AI & Research

Using AI Responsibly in Accounting Research

Agentic AI tools are changing how empirical research gets done. This page collects my work on using them well: doctoral teaching, and an open-source toolkit for researchers.

AI agents can compress the mechanical parts of empirical work — literature reviews, data pipelines, replication, drafting — but they do not lower the bar for what counts as evidence. The standards are the same as they have always been: every number verified, every step reproducible, every claim owned by the researcher. My teaching and tools focus on workflows that get the speed without giving up the rigor.

Yale SOM · PhD Intensive

Claude Code for Accounting Research

This hands-on PhD course focuses on agentic AI for empirical accounting research. Five modules of hands-on work incorporate AI into the research pipeline responsibly: the course starts at first install and ends at a verified WRDS-to-Stata pipeline, covering reproducible pipelines, text measures from EDGAR, licensed data on your own account, and the verification protocol that makes the speed usable in a paper.

The course closes with a seminar on the economics of knowledge when machines can produce it. For the consumer, AI-created knowledge is privately cheap but socially valuable only if it can be verified and trusted. For the producer, AI lowers the private cost of research, while the social return depends on quality, congestion, and the allocation of scholarly attention. The seminar asks where these private and social margins diverge in the academic labor market, and what remains scarce.

Course site coming soon
Open Source · Toolkit

Barrios Skills: An AI Agent Toolkit for Research

A free, open-source collection of 40 agent skills and a WRDS data connector for empirical researchers in economics and accounting. The skills cover econometric workflows, academic writing, document preparation, and data visualization; the connector lets agents query Compustat, CRSP, and other WRDS databases with your own account. Works with Claude, Cursor, and Codex.

Explore the Toolkit