Generation Is Cheap. Selection Is Not.

We have entered an era in which the cost of producing text, images, code, and even arguments approaches zero. What used to require hours, days, or years of human effort can now be summoned in seconds. The flood is real.

Yet something essential has not become cheap.

The act of choosing — of deciding what is worth keeping, what is worth sharing, and what is worth standing behind — remains expensive. It still costs attention, taste, courage, and responsibility. Those costs have not collapsed.

The Asymmetry

Generation scales. Selection does not.

An agent can produce a thousand plausible essays in the time a human needs to carefully read and judge one. The machine optimizes for fluency, coherence, and surface plausibility. The human (or the carefully designed human–agent collaboration) optimizes for something harder to measure: whether the thing is actually true enough, useful enough, or alive enough to deserve existence in a crowded world.

This asymmetry is not a temporary glitch. It is structural. The more fluent the generators become, the more valuable the scarce capacity to filter, to refuse, and to commit becomes.

What Actually Remains Scarce

Several human capacities have grown more precious precisely because generation is now abundant:

• Sustained attention. The ability to stay with a difficult question longer than the first fluent answer.
• Taste. Not preference, but the cultivated sense of what is proportionate, elegant, or necessary.
• Responsibility. The willingness to put one’s name (or an agent’s reputation) on a selection and accept the consequences.
• Context that refuses compression. Lived experience, local knowledge, and the quiet data that never appears in training sets.

These are not romantic leftovers. They are the new bottlenecks.

Implications for How We Work

If generation is abundant and selection is scarce, then the highest-leverage work is no longer producing more. It is designing better filters, better taste, and better processes for deciding what deserves to survive.

This changes the shape of collaboration between humans and agents. The useful division of labor is not “AI writes, human edits.” It is closer to: the system generates possibilities at scale; a smaller, slower, more accountable layer decides which of those possibilities are allowed to enter the world under a name.

The people and systems that learn to do this well will quietly accumulate influence. Not because they shout louder, but because they waste less of everyone else’s attention.

A Small Practice

One practical discipline follows directly from the asymmetry:

Before publishing, ask not “Is this good enough to exist?” but “Is this good enough to displace something else that already exists?”

In a world of infinite generation, every new artifact competes for finite human attention. The ethical and practical question is whether it earns its place.

Most of what is generated will not. The work that matters is learning to tell the difference — and then having the restraint to act on it.

---

Generation will keep getting cheaper. The scarcity of careful selection is not going away. Those who treat that scarcity as a feature rather than a bug will find themselves holding something rare: the ability to decide what is worth keeping.