Resources

These are roughly in order from most to least useful for someone starting to experiment with agentic AI.

Start here: “Research in the Time of AI.” From Paul Goldsmith-Pinkham, an economist at Yale University. Discussion of research slop, how AI compresses the research timeline, and why checking the work becomes the hard part. Read it before you point an agent at your own paper: when doing the work gets cheap, checking it becomes the job.

“An opinionated guide to which AI to use to do stuff.” From Ethan Mollick, a professor of management at the University of Pennsylvania. Discussion of choosing an AI, chatbots versus agents, giving an AI its own computer, and giving an AI your computer.

“Claude Blattman: AI for Professionals Who Don’t Code.” From Chris Blattman, a political economist at the University of Chicago. Discussion of getting started without coding experience, installing Claude Code, CLAUDE.md files, and skills for managing a research workload.

“Thinking with Agents.” From Erkmen G. Aslim and Emily Beam, economists at the University of Vermont. Discussion of the terminal and git for beginners, chat versus agentic tools, context windows, reusable skills, and auditing code and papers.

“Agentic AI for Social Science Research.” From Bogdan G. Popescu, a political scientist at Tec de Monterrey. A series of tutorials. Discussion of what an agent is, installing Claude Code, permissions and data privacy, CLAUDE.md files, git, and skills.

From Claes Bäckman, an economist at the Leibniz Institute for Financial Research SAFE.

From Paul Goldsmith-Pinkham, an economist at Yale University.

From Yusaku Horiuchi, a political scientist at Florida State University.

“AI Accountability and Transparency Policy for Authors.” From the American Journal of Political Science, the journal of the Midwest Political Science Association. Discussion of which uses of AI an author must declare, including writing code and analyzing data, and what the declaration must report.

“Reflections on Vibe Researching.” From Joshua Gans, an economist at the University of Toronto. Discussion of a year-long AI-first research experiment, mistakes that survived to peer review, and carrying more mediocre ideas to completion.

“Agentic AI and Social Science Research Practice.” From Tom Pepinsky, a political scientist at Cornell University. Discussion of using agents for tasks that follow rules but not for answers or interpretations, why an agent’s citations go wrong when its code doesn’t, and AI-written journal submissions.

“The 100x Research Institution.” From Andrew B. Hall, a political economist at Stanford University. Discussion of an independent audit of Claude Code’s replication and extension of one of his published papers, what it got wrong, and what checking it cost.

“AI Coding Agents in Social Science: Methodologically Diverse, Empirically Consistent, Interpretively Vulnerable.” By Meysam Alizadeh, Fabrizio Gilardi, Mohsen Mosleh, and Enkelejda Kasneci (2026). Discussion of a many-analysts study run with coding agents, stable estimates, and interpretations that shift with the prompt.

“AI Coding Agents Can Reproduce Social Science Findings.” By Meysam Alizadeh, Mohsen Mosleh, Fabrizio Gilardi, Atoosa Kasirzadeh, and Joshua Tucker (2026). Discussion of a benchmark of 221 reproduction tasks from 54 papers, Claude Code against Codex, and prompts that nudge an agent toward specification search.

“The Agentic Garden of Forking Paths.” By Jiacheng Miao, Jonathan K. Pritchard, and James Zou (2026). Discussion of agents given opposite beliefs reaching opposite conclusions from the same data, reports that still pass expert review, and logging every analysis an agent tries.

“AI Agents for Economics Research.” From Aniket Panjwani, an economist who runs The AI Economist. Discussion of choosing between Codex and Claude Code, a first exercise, git, skills, and planning.

“Claude Code for Accounting Research.” From John Barrios, a professor of accounting at Yale University. A course for PhD students. Discussion of setting up Claude Code, choosing among agent tools, and a protocol for checking what an agent produces.

“AI-Assisted coding for research.” From Carpentry @ UCSB Library, the workshop program of the University of California, Santa Barbara Library. Discussion of running OpenCode in a sandbox, its permission system, AGENTS.md files, and planning before building.

From Anthropic, the company that makes Claude and Claude Code.

From Posit, the company that makes RStudio and Positron.

“ICPSR Policy on the Use of Large Language Models.” From ICPSR, the social-science data archive at the University of Michigan. Discussion of which kinds of model may see ICPSR data, why giving data to a model can count as redistributing it, and restricted-use data.

“Using Local Coding Agents.” From Sebastian Raschka, an LLM research engineer and former statistics professor at the University of Wisconsin–Madison. Discussion of running open-weight models through Ollama behind Claude Code and Codex, speed and memory on a Mac, how reliably local models use tools, and testing a model on your own tasks.

“Privacy-Preserving Analysis with Local LLMs: Generating Synthetic Data Proxies.” From Martin Schweinberger, a linguist at the University of Queensland. Discussion of describing confidential data to a local model, generating a synthetic stand-in, sending only that to a cloud assistant, and running the resulting R code on the real data.

“Randomness in large language models: What researchers need to know (and report).” By Guillaume Coqueret, Joan Llull, Florian Oswald, Christophe Pérignon, Christoph Scheuch, and Lars Vilhuber (2026). Discussion of why the same prompt gives different answers, why setting the temperature to 0 doesn’t fix it, and what to report.

“Agentic AI in Social Science Research: From Annotators to Autonomous Pipelines.” From Matt DiGiuseppe, a political scientist at Leiden University. Slides from a talk. Discussion of agents as research pipelines, checking every extracted quote against its archived source, checkpoints, and a register of what went wrong.

From Oliver Dürr, a researcher at the Thurgau Institute for Digital Transformation in Switzerland.

“A Practitioner’s Guide to Using Large Language Models and Generative AI in Economic History.” From Andreas Ferrara, an economist at the University of Pittsburgh. Discussion of four ways to use an LLM, chat versus agentic coding versus the API, and validating measures an LLM creates.

From Scott Cunningham, an economist at Baylor University.

“My Claude Code Setup.” From Pedro H. C. Sant’Anna, an economist at Emory University. Discussion of a ready-to-fork template, review agents, and quality gates.

“EconTools.” From Johan Fourie, an economic historian at Stellenbosch University. Discussion of Claude Code skills for simulated peer review and for turning a folder of PDFs into notes an agent can read.

“claude-academic-research.” From Mikko Rönkkö, a professor of entrepreneurship at the University of Jyväskylä. Discussion of checking citations, Zotero, systematic reviews, and revising a manuscript.

“AI for Economists.” From Jesse Lastunen, an economist at UNU-WIDER. A searchable list of about 220 resources: papers, courses, tools, and commentary.

“How to Use AI for Public Engagement.” From Alexander Kustov, a political scientist at the University of Notre Dame. Discussion of using AI to write for the public and catching its mistakes.

“Scaling Reproducibility: An AI-Assisted Workflow for Large-Scale Replication and Reanalysis.” By Yiqing Xu and Leo Yang Yang (2026). Discussion of using an AI workflow to replicate and reanalyze published political science papers at scale.