Getting Started with Claude Code: A Researcher's Setup Guide
Economists, political scientists, and sociologists have been quietly adopting Claude Code — not as a novelty, but as what Princeton economist Paul Goldsmith-Pinkham calls "a research assistant that happens to live on your computer." His new Markus Academy series is the first Princeton-affiliated, video-backed tutorial sequence on Claude Code built specifically for empirical researchers, and it comes with a refreshingly honest framing: not "AI wrote my paper in 30 minutes" but "here's why researchers who ignore this are leaving real capability on the table."
The setup guide zeroes in on two trust problems unique to research contexts. First, many datasets contain sensitive or regulated data that should never leave a local machine — which makes Claude Code's local filesystem access both a major advantage over browser-based tools and a genuine compliance risk if misconfigured. Goldsmith-Pinkham's advice is pointed: never run claude from a high-level directory with PII data underneath it, and use --add-dir to explicitly scope context to safe folders only. Second, context window degradation is more dangerous for researchers than for developers — a broken build is visible, but silently wrong regression results are not. The guide recommends clearing context every 45–60 minutes or after any major phase transition (data cleaning, analysis, visualization) to avoid subtle drift accumulating undetected.
The Markus Academy audience — empirical social scientists accustomed to Stata, R, and Python but not terminal-first tooling — represents a large and underserved segment of Claude Code's growing user base. Anthropic's own subscriber data points to accelerating Pro tier adoption outside traditional developer communities, and a Princeton-affiliated series carries the kind of institutional credibility that will circulate through university computing workshops and department Slack channels alike. Part 1 is live now; more installments are forthcoming.
Read the full article at Substack (Paul Goldsmith-Pinkham) →