7:25-15:50
Watch closely
Understand what Codex actually is
This is the conceptual anchor: Codex is not just a chatbot. It sees your files, edits them, runs commands, and can do the same local work you would otherwise do by hand.
Guide to Codex for Economists
A practical map through the Codex Desktop course: what to watch closely, what to skim, and what to do with your own research folders afterward.
How to use this course
Do not watch it like an app developer.
The course is broad because Codex is broad. As an economist, your goal is narrower: learn enough of the interface, permissions, skills, Git, and review loop to make your research folders safer and faster.
When the course turns into a web app build, do not worry about memorizing the stack. Watch for the workflow: plan the work, let Codex operate, inspect the diff, and keep only what improves the project.
Open the full course on YouTube →§ 01
Timestamped viewing plan
15:50-56:05
Watch closely
Learn the Desktop app, context, voice, and permissions
This is the part that makes the tool usable day to day. Pay special attention to context windows, compaction, voice input, permissions, guardrails, and terminal basics.
56:05-1:06:22
Skim, then configure
Project instructions and AGENTS.md
The course explains why giant instruction files can backfire. For research projects, keep instructions short: commands to run, files never to edit, what counts as a valid output, and how results should be checked.
1:06:22-2:17:00
Watch closely
Plugins, computer use, skills, MCPs, and subagents
For economists, and really for everyone, skills are the highest-leverage agentic coding concept to learn. Watch enough of plugins and MCPs to know what they are, then in particular pay close attention when the course turns to skills.
2:17:00-2:33:25
Do not skip
Git and GitHub
Git is essential for agentic coding. You do not need to memorize Git commands, but you do need to understand how to get Codex to use Git: checkpoints, branches, private repos, and reviewing diffs.
2:33:25-3:16:24
Watch selectively
Deciding what to build, cloud delegation, issues, and automations
Skip the section on cloud delegation. Deciding what to build can be useful for seeing the ideation, brainstorming, and planning workflow. Some economists like GitHub issues as a way to outsource memory of the backlog of tasks on research projects; if you have or prefer other systems, feel free to skip.
3:16:24-3:59:58
Skim first
The app build
Here, I am building a social media content creation app. It is unlikely to be of use to most of you, but more economists are thinking about how to communicate their research to a wider audience, so if you are interested, it will be helpful to see what a quick build looks like with a modern stack.
§ 02
Practical exercises for economists
Research memo
Turn a transcript into a research memo
Use Claude Code with yt-dlp on the transcript of some economics seminar on YouTube. Ask for a one-page memo with claims, citations to check, open questions, and follow-up readings.
LaTeX
Create one LaTeX skill
Write a skill that tells Claude Code how you prefer to have your tables formatted, your Beamer decks styled, or how to convert papers to Beamer decks according to your preferences.
Connectors
Connect one real knowledge surface
Try a Claude Connector, MCP, or Codex plugin for email, calendar, Drive, or Notion. Keep the task narrow: seminar prep, referee-report triage, or coauthor follow-up.
Project hygiene
Clean up an old project repo
Ask Claude Code for an evaluation of a pre-agentic coding era project folder's code. Then ask it how it would suggest cleaning it up, and implement its suggestions.
Use the deck as the course map.
The slide deck above is the full public deck I used for the course. The timestamped plan on this page is the economist-specific layer: where to slow down, where to skim, and what exercises to do after watching.