Taste Is the New Bottleneck When Cheap Execution Floods Work

Taste is becoming the binding constraint in a field where output is no longer scarce. When a system can generate a hundred competent variants in the time it once took to make one, the limiting input shifts from production to discernment: knowing which version is worth keeping, which to cut, and which should never have been proposed in the first place.

This is a quiet inversion, and it changes who matters in the room. For most of the discipline's history, the hard part was making the thing at all. Now the hard part is choosing among the many things that arrive, finished and plausible, each one asking for a verdict it cannot supply on its own.

• Generation has flattened to near-zero marginal cost; selection has not, and shows no sign of doing so.
• The editor now outranks the generator, because a considered rejection carries more signal than a fresh creation.
• Taste functions as a compression of ten thousand prior judgments into one fast, pre-analytic call.
• A field rich in candidates starves for the criteria that rank them against each other.
• Discernment, once treated as a flourish, is hardening into a measurable and teachable discipline.
• Abundance does not raise the average; left unmanaged, it quietly lowers the bar.

Why did execution stop being the scarce input?

Execution is the property of turning an intention into a finished artifact. For decades that step was expensive enough to act as a natural filter: only ideas someone genuinely believed in survived the cost of being built. The expense was the gate, and the gate did real work — it killed weak ideas before they consumed attention.

That gate has largely dissolved. When variants are cheap, the population of plausible-but-wrong artifacts explodes. A team can now hold forty drafts of the same paragraph, eight visual directions, three architectures that all compile and pass their tests. Abundance does not resolve the question of what is good; it sharpens it, because the cost of looking at everything rises even as the cost of making everything falls toward zero.

The constraint relocates. It moves from the hands that build to the eye that judges, and the eye is slower to scale than any generator. One tradeoff that surfaces is that throughput no longer correlates with progress — a system can produce all day and still drift sideways, because nothing in raw generation knows where it should stop, or whether it should have started.

What replaces the cost gate is a judgment gate, and judgment does not come for free. It has to be paid for in attention, and attention is the one resource that did not get cheaper. So the field finds itself richer in artifacts and poorer in the thing that was always doing the quiet work of saying which artifacts deserved to exist.

What does taste actually mean as a working capacity?

Taste refers to the learned ability to predict, quickly and before full analysis, whether a thing belongs. It is not preference and not opinion. It is a compressed model of consequences — an internalized sense of which choices age well and which collapse on second contact, formed long before anyone can articulate the reasons.

Seen this way, taste is less mysterious than it sounds. It behaves like any other expert judgment: built from volume, calibrated by feedback, and degraded when either is missing. The person with taste has simply seen enough outcomes to feel the difference between a choice that merely satisfies in the moment and one that holds under pressure over time. The feeling is fast; the apprenticeship behind it was not.

This matters because a capacity built from feedback can also be starved of it. In a generation-rich setting, output and consequence drift apart — things ship faster than their results return. Taste that stops receiving outcomes stops learning, and begins to fossilize into mere habit. Keeping the loop short is itself a discipline.

Discernment debt

The backlog of unjudged output that accumulates when generation outpaces the human capacity to evaluate it, and quietly compounds.

Negative selection

The discipline of defining what to reject first, using exclusion rather than approval as the primary instrument of quality.

Criterion drift

The slow loosening of standards that happens when abundance makes nearly every option look acceptable in isolation.

Why does the editor now outrank the generator?

A generator answers the question "what could this be?" An editor answers the harder one: "of everything this could be, what should it be?" The first question has been largely automated. The second remains stubbornly human, because it requires a stake in the outcome and a memory of what went wrong the last several times.

Rejection carries information that acceptance does not. When someone cuts forty drafts to keep one, the act of cutting encodes a theory of quality that the surviving draft alone could never express. That theory — the reasoning behind the cut — is the scarce asset. It cannot be generated by the system that produced the candidates; it can only be held by someone willing to be accountable for the call.

This reframes seniority. The valuable contributor in a generation-rich room is rarely the fastest producer; it is the person whose "no" is trusted. Their refusal compresses years of experience into a single decision others can build on without re-deriving it. Authority migrates from making to ruling on what was made, and that migration is already underway.

How do you turn taste from a flourish into a discipline?

A flourish is decorative and unaccountable; a discipline is repeatable and inspectable. The move from one to the other happens when judgment stops living only in someone's gut and starts leaving a trail others can read, question, and inherit. Privately held taste protects no one once its holder is busy or gone.

The practice has a rough order of operations, and the order is not arbitrary:

1. State the rejection criteria before generating anything, so the standard precedes the temptation of sheer volume.
2. Generate broadly, but cap the field — an unbounded candidate set defeats the very eye that must read it.
3. Cut first by exclusion, removing what fails a stated rule before comparing what survives on finer grounds.
4. Record the reason for each cut, building a written grammar of "no" that future work can reuse instead of rediscover.
5. Revisit the criteria when the same kind of artifact keeps surviving the filter and disappointing in use.

Done consistently, this turns taste into something teachable. The criteria become an artifact in their own right — more durable than any single output they were used to filter, and far easier to hand to the next person.

None of this removes the gut; it disciplines it. The fast pre-analytic call still happens first, but now it is checked against a stated standard and corrected when the two diverge. Over time the written rules and the trained instinct converge, and the practitioner can trust the instinct precisely because it has been audited often enough to earn that trust.

How does abundance quietly erode standards?

The danger of cheap generation is not bad output; it is acceptable output, arriving faster than anyone can rank it. Each individual variant looks fine. Compared only against its near-identical siblings, every choice seems defensible, and the bar slides downward by degrees no one quite notices in the moment.

This is criterion drift, and it is corrosive precisely because it is comfortable. Standards rarely fail in a single dramatic decision. They erode through a thousand small acceptances, each justified by the local fact that the thing in front of you is not, strictly speaking, wrong. Abundance supplies an endless stream of not-actually-wrong, and the stream is patient.

The counterweight is an external reference held apart from the flow of candidates — a fixed sense of excellent that does not adjust itself to whatever the generator happens to be producing this week. Taste that floats with its inputs is not taste; it is an average dressed as a judgment. The whole value of discernment is that it refuses to be one, and refusal takes effort.

There is a practical tell for drift worth watching. When a reviewer finds themselves approving work because nothing is wrong with it rather than because something is right about it, the standard has already started slipping. Excellence is a positive claim; the absence of error is not. A field that conflates the two will keep shipping competent, forgettable output and wonder where its edge went.

Where the binding constraint moves next

As generation continues to commoditize, the premium on judgment will tend to concentrate rather than spread evenly. The roles that hold up are the ones that own a point of view: a defensible sense of what is worth making and the nerve to discard the rest. That posture is harder to copy than any technique, because it is built from accumulated outcomes rather than acquired skills.

Expect the grammar of "no" to become explicit infrastructure — criteria written down, versioned, and argued over, the way code and copy already are. Teams that treat their rejection rules as a shared asset will compound an advantage that pure output cannot match, because the rules travel and the instinct, for a while, does not.

Expect the grammar of "no" to become explicit infrastructure — criteria written down, versioned, and argued over, the way code and copy already are. What gets harder is staying calibrated when the cost of producing more drops to nothing and the social pressure to accept it rises in step. The scarce thing was never the making. Increasingly, it is the patience to look hard at everything that can be made, and the willingness to leave most of it deliberately unmade.