Back to Work

AI for Research

2026

Coordinator & Teacher • Amsterdam Business School, University of Amsterdam

EducationAgentic AIClaude CodeReproducibilityReplicationOpenAlex

Overview

AI for Research is a course on using agentic AI tools, such as Claude Code, in research. Students learn to let an agent work in a project folder, keep the work reproducible, and check every result before it goes into a paper.

The tools read files, write code and run it. That makes them useful, and it makes mistakes easier to miss. The course is for researchers who want to use them without losing control of their work. I coordinated the first edition, in autumn 2026, and taught it together with colleagues from the University of Amsterdam.

Structure

Session 1
Foundations

Basics of AI use, and defining the structure of your work so an agent can operate inside it.

Session 2
Literature

Using Claude tools and vibe coding for analysis, including a literature tool built on OpenAlex.

Session 3
Replication

Reading research by rebuilding it, using Card and Krueger (1994) as the case.

Labs

•

Turn a paper into a newspaper article

•

Build a small literature tool on OpenAlex that scores abstracts against your own statements

•

Rebuild a table from a published economics paper and see what one choice does to it

Reading Research by Replicating It

In the third session participants rebuild the numbers of Card and Krueger (1994), the New Jersey minimum-wage study of 410 fast-food restaurants. An agent rewrites the authors' SAS program in Python, and each pair then changes one choice and explains what it does. Every decision goes into a logbook. What the room found:

•

All five models of Table 4 reproduce, including n = 357 and the New Jersey estimate of 2.33 (SE 1.19).

•

Removing the wage-data rule adds 27 stores and moves the estimate from 2.33 to 2.75.

•

Heteroskedasticity-robust (HC1) standard errors give 1.45 instead of 1.19. Both 95% intervals still include zero.

•

The paper and its SAS code disagree on model (v): 11.91 (7.39) from the code, 11.98 (7.42) from the footnote.

Next Edition

A second iteration is planned for the first quarter of 2027. Dates are not announced yet. The course website has the details and a way to register your interest.

ai4research.strezoski.com