Claude Code for AI Research
Coding agents rerun baselines, sweep seeds and write the scripts behind results tables, then report on code you did not watch run. This course teaches you to hand that work to Claude Code and still defend each number: reproduced first, measured on an evaluation the agent cannot edit, reported with n and an error bar, and reviewed in a fresh context. You practice on the Chapter 16 evaluation code from the book's public repository, which runs offline with uv on Python 3.12.
OutcomeYou leave with: a branch of the book's course repository with deny rules, a hook, a CLAUDE.md and a skill, plus a lab notebook with a reproduced Chapter 16 result, one measured experiment with intervals and a seeded-bug review score
Rod Rivera
Professor
Desk
Every lesson is free. Lessons marked open need no account; the rest open once you register.
What You'll Learn
- How an agent loop differs from a chat exchange, and why reliability over n steps compounds as p^n
- How to install Claude Code from a blank machine and run the practice repository's checks with uv on Python 3.12
- How to reproduce a recorded result before changing anything, and record it in a lab notebook
- How to block edits to evaluation files with permission deny rules, a PreToolUse hook and a hash check, and what each layer misses
- How to write the hypothesis, metric, baseline, evaluation set and decision rule before the agent edits, and review its plan in plan mode
- How to size an experiment to one question and compute the n it needs
- How to demand the command, commit, seeds, eval-file hash and n, and derive the binomial interval and McNemar's paired test
- How to write a CLAUDE.md and a skill for a research repository, and check that they loaded and were followed
- How to run experiments in parallel with worktrees, subagents and headless runs, and what that costs in tokens
- How to review agent-written research code with a fresh-context reviewer, measured on bugs you plant
- How to present agent-assisted work honestly in an interview loop or a take-home
Prerequisites
- Python you can read, and basic git (commit, branch, diff)
- No prior terminal setup required; Lesson 02 starts from a blank machine
- A Claude Code plan or API access; this free course does not pay for the tool
Syllabus
Read the rest of Claude Code for AI Research
The first lessons need no account. Sign in free to read the rest and keep your place on any device.
