, and be accountable for it in public. A soft center line marks that the two cooperate. Caption: the ranking can move, the accountability does not.WHAT A MODEL DOES WELLWHAT STAYS YOURSProduce ten optionsten finished-looking versions, fastScore them against patternsrank each one by learned rulesPredict what tends to landwhat a discerning audience prefersChoose the one that shipspick, and know why this oneDefend it to the roommake the case others can weighBe accountable in publicwilling to be wrong, in the openThe ranking can move. The accountability does not.

So the ten options are not the problem, and they were never going to be your edge, because everyone gets the same ten. Choosing, defending the choice, and being accountable for it are the job, and all three get better with the reps. That is the whole case for treating taste as a muscle: if it were a gift, there would be nothing to do about any of this, and the fact that there is plenty to do is the good news.

Where this goes next

Taste is one of five fundamentals our essay The Discipline argues AI has not commoditized, alongside product sense, strategy, product-market fit, and intentional differentiation, and it is the one the current moment sharpens most directly rather than threatens. Read that piece for how taste sits among the others and why the whole set is the ground the new discipline stands on. The smallest version of the training is the one to start this week: keep a short product diary, run a weekly teardown of something you admire, and cut one thing from your own work every week. The deeper version, building taste deliberately on real shipped products where the system pushes back, is the work The Builder's Stack is built to put you through, because the only place the muscle grows is on something real.

Sources