Agentic Coding Tools
for Researchers

Abstract

Agentic coding tools operate directly in your codebase, execute multi-step tasks, and maintain context across a project. Researchers can use them for ideation, downloading and cleaning data, running analyses, drafting papers, and many other tasks that make up the daily work of empirical research. This four-day workshop introduces the tools (primarily Claude Code, with brief comparison to Cursor and Codex), what makes them distinct from conversational AI, how to set up a working environment with configuration files like CLAUDE.md and agents.md, how to extend them with plugins and custom skills, and how to work with them responsibly: breaking research tasks into structured plans, validating those plans before execution, and writing tests that verify agent output. Participants apply what they learn to projects using real data and tasks drawn from empirical social science.

Outline

1. § i.
   Foundations & first real workflow. Set up a working environment, generate your own CLAUDE.md, clean a real dataset with verification checks built into the script.
2. § ii.
   From data to compiled output. Run regressions, build publication-quality tables and figures, compile a Markdown or LaTeX document end-to-end.
3. § iii.
   Customizing & extending. Build a multi-level CLAUDE.md, persistent memory, and custom commands; install community skills and write your own.
4. § iv.
   Applied research & capstone. APIs, drafting prose, and ninety minutes of focused work on your own research project.

Participate

Registration is administered by the Consortium for the Advancement of Research Methods and Analysis. Tuition and seat counts are set by the consortium.

Instructor

Justin Frake is Assistant Professor of Strategy at the University of Michigan’s Ross School of Business, an Associate Editor at Management Science, and a Senior Editor at Organization Science. He studies how individual perceptions and behaviors produce organizational outcomes; his research topics include strategic human capital, misconduct, and causal inference.

Mailing list

Workshop materials, errata, and future cohort announcements. Infrequent. Unsubscribe at any time.