Claude Code Series (Part 10): Producing Highly Effective Decks for My Data Science Class

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

  • Original: https://causalinf.substack.com/p/claude-code-series-part-10-making

  • Key Ideas

  • Lecture deck creation via exhaustive dictation -- talking ideas, objectives, pedagogy, and audience into existence rather than "vibe coding"
  • Claude Code created an original .sty Beamer style file on the fly when asked to change aesthetics mid-session
  • Claude Code produces ambitious TikZ graphics (e.g., a filing cabinet) from anecdotal descriptions without explicit requests
  • The marginal cost curve of deck production shifts down and flattens while the marginal benefit curve shifts up, so time-adjusted quality rises
  • Optimal deck production is where MB = MC across slides; Claude Code shifts both curves favorably, making the equilibrium quality higher

  • The 1h40m video demonstrates the full live-coding process for an undergraduate data science class at Harvard

  • Summary

Cunningham records and narrates a 1-hour-40-minute live session of building Beamer lecture slides for an undergraduate data science course using Claude Code. He emphasizes that the process is dictation -- continuously talking through ideas, objectives, audience, and pedagogical beliefs while iterating on slide content, order, and aesthetics. Claude Code does not merely "know LaTeX"; it has absorbed tacit knowledge about what makes great presentation decks work, including narrative flow, cognitive density, and visual rhetoric.

A highlight is Claude Code spontaneously creating a custom .sty style file when asked to change the deck's look, and producing complex TikZ graphics from loose verbal descriptions. Cunningham frames the productivity gain through an economic lens: Claude Code shifts the marginal benefit curve of deck-making upward (higher quality per unit of time) and the marginal cost curve downward and flatter (less strain at higher effort levels). The net effect is ambiguous on total time spent but unambiguously positive on output quality. He argues this matters especially for "modal academics" who find deck creation tedious and whose slides are therefore uniformly mediocre.

  • Relevance to Economics Research

The post provides a concrete workflow for economists who rely on Beamer presentations for seminars, job talks, and lectures. The economic framing of the productivity shift (MB/MC analysis of deck creation) is itself a useful pedagogical device. The demonstration that an AI agent can handle TikZ, custom styling, and narrative structure lowers the barrier for researchers who want higher-quality visual communication but lack design skills or patience.