AI coding agents such as Claude Code and ChatGPT Codex can now execute large portions of high-quality policy analyses. Yet using these tools effectively requires statistical judgment and discernment, combined with a working knowledge of the capabilities and limitations of agentic AI. This course will teach students to effectively direct AI agents to conduct rigorous policy analysis while verifying, refining, and communicating the results. This course will not teach statistical methods; for that reason, a strong foundation in statistical analysis and causal inference is required. Two sections are offered: one required for MPA/ID students and one available as an elective.
Students will use AI coding agents to work through complete analytical pipelines, from data acquisition and cleaning through modeling, robustness checks, and communication of findings to policymakers. The course emphasizes four cross-cutting skills: structuring effective agentic workflows, managing cognitive debt, verification-first analysis, and reproducibility and documentation. The course will also introduce the use of AI as a research methodology, such as the use of large language models to classify text at scale, though this will not be the primary focus. This is an applied, project-based course; students will complete a capstone analysis using real policy data.
This course does not teach statistical methods. It assumes a strong background in statistical analysis and causal inference and focuses on the use of agentic AI tools to conduct policy-relevant quantitative analysis. Students without this background should complete the quantitative methods sequence (API-201/202M/203M or API-209/210, or equivalent) before enrolling.