Before the Algorithm There Was the Index: How the Card Catalog Solved a ‘New’ Problem

The problem every founder describes as new—too much information, too little signal, decisions made on noise rather than evidence—is not new. It was solved, with considerable elegance, in the late nineteenth century. The solution was a wooden cabinet full of three-by-five inch cards. Understanding how it worked, and why it was eventually abandoned, is one of the more useful pieces of intellectual history available to anyone who builds things for a living.

Why the “new” crisis is ancient

Every generation believes it invented information overload. None of them did. The story of organizing information runs from the layout of papyrus scrolls at the Library of Alexandria and playing cards with notes on the back that served librarians during the chaos of the French Revolution, all the way to the doorstep of the digital information retrieval we use today. The pattern is consistent: a surge in the volume of recorded knowledge outruns the retrieval systems built to manage it, someone engineers a new architecture, and the crisis recedes—until the next surge.

When the Library of Congress assumed copyright registration and deposit responsibilities in 1870, the huge increase in the number of books and other items acquired by the Library rendered the most recently published catalogs quickly obsolete. The institution faced, in miniature, exactly what a Series A startup faces when its Slack channels, dashboards, newsletters, and investor updates begin generating more inputs than any leadership team can process: the retrieval problem. The question is never “do we have the information?” It is always “can we find the right piece at the moment we need to decide?”

Information overload has been defined as the exceeding of processing capabilities given an amount of information, specifically occurring when information-processing requirements surpass an individual’s cognitive capacity. That definition, drawn from contemporary cognitive science, describes what nineteenth-century librarians were solving with wood and paper. The medium changes. The constraint does not.

What the card catalog actually was—and was not

Most people remember the card catalog as a storage device. That is the wrong frame. The card catalog evolved out of the need for a standardized system to manage rapidly expanding libraries, serving as both a repository for data and a search tool in a pre-digital age. The distinction matters enormously. A repository is passive. A search tool is active—it is built around the moment of retrieval, not the moment of deposit.

Since 1873, Melvil Dewey had been working at the Amherst College Library to develop a system for classifying bibliographic works and organizing card catalogs. He published the first edition of what we now know as Dewey Decimal Classification in 1876. The innovation of Dewey’s system was the categorization of published works into knowledge subjects (“classes”) represented by a number (“decimal”).¹ The Decimal Classification introduced the concepts of relative location and relative index, which allow new books to be added to a library in their appropriate location based on subject. In operational terms: every new piece of information was classified at the point of entry, not at the point of search. The cognitive work happened upstream.

The ambition scaled further. Paul Otlet, a Belgian lawyer and documentation pioneer, sought to organize and interconnect the world’s knowledge through systematic indexing and classification. Collaborating with Henri La Fontaine, he established the International Institute of Bibliography in 1895, which evolved into the Mundaneum, an ambitious repository intended to catalog all human knowledge on millions of index cards.² The Répertoire Bibliographique Universel steadily grew to 13 million index cards in 1927; by its final year, 1934, it had reached more than 15 million. The effort met with early success, even attracting a healthy business in mail-order research services, in which users would submit search queries for a fee. This was, in effect, a search engine—operated by human indexers, powered by classification logic, and constrained by the discipline of the query. You had to know what you were looking for before you asked.

That constraint was not a limitation. It was the feature. The card catalog system was fundamentally designed around known-item searching. You knew what you were looking for: a specific author, a specific title, or a specific subject. The catalog helped you determine if the library owned it and where to find it. The card catalog forced you to form a question before it gave you an answer. Every modern feed-based platform does the opposite: it supplies answers before you have formed a question, and then optimizes to keep you consuming answers indefinitely.

The algorithm’s structural problem

The term “filter bubble” was coined by internet activist Eli Pariser circa 2010. Pariser’s book The Filter Bubble (2011) predicted that individualized personalization by algorithmic filtering would lead to intellectual isolation and social fragmentation.³ The prediction has been debated empirically—there is scant empirical evidence for the existence of filter bubbles in their most extreme form, and search and social media users generally appear to encounter a more diverse media diet than often assumed—but the debate misses the more important operational point for founders.

The problem is not primarily that algorithms show you the wrong things. The problem is that they show you things before you have decided what you need. Recommendation systems, by tailoring content based on user preferences, inadvertently contribute to the creation of digital environments where users are predominantly exposed to viewpoints that align with their own.⁴ For a citizen, this is a democratic concern. For a founder making capital allocation decisions, it is a cognitive architecture problem: the feed is optimized for engagement, not for the quality of the decision that follows from it.

In terms of decision behavior, information overload can occupy numerous cognitive resources and damage decision quality. Many studies show that when people are given more information than they can handle, they suffer from negative effects such as confusion, poor judgment, and bad decisions. The feed does not cause this by malice. It causes it by design: engagement metrics reward volume and recency, not relevance to the specific decision in front of the user. A founder scrolling a LinkedIn feed before a pricing decision is not doing research. They are depleting the cognitive budget they need to make the decision well.

Simon’s theorem and why it matters for operators

The theoretical foundation for all of this was laid not by a librarian but by an economist. In honor of the 75th anniversary of Herbert Simon’s seminal book Administrative Behavior, first published in 1947, the work was recognized as “epoch-making” by the Royal Swedish Academy of Sciences, which awarded Simon the Nobel Prize in Economics in 1978.⁵

Simon emphasized the limits to rationality that real-life administrators face with regard to memory, attention, and capacity. He argued that decision-makers face three key constraints: limited knowledge of alternatives and outcomes; cognitive limitations in processing complex information; and constraints on time and computational resources. His response to this was the concept of satisficing—a portmanteau of “satisfying” and “sufficing.” Satisficing is the strategy of considering the options available to you for choice until you find one that meets or exceeds a predefined threshold—your aspiration level—for a minimally acceptable outcome.

