MY580: Agentic AI for Social Science and Data Science Research

  • Author/Source: Daniel de Kadt (London School of Economics), workshop repo on GitHub

  • Original: https://github.com/ddekadt/MY580_agentic_ai/tree/main

  • Key Ideas

  • MY580 is a hands-on workshop at LSE introducing social science and data science researchers to agentic AI through a free, terminal-based coding assistant.
  • Primary tool is Gemini CLI (Google's free terminal-based agent as of May 2026), chosen for accessibility — but the principles are vendor-agnostic and apply to Claude Code, Codex, or other agentic assistants.
  • Workshop covers how modern AI agents plan, use tools, and execute multi-step tasks, using Python and R as the analysis languages.
  • Recommended IDE is Positron (Posit's modern fork of VS Code aimed at data scientists).
  • Pedagogical scaffolding: students get a starter project_student/ folder with a blank GEMINI.md (the Gemini analogue of CLAUDE.md) and pre-staged raw data, plus a reference project_demo/ showing a worked example.

  • Setup is intentionally low-friction: ~10-minute install, only requires a personal Google account for the free tier, runs on any laptop.

  • Summary

The MY580 workshop materials, authored by Daniel de Kadt at the London School of Economics, package a one-day introduction to agentic AI for social science and data science researchers. Unlike most agentic AI tutorials in the academic-research orbit, which assume Anthropic's Claude Code, MY580 standardizes on Gemini CLI specifically because Google's free tier removes the cost barrier for student adoption. The author is explicit that the choice is pragmatic, not ideological: the workshop teaches general principles of agentic AI use, and students with paid Anthropic or OpenAI accounts are encouraged to substitute their preferred tool.

The repo bundles OS-specific setup guides (Windows and macOS), starter and demo project folders, and a workshop slide deck (MY580_agentic_ai.html). The split between project_student/ (a blank GEMINI.md plus raw data) and project_demo/ (a worked Module 4 example) mirrors the increasingly standard pattern in agentic-AI pedagogy: give students a configuration file scaffold, pre-staged data, and a reference implementation they can read but not edit. The choice of Positron as the recommended IDE is notable — it signals that the workshop treats agentic AI as part of a data-science stack, not a pure software-engineering one.

  • Relevance to Economics Research

MY580 is one of the first agentic-AI workshops at a major social science department (LSE Methodology) and represents how methods training is starting to absorb agentic AI as a core competency rather than an elective. For economists building similar workshops, three design choices are instructive: (1) standardizing on a free tool to maximize student access, even at the cost of using a less-capable agent; (2) using a GEMINI.md / CLAUDE.md-style configuration file as the central pedagogical artifact students fill in themselves; and (3) explicitly framing the workshop as vendor-agnostic principles rather than tool-specific tricks, hedging against the rapid pace of capability change. The pre-staged data/raw/ pattern and the demo/student folder split are directly reusable for an economics-research equivalent.