Getting Started with Claude Code
Welcome to Day 1 of the AI Agents workshop. This is the on-ramp: what a coding agent is, how to think about it, and how to get your first Claude Code session going — either with a Claude Max subscription or with Anthropic API tokens (pay-as-you-go, around $5 covers the whole workshop).
If you have never used a coding agent from the terminal before, start here. When you’re comfortable, move on to the AI Agents workshop (Day 2) to turn a messy research folder into an AEA-compliant replication package.
Materials
- Tutorial. Step-by-step guide: mental models, both setup paths, first session, the
CLAUDE.mdfile, pitfalls, cost management, and a small hands-on exercise. - Slides. Accompanying slide deck.
- TUTOR.md. Optional: an instruction file that lets Claude Code walk you through your very first session, interactively.
What you’ll leave with
- Claude Code running locally against either Claude Max or an Anthropic API key.
- A mental model of what a coding agent is and isn’t: what it’s good at, where it fails, and what remains your job.
- Four prompt patterns you’ll reuse every day: vague-start, scripts-not-commands, style-by-reference, small-iterations.
- A
CLAUDE.mdof your own and a first feel for directing the agent through it. - A checklist of pitfalls so you avoid the ones that bite beginners.
Prerequisites
- Docker container from the Docker workshop, up and running. The image bundles Claude Code.
- Either a Claude Max subscription, or a credit card to load $5 into an Anthropic Console account. We set up both paths in the tutorial.
- Roughly 90 minutes.
About
This workshop is aimed at Master’s and PhD students in economics who have used ChatGPT in a browser but have never driven a coding agent from the terminal. It complements Day 2 by starting one step earlier: with how to think about agents. It draws on Paul Goldsmith-Pinkham’s “Getting Started with Claude Code” and “From an empty folder to a figure” for the pedagogy, and on the earendil-works/pi-tutorial for the agent-as-tutor pattern shipped as TUTOR.md.
Want to try non-Claude models (GPT-5, DeepSeek, Gemini)? The Day 2 tutorial appendix has a self-contained setup for Aider + OpenRouter: one API key, ~300 models, per-key spend caps.
Feel free to open an issue on GitHub if anything is unclear.