How Do Scientists Use Claude Code?

  • Author/Source: Charles Yang (Renaissance Philanthropy), via Republic of Science Substack (2026-03-05)

  • Original: https://republicofscience.substack.com/p/how-do-scientists-use-claude-code

  • Key Ideas

  • As of February 2026, approximately 2.1% of scientists with ORCID-linked GitHub profiles use Claude Code; extrapolating current growth rates suggests ~10% by end of 2026.
  • Adoption follows a U-shaped curve by seniority: early-career scientists and senior (post-tenure) scientists adopt at higher rates than mid-career researchers (3-10 years since first publication), likely reflecting higher risk aversion during the tenure-track period.
  • Economists and social scientists are the highest adopters by field (up to 3.4%), while mathematics and environmental science are the lowest (~1.4%). Adoption is surprisingly uniform across fields.
  • U.S.-based Tier 1 universities lead institutional adoption; most scientific institutions have zero public Claude Code users.
  • Scientists primarily use Claude Code for Python, R, and shell scripts --- languages common in scientific computing --- and are more likely to work across multiple repos than non-scientist users.

  • "Winning the AI for Science race" is framed as downstream of getting STEM graduate students to use the best AI coding tools; hackathons are highlighted as an effective adoption mechanism.

  • Summary

Yang presents the first empirical measurement of Claude Code adoption among scientists, leveraging the fact that Claude Code commits are default co-authored (making them identifiable in Git history) and using ORCID-GitHub profile links to identify active researchers. The study identified 15,933 scientists from ~14 million ORCID profiles who maintain active GitHub accounts and recent peer-reviewed publications, of whom 331 (2.1%) had made Claude Code commits by February 2026.

The analysis reveals several patterns. Adoption is growing steadily but remains low in absolute terms, with usage tied to academic cycles (a notable bump in late January 2026 as new semesters began). The U-shaped seniority curve suggests that mid-career researchers --- precisely those who might benefit most from productivity tools --- face the strongest barriers to adoption, whether from risk aversion, time constraints, or institutional culture. By field, economists and social scientists lead adoption, which may partly reflect ORCID usage patterns but is consistent with the broader narrative of economics being an early-adopter discipline for AI tools.

Yang frames the policy implications starkly: accelerating AI coding tool adoption among scientists should be a "national imperative." He highlights hackathons as effective adoption drivers and notes that enterprise access (per-seat token limits, institutional agreements) remains a significant barrier that AI labs have not yet cracked. The countries whose graduate students internalize these tools first will compound that advantage for years.

  • Relevance to Economics Research

This article provides the only empirical data on AI coding tool adoption among researchers, and the finding that economists are the highest adopters is directly relevant. For economics departments and individual researchers, the data suggests that adoption is still very early --- there is significant first-mover advantage available. The U-shaped seniority curve has implications for faculty hiring and PhD training: departments that facilitate AI tool adoption for mid-career faculty and graduate students may gain a meaningful research productivity edge. The enterprise access barrier is also relevant for universities negotiating institutional AI subscriptions.