About & Attribution

About this course

Claude Code for Accounting Research: AI Agents for the Empirical Researcher is a five-module, hands-on intensive plus a closing half-day seminar. It is built for accounting PhD students and faculty with mixed Stata/Python backgrounds, and it assumes no experience with agentic AI tools. It teaches Claude Code as a research assistant that runs in the terminal. It reads and writes files, runs your existing code, and iterates. The course spends as much time on verification as on capability, because faster output is of no use if you can’t trust it. The five build modules cover foundations and the tool ladder; an empirical loop on free SEC data; EDGAR text as a measured panel, plus a first Skill; WRDS at scale through an MCP server with a Stata bridge; and writing, safety and verification as a capstone. The closing seminar asks how an agent this capable changes the way accounting research gets produced, and which parts of the job stay with the researcher.

Instructor

John Barrios, Yale School of Management.

Attribution

Original authorship. The notes, labs, slides, and figures on this site are original teaching material written for this course. Where this course draws on an external idea, technique, or example, it is cited in the text or listed on the Resources page rather than reproduced.

Goldsmith-Pinkham / Markus Academy series. Several of this course’s teaching examples and prompt patterns were inspired by Paul Goldsmith-Pinkham’s “Claude Code for Economists” series for the Markus Academy (Ep. 162).

Barrios Skills collection. The Skills catalog is a snapshot of the Barrios Skills collection, included under its MIT license (reproduced in full at skills/LICENSE.txt and linked from the catalog index). The live collection at barrios88.github.io/barrios-skills is the canonical source and is updated regularly. The copy on this site was taken when the course was built and is not updated.

Figure and exhibit policy. Figures on this site (the tool ladder, the empirical loop, the verification-debt 2×2, and the rest) are original diagrams drawn for this course. A small number of exhibits show what a Claude Code or Claude Desktop session looks like in practice, for example the Module 1 terminal-anatomy and permission-prompt exhibits. These are mock-ups of a typical interaction, not screenshots or transcripts of a real session. Each caption says so.

External links. The links on the Resources page point to real, existing projects. That includes the one item flagged as unvetted rather than recommended.

No Yale trademarks. This is a personal course site by a Yale faculty member, not an official Yale or Yale School of Management site.

Contact

For questions about the course or its materials, or to ask about reusing or adapting something on this site, contact John Barrios through his Yale School of Management faculty listing.