Taste Is Inarticulation With Good PR

Ask anyone in product, design, or tech generally what humans will still bring that AI can't replicate, and the answer increasingly comes back to "taste". Paul Graham says taste will become more important in the AI age. Lenny Rachitsky put it first on his list of timeless skills: "[it's] the bottleneck when AI generates unlimited outputs." Garry Tan frames it as one of only two things humans still need to do. The word has calcified into a consensus that's starting to feel a lot less like a profound insight and a lot more like a security blanket.

What gets called "taste" bundles two things together. There's the generative side: the lived experience, the real-world inspiration, the unexpected connections a human brings to the work. A chef drawing on a memory from decades ago; a designer making a connection no one else would have made. That's genuinely human and I'm not disputing it. Then there's the evaluative side, where taste comes in: the judgment of whether what's in front of you is good or not — whether the dish works, whether the copy lands, whether the architecture holds. They get conflated because for creators they run in a tight loop, evaluating as they make, but they're doing different work.

I don't have a problem with taste as a concept; I struggle with how the argument is being made. People try to define what taste is ("judgment under uncertainty", "knowing what good looks like before others can see the pattern"), but trace any of those definitions back and you land in the same place: recognition happening faster than articulation.

"I'll Know It When I See It"

The most honest definition of taste, even among those who have tried to formalize it, is the same: "I'll know it when I see it." That's Justice Potter Stewart's line from Jacobellis v. Ohio (1964) as an admission that he couldn't define what he was ruling on, only recognize it. It's stuck around because it captures something honest about pattern recognition, but as a definition of a competitive advantage, it's not doing any real work.

"I'll know it when I see it" just means recognition is happening faster than narration. It describes the timing, not the mechanism. And when taste advocates use it, they're usually describing the same underlying thing: accumulated pattern recognition that they haven't converted to language yet.

A senior designer looks at a layout and something is off. A strong editor reads a sentence and immediately knows it doesn't land. An experienced PM reviews a roadmap and senses that the priorities are wrong. They're often right, and the reactions are valid, but those instincts aren't beyond replication. They're the output of processing that hasn't been articulated yet. The gap between the feeling and language is a gap of effort, not a gap of human essence, even when those feelings are drawing on experience that's genuinely hard to put into words.

"Taste" is Inarticulation with Good PR

The causal chain that doesn't hold up is the move from "can't currently be articulated" to "can't be articulated in principle" to "therefore exclusively human." Each one of those steps is a reach, and the current discourse on taste is quietly proving it. There's a practice being explored now called "taste extraction" where you use AI to run A/B interviews, surface why you only prefer one option over another, and build a document that captures your judgment so a system can apply it at scale. The output gets called an "Essence document": a written model of your taste that compounds across hundreds of outputs.

If taste can be extracted, interviewed out of you, and encoded in a document a system can use, it was never really inarticulate. It was unexported cognition sitting in your head because no one ever pushed you to get it out. The barrier was never some irreducible human capacity — it was effort. Most people with genuinely good taste have never been required to examine what's producing it, which is very different from it being unexaminable.

"Taste as moat" is a bet on perpetual inarticulation: that people will never do the hard work of understanding their own criteria. Articulated criteria will still require judgment to apply — that's true of expertise at every level — but explicit criteria with some residual judgment is a very different thing from a black box. One you can manage; the other you can't.

What This Costs You as a Manager

If your team's quality standards live in someone's head as "taste", you've built a black box into your process, and black boxes make bad systems.

Think about what it would look like to give a team member feedback the way taste usually gets invoked: "This isn't quite right, but I'll know it when I see it." No serious manager actually talks that way. Good feedback sounds like describing what good looks like, what's missing, and why. The ability to articulate what "good" means, specifically rather than vaguely, is a basic management competency. We already accept this in every other domain. Calling it "taste" is just one way of exempting one domain from that standard.

And if we do continue to chalk the gap between current and desired outputs up to "taste", the consequences will continue to compound. You can't hire for it reliably because you can't define what you're screening for. You can't develop people towards it because you can't specify what they need to learn, and you can't scale it because it stops where that person's instincts stop. When taste is a quality bar, the person holding the taste becomes the bottleneck and they've made themselves undebuggable.

The argument also quietly excludes junior team members. Juniors can create — they bring energy, fresh perspective, and things you haven't thought of. What they haven't built yet is the evaluative function: the judgment of what good looks like. Developing them into senior practitioners means naming those criteria explicitly enough that they can learn it. If you can't name it, you're not teaching. You're just hoping it gets absorbed by osmosis. And without that pipeline, you've closed off the path that produces the next generation of people with the judgment you say you need.

You can't manage what you can't describe. Taste without articulation is just opinion with a more flattering name.

Judgment Without Language Doesn't Scale

None of this dismisses underlying judgment. Pattern recognition happens, and the ability to evaluate faster than you can explain is a fundamental part of the human experience, and you don't have to narrate it while you're doing it. The articulation is the reflective work you do afterward. The generative side — the lived experience, what you bring to the work — is yours. The evaluative side, the part that tells you whether the output is good, is what you owe your organization. Claiming that gap is permanent, and therefore makes you irreplaceable, is a cop-out.

The people who will be genuinely hard to replace aren't the ones who have good taste. It's the ones who can take their taste apart, name its components, and convert it into something that scales, which is arguably the harder skill. It requires interrogating not only whether something is good, but what specifically makes it good, what principles that it reflects, and whether that principle holds in adjacent cases. This will be the defining quality of those who are just good consumers of these tools versus the people who are building the systems that scale beyond themselves.

The good news is that articulation is learnable. You can build the practice of pressure testing your own criteria, writing them down and watching them sharpen under scrutiny. The question isn't whether your judgment can be put into words. It's whether you'll do the work to put it there.