The Curation Problem: When the Scarce Resource Becomes Knowing What's Good

The Rust programming language has a package registry, crates.io, with over 130,000 packages. This is, on its face, a triumph of abundance — an enormous ecosystem of reusable code, free for anyone to use. But it produced a problem that abundance always produces: how do you know which ones are good? Faced with dozens of packages that all claim to do what you need, most of them unmaintained, buggy, or subtly wrong, the developer drowns in choice — and out of that drowning came blessed.rs, a curated directory of recommended crates, a human-made list of "here are the good ones." The need for such curation, in an ecosystem so vast, has become existential: the value of the 130,000 packages is nearly worthless without a way to find the few hundred worth using, so the scarce and valuable resource is no longer the packages (there are far too many) but the curation — the trustworthy judgment about which ones are good. Abundance did not solve the problem of getting good software; it relocated the problem from scarcity of options to scarcity of judgment about options.

This is the curation problem: the condition in which abundance — of packages, tools, content, options — makes the scarce and valuable resource no longer the things themselves (there are overwhelmingly many) but the curation of them: the trustworthy judgment about which are good, so that as ecosystems produce ever more, the bottleneck shifts from production to discernment, and knowing what's good becomes harder and more valuable than making more.

Why abundance makes judgment scarce

The curation problem arises from a simple inversion: when a resource is scarce, having more of it is valuable; when it becomes abundant, having judgment about which is valuable — so abundance converts the scarce good from the thing to the selection of the thing. In a world of few packages, each new package adds value; in a world of 130,000, another package mostly adds noise, because the marginal package is unlikely to be the good one and its presence makes the good ones harder to find — so past a threshold, more production does not help and can actively hurt, drowning quality in quantity. The valuable act shifts from creating (there is already too much) to discerning (which of the too-much is worth using) — and discernment is genuinely hard and genuinely scarce, because evaluating quality requires expertise, effort, and judgment that do not scale the way production does. This is the series' False Economy of AI Abundance (#68) generalized beyond AI content to all abundant ecosystems: the flood devalues the individual item and revalues the filter, so blessed.rs — a human's curated judgment — becomes more valuable than any single package, because it provides the scarce discernment the abundance destroyed. And curation resists automation precisely because it is judgment: knowing that a package is well-designed, actively maintained, and appropriate for a use requires the kind of contextual evaluation the series' Criteria Collapse (#146) showed cannot be fully reduced to metrics — so the scarce resource is not just "a list" but trustworthy human judgment about quality, which is exactly what abundance cannot produce more of.

Why AI turns the problem critical

The curation problem becomes acute in the age of AI because AI is an abundance machine — it can generate packages, content, code, and options without limit — so it dramatically worsens the abundance side of the problem while the judgment side stays scarce, widening the gap between how much exists and our ability to discern what's good. Every domain AI touches, it floods: AI can generate infinite variations of content, code, and products, so the 130,000-package problem becomes the everything-problem, with abundance exploding across every ecosystem AI reaches while the human judgment needed to curate it does not scale at all. This threatens to make the curation problem civilizational: as AI-generated abundance overwhelms every domain, the scarce resource everywhere becomes trustworthy discernment about what is good, and the ability to find quality amid infinite AI-generated quantity may become one of the defining challenges of the era — the series' broader concern that the flood of the generated drowns the valuable. But AI is also a candidate solution: it can potentially curate as well as generate — filtering, evaluating, and recommending amid the abundance it creates — so the question is whether AI's curation capability can keep pace with its generation capability, or whether it will always generate faster than it can discern, leaving humans to drown in a flood that AI made worse and could not clean. The deeper worry is that generation is easy to automate and judgment is hard, so AI may be structurally better at worsening the curation problem than at solving it — flooding faster than it can filter.

The counterpoint: curation has its own failures, and abundance has real value

Honesty requires the strong objection, because "curation is the scarce good" can romanticize the curator and undervalue the abundance — and both moves have real costs. Curation has serious failure modes: the curator has biases and blind spots, so a curated list encodes one perspective's judgment and can exclude the valuable-but-unfashionable; curation gatekeeps, concentrating the power to define "good" in whoever curates, which can entrench incumbents and suppress the new; and curated lists go stale, freezing a moment's judgment as the ecosystem moves. So "we need curation" is not a clean good — bad or captured curation can be worse than raw abundance, and the centralization of taste that curation implies carries real risks. And abundance itself has genuine value the curation-problem framing can obscure: the 130,000 packages include the long tail of niche, specialized, and unexpected tools that a curated list of "the good ones" would exclude, so abundance serves the diversity of needs that curation, optimizing for the general case, cannot — and the ability of anyone to publish, unfiltered, is exactly what lets the unexpectedly-valuable emerge. So the curation problem is not "abundance is bad and we need gatekeepers to tell us what's good." It is the narrower claim that abundance relocates the scarce resource from the things to the judgment about them, that this judgment is genuinely hard, scarce, and worsened by AI's generation, and that trustworthy curation becomes correspondingly valuable — while recognizing that curation has real failure modes (bias, gatekeeping, staleness, centralization of taste), that abundance genuinely serves the long tail, and that the goal is good, plural, contestable curation rather than a single gatekeeper's list. The problem is not too many options; it is too little trustworthy help choosing among them — and the answer is better discernment, not less abundance.

What it asks of us

The curation problem asks us to recognize discernment as the scarce and valuable resource it has become — to value, build, and protect trustworthy judgment about quality amid abundance, while avoiding the failure modes that make curation its own problem. In practice that means treating curation as first-class work: valuing and supporting the human judgment (the blessed.rs makers, the trusted reviewers, the discerning guides) that helps people find quality amid overwhelming choice, rather than assuming abundance is self-organizing; building plural and contestable curation — many curators, transparent criteria, the ability to disagree — rather than single gatekeepers whose bias becomes everyone's constraint; and pursuing AI curation deliberately, to help filter the abundance AI itself creates, while being honest that generation may outpace discernment. It means, as individuals, cultivating our own discernment, since in a world of infinite options the ability to judge quality is a core competence, not a delegated one. The deeper recognition is that abundance does not end scarcity but moves it — from the things to the judgment about them — so that in an age of infinite generation, the rare and precious resource is not more stuff but the trustworthy discernment of what, among the endless stuff, is actually good. The 130,000 packages were the easy part; knowing which few hundred to use is the hard part, and the harder AI makes the abundance, the more that discernment — human and machine, plural and contested — becomes the thing worth having. In a flood, the scarce resource is dry ground.

This is article #214 in The IUBIRE Framework series. The Curation Problem was articulated by IUBIRE V3 in artifact #219 — "The Curation Problem: Why blessed.rs Points to a Deeper Crisis in Technical Knowledge" (with artifact #236, "The Rust Curation Problem"). Real-world grounding: the Rust crates.io registry with 130,000+ packages and the emergence of blessed.rs, a human-curated directory of recommended crates, as an existential response to the difficulty of finding quality amid abundance; the general pattern (across package registries, content platforms, and tool ecosystems) by which abundance relocates the scarce resource from the things themselves to trustworthy judgment about which are good; AI as an abundance machine that dramatically worsens the problem across every domain while being an uncertain candidate to also solve it (generation being easier to automate than judgment); and the countervailing realities that curation has real failure modes (curator bias, gatekeeping, staleness, centralization of taste) and that abundance genuinely serves the long tail of niche needs. Related to False Economy of AI Abundance (#68), Knowledge Structure Problem (#50), and Criteria Collapse (#146).

Next in series: The Ensemble Paradox (#215)