Claude Code As the Agentic RA: Using AI Coding Agents to Democratize Quantitative Methods In Social Work Research

Social work researchers face persistent methodological barriers, such as limited research assistant budgets and steep learning curves for statistical software. These constraints disproportionately affect faculty and doctoral students with resource-limited infrastructure. Agentic AI coding tools, autonomous software agents that write, execute, and debug code through natural language instruction, represent an emerging methodological innovation for quantitative social work research. Unlike conversational AI, agentic tools operate within a researcher’s computing environment, acquiring data from public APIs, cleaning datasets, running statistical models, and generating publication-ready scripts in Jupyter Notebook, R, Stata, or Python. Central to this workshop is the principal investigator model, where these tools functionally resemble research assistants capable of executing graduate-level tasks.

Participants will: (1) distinguish between conversational AI and agentic AI coding tools and their appropriate applications across the research lifecycle; (2) develop structured analytic plans suitable for delegation to an AI coding agent; (3) observe live, end-to-end secondary data analyses using multiple publicly available sources, such as the U.S. Census Bureau American Community Survey API, CDC WONDER, and Bureau of Labor Statistics, conducted via natural language prompting; and (4) critically examine ethical considerations including data privacy protocols, AI disclosure in scholarly manuscripts, and frameworks for distinguishing delegated analytic tasks from intellectual contributions.

This workshop uses an experiential design across three phases. Phase I establishes a conceptual and ethical foundation through structured discussion. Participants locate themselves on an AI adoption taxonomy, ranging from abstinence to prescription, and examine what constitutes “good enough” performance when the benchmark is research assistant task execution rather than full researcher replacement. The principal investigator model is introduced, paralleling how researchers already navigate delegation and the distinction between tasks and intellectual contributions. Phase II is a live demonstration of a complete secondary data analysis. Through the development of a structured abstract and analytic plan, participants will observe the AI agent execute each methodological step in real time: data acquisition from multiple federal APIs, variable construction, dataset merging, cleaning, descriptive statistics, bivariate analysis, and multivariate modeling, including OLS regression and machine learning with SHAP interpretation. Phase III addresses data privacy and responsible implementation. Participants examine privacy protocols, learn safeguards such as synthetic datasets and data shells, and observe a demonstration of a local, private-by-design alternative. The session closes with manuscript disclosure language and planning for tool integration.

This workshop addresses the conference theme of advancing rigorous inquiry for the common good by introducing an emerging methodology that lowers barriers to reproducible, sound research. By equipping scholars with skills in AI-augmented analysis, the session supports broader participation in data-driven social work research, particularly for those without dedicated support. Participants leave with practical strategies for integrating these tools while maintaining the oversight, transparency, and ethical rigor that social work scholarship demands.