Video summary
How to Use OpenAI Codex (Step-by-Step Tutorial)
Main summary
Key takeaways
Summary
Kevin Stratvert presents a beginner-friendly, step-by-step tutorial on using OpenAI Codex to turn a plain-language idea into a working software project without needing to write code.
Tutorial workflow and features
-
Get started and create a project
- The tutorial uses the ChatGPT desktop app on Windows or Mac, with Codex selected from the app.
- A Codex project is a local folder containing the files Codex can work with. After creating and trusting a folder, users can describe a project in ordinary language.
- Unlike a standard chat, Codex can work directly with project files, run commands, and take actions.
-
Build and refine a game
- Kevin prompts Codex to create a browser-based cookie-shop management game. Players start with $100 and make daily decisions about baking quantities, prices, and advertising.
- Codex creates the necessary files, and Kevin previews the game by opening its
index.htmlfile. - He then asks Codex to add animated customers. The tutorial demonstrates an iterative workflow: describe a feature, review and test the result, then provide feedback for further changes.
- Codex shows which files it changed and highlights additions and removals.
-
Track changes with Git
- Codex’s change view displays file differences, with green for additions and red for removals.
- Kevin asks Codex to create a Git commit—a saved checkpoint—without manually entering Git commands. Commits provide a way to return to earlier project states if later changes cause problems.
-
Use multiple agents
- Kevin assigns three improvements to separate agents: cookie varieties, shop upgrades, and visual polish.
- Each agent works in an isolated worktree, or separate copy of the project, so tasks can run in parallel without altering the main project.
- He reviews each result and brings only the cookie-variety feature into the main project, leaving the other changes separate.
-
Create and reuse Skills
- A Skill is a reusable set of instructions for a recurring task.
- Kevin creates a “Game Tester” Skill to check whether the game launches, controls and calculations work, and players cannot spend or sell resources they do not have.
- Codex runs the checks and reports issues without changing the project. Skills can also be used for other repeatable work, such as code reviews or documentation.
-
Hand work to Codex Cloud
- The tutorial demonstrates connecting GitHub, creating a repository, and pushing the local project so Codex Cloud can access it.
- Kevin switches to a cloud environment and assigns a customer-satisfaction feature. The task runs remotely, and its results can be reviewed later in the desktop app or on the web.
- The demonstrated workflow is to work locally, push to GitHub, and then delegate tasks to Codex Cloud when remote work is useful.
-
Explore other ways to use Codex
- Codex is also available on the web, through a command-line interface, and as an extension for Visual Studio Code.
- Kevin recommends the desktop app as an approachable starting point.
Key takeaway
The tutorial shows a practical Codex workflow: start with a natural-language description, let Codex build or modify files, test the result, use Git for checkpoints, and optionally delegate parallel or remote tasks. The cookie-shop game is an example; Kevin notes that the same approach can be applied to apps, websites, automation, and other software ideas.
Main speaker
- Kevin Stratvert — presenter and guide.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.