The Failure of the Filing Cabinet Metaphor

Modern digital documentation relies on a 1970s desktop metaphor—folders and hierarchies—that assumes knowledge exists in a single, logical location. This model is fundamentally flawed because complex information, such as accessibility decisions, rarely belongs to one category. It intersects across design, engineering, and support, forcing creators to choose an arbitrary 'home' for documents, which inevitably makes them harder to find for others.

Information Foraging vs. Hierarchical Browsing

Drawing on the theory of Information Foraging by Peter Pirolli and Stuart Card, the author argues that users do not systematically browse hierarchies. Instead, they act like foragers, following cues and signs until they find what they need. When documentation is buried deep within a tree structure, users abandon the search, ask colleagues, or create duplicates. The 'correct' location is often only obvious to the person who created the folder, not the person trying to retrieve the information.

Designing Knowledge Graphs for Humans and AI

AI retrieval systems have exposed the limitations of folder-based storage by prioritizing meaning, context, and relationships over physical location. To build effective documentation, teams should treat their systems as knowledge graphs rather than isolated assets.

Key strategies for building discoverable knowledge include:

• Multiple Entry Points: Ensure information is reachable via search, metadata, tagging, cross-linking, and semantic relationships.
• Non-Linear Connections: Use tools that emphasize links and tags (like Obsidian) to show how ideas intersect, rather than forcing them into a single hierarchy.
• Accessibility Principles: Apply the same logic used in accessible design—never rely on a single path to communicate meaning. If a user can only find information through one specific navigation route, the system is fragile.

Ultimately, the goal is to stop obsessing over the 'perfect' storage location. By focusing on clear structure, meaningful headings, and consistent metadata, designers can create a body of knowledge that is equally discoverable for humans and machine-learning models.