How to Use OpenAI Codex (Step-by-Step Tutorial)

About this video

Kevin builds a browser game called Kevin’s Cookie Shop Tycoon to teach OpenAI Codex in the ChatGPT desktop app on Windows and Mac, noting Codex is also available on the web, via CLI, and in a code editor. After creating a project — a local folder Codex works in — and clicking Trust Folder, he prompts in plain English for a game where the player starts with $100, runs a shop for 30 days, and each day decides cookies to bake, price, and advertising spend, then sees sales, revenue, expenses, profit and cash. Codex finishes in about three minutes and 10 seconds, creating the files and an index.html to open. He then iterates by prompting for animated customers reflecting cookies sold, a change that takes a little under two minutes and modifies game.js, index.html and the README, adding about 47 lines and removing three. He shows integrated Git: view changes and diffs (green added, red removed), and asks Codex to commit with the message ‘Add animated characters’ as a checkpoint. Next he runs three agents in parallel in isolated worktrees — cookie varieties, a shop upgrade system, and visual polish — finishing in about five minutes, previews each version, and merges only cookie varieties into main. He creates a reusable Game Tester skill to check launch, controls, calculations, and that players cannot spend or sell what they do not have, reporting problems without fixing them. Finally he connects GitHub via Plugins, has Codex create and push a Cookie Shop Tycoon repository, switches a new chat to cloud, creates an environment, and delegates a customer satisfaction system remotely.

Key takeaways

• Start in the ChatGPT desktop app’s Codex view: add a project folder, trust it, and describe the build in plain English — no language or file choices needed
• Build iteratively: prompt, test the result (e.g., open index.html), then prompt for changes instead of starting over
• Use integrated Git commits as checkpoints so you can view diffs by last turn and roll back if something breaks
• Split larger work across parallel agents in isolated worktrees, preview each result, and merge only the features you want to keep
• Save repeatable instructions as skills and hand off to Codex Cloud via a connected GitHub repository when you want work to run remotely

Episode details: published October 8, 2026 · 17:11 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Kevin Stratvert’s own claims, shared for education — they are not typical or guaranteed.

← Previous: How to Edit Videos with ChatGPT · Next: How to Categorize Transactions in QuickBooks Online (2026) →

All Kevin Stratvert videos · Kevin Stratvert profile · Kevin Stratvert on YouTube