Claude Code 16: The Memory Foam Mattress Theory of Claude Code

  • Author/Source: Scott Cunningham (Baylor University), via Substack ("Causal Inference")

  • Original: https://causalinf.substack.com/p/claude-code-16-the-memory-foam-mattress

  • Key Ideas

  • Claude Code is "endogenous software" — like a memory foam mattress, it conforms to the user rather than requiring the user to learn fixed rules
  • There is no real onramp or starter kit for Claude Code; other people's workflows reflect their own shape, not yours
  • The current writing about Claude Code is dominated by incumbents (software engineers), leaving the extensive margin of social scientists underserved
  • Three economics concepts frame adoption: incumbents vs. entrants, average vs. marginal user, intensive vs. extensive margin
  • The /insights command analyzes six weeks of work patterns and produces a personalized workflow report

  • Story and demonstration beat documentation for helping new users adopt Claude Code

  • Summary

Cunningham argues that Claude Code is fundamentally different from any prior software because it is endogenous — it adapts to the user's working style rather than imposing a fixed set of rules to learn. He draws on three economics concepts (incumbents vs. entrants, average vs. marginal user, intensive vs. extensive margin) to explain why most existing guides and starter kits fail new adopters: they reflect the workflows of software engineers who were already there, not the needs of social scientists just arriving.

The central metaphor is a memory foam mattress. Just as memory foam conforms to the sleeper's body over time, Claude Code learns and adapts to each user's unique style of work. This means there is no transferable "correct" configuration — only the shape that formed around you. Cunningham shares his own /insights report, generated after an estimated 1,642 hours of use, as evidence that the tool produces deeply personalized workflow patterns. He concludes that Claude Code cannot be taught; it can only be learned, and the best way to learn is to start using it on your own directories and tasks, trusting that it will figure you out.

  • Relevance to Economics Research

The essay reframes AI tool adoption through familiar economics concepts (extensive margin, experience goods, incumbent advantages) and argues that the barrier to adoption for applied social scientists is not technical skill but the alien nature of endogenous software. For researchers whose work lives in directories — papers, data, slides — Claude Code offers transformative potential, but only through direct sustained use rather than following others' workflows.