Simon’s insight was not that humans are irrational. It was that rationality is bounded by the architecture of the information environment. Change the architecture, and you change the quality of the decisions that emerge from it. The card catalog was, in Simon’s terms, a satisficing machine: it gave you enough structured information to make a good decision about where to look next, without overwhelming you with everything the library contained. The algorithm is the opposite: it gives you everything, optimized for the metric that keeps you on the platform, and leaves the satisficing work entirely to you—at the worst possible moment, when your cognitive reserves are already depleted.

A consistent finding emerges across diverse domains, including clinical medicine, financial decision-making, public policy, and experimental psychology: performance degradation occurs predictably when cognitive demands exceed working memory capacity. Founders are not exempt from this finding. As a startup founder, there is a high volume of choices you have to make daily—perhaps hourly. With a continuous high volume of decisions involved, each one may seem increasingly difficult.⁶

What the card catalog’s architecture actually looked like—and how to replicate it

The card catalog had five structural properties that made it work. Each has a direct operational analogue for a modern founder or operator.

1. Classification at the point of entry, not the point of retrieval. Knowledge management taxonomy is a systematic classification framework that organizes content using controlled vocabulary, hierarchical relationships, and metadata—making information findable regardless of which tool it lives in, who created it, or when it was written. In practice: every piece of information that enters your decision environment should be tagged at the moment it arrives—by decision type, by time horizon, by reversibility. Unclassified inputs are noise.

2. Controlled vocabulary. The subject headings, controlled by the Library of Congress, imposed a particular intellectual organization on knowledge. Related subjects were kept together, but the relationships were hierarchical and relatively rigid. The rigidity was not a bug. It prevented the proliferation of synonyms that makes modern knowledge bases unsearchable. For a founder, this means committing to a fixed set of decision categories—market, product, people, capital—and refusing to let every new framework generate a new taxonomy.

3. Standardized metadata on every card. Every book in the collection had a standardized card listing, relevant metadata, and cross-referenced topics. The equivalent for a founder is a decision log: a record of what was decided, on what evidence, by whom, and with what expected outcome. Without this, the organization relitigates the same decisions repeatedly, consuming cognitive resources that should be reserved for new problems.

4. Tiered access by decision type. The catalog distinguished between author, title, and subject searches—three different entry points for three different types of questions. A useful founder rule is simple: if a decision is reversible, make it fast. If it is expensive to reverse, slow down. The card catalog never mixed these categories. Neither should a founder’s information diet.

5. Batched, intentional retrieval. The Library of Congress card catalog grew to more than 22 million cards in 22,000 drawers before the last card was added in 1980.⁷ Nobody browsed 22 million cards. They arrived with a question, retrieved what they needed, and left. The most immediately implementable structural solution to decision fatigue is decision batching: grouping all decisions of a similar type into a single weekly or daily block, rather than addressing them as they arrive. The cognitive cost of decisions is not just in the decision itself—it is in the context-switching required to address a decision from a different domain.

The transition that cost something

The card catalog was retired not because it failed but because digital systems offered capabilities it could not match. Digital catalogs offered obvious advantages: keyword searching across multiple fields, remote access from home computers, real-time availability information, the ability to search millions of records simultaneously rather than being limited to a single library’s holdings. These are genuine gains. But something was lost in this transition that goes beyond mere nostalgia for analog technology. The card catalog was not simply a less efficient version of the online catalog. It was a fundamentally different kind of information system, one that encoded particular ways of thinking about knowledge and encouraged certain research behaviors.

The behavior it encouraged was intentional retrieval. The behavior the open web encourages is ambient consumption. These are not equivalent. As Pariser argued, we will increasingly each live in our own unique information universe. Our past interests will determine what we are exposed to in the future, leaving less room for the unexpected encounters that spark creativity, innovation and the democratic exchange of ideas. For a founder, the loss of unexpected encounters is not primarily a democratic problem. It is a strategy problem: the signal that would have changed your mind about a market, a hire, or a product direction is being filtered out by an algorithm that has learned you prefer confirmation.

OCLC printed its last library catalog cards on October 1, 2015, ending an era that lasted more than 150 years. The era ended. The underlying design logic did not become obsolete. The card catalog’s lesson is not “go analog.” It is “design your information environment around the decision, not around the feed.”

What this means

The forward-looking close

The next wave of AI-assisted tools will not solve the information architecture problem by default. They will intensify it. Generative AI surfaces more plausible-sounding inputs faster than any previous technology; without a classification system and retrieval discipline already in place, the result is not better decisions—it is faster noise. The founders who will navigate this well are not the ones who consume the most, or the ones who trust the algorithm to curate for them. They are the ones who, like the librarians of the 1870s facing a copyright-deposit avalanche, ask the structural question first: what is the retrieval system, and is it built around the decision or around the feed?

Business Growth Accelerator (a FounderWise brand) works with founders and operators on exactly this kind of decision architecture—the systems, criteria, and information structures that determine whether a company’s cognitive resources are spent on signal or on noise. The card catalog did not make knowledge easier to accumulate. It made knowledge easier to use. That is the standard worth recovering.