How to Use OpenAI Codex in VS Code for Free: Setup, Demo, and Review

Introduction to OpenAI Codex

• 💡 The video discusses Matt Shumer's viral essay claiming AI, specifically GPT 5.3 Codeex, will cause significant job market disruption.
• 🎯 Shumer stated that AI agents could build "the finished thing" from plain English, even possessing human-like judgment and taste.
• ✅ OpenAI Codex is currently available for free access through a ChatGPT account, allowing users to test its capabilities.

Setting Up Codex in VS Code

• 🛠️ First, users need to install Visual Studio Code on their preferred operating system (Windows, Mac, or Linux).
• 🚀 Next, navigate to openai.com/codex, click "Try in your IDE," and select VS Code from the dropdown menu.
• 🔑 The process involves installing the Codex extension from the marketplace and then signing in with a ChatGPT account.

Practical Demo: React Wordpad

• 💻 The demonstration involved prompting Codex to create a React Wordpad clone, complete with a ribbon interface and formatting options.
• ⚠️ Codex provides an important safety feature by asking for permission to run commands like npm install, allowing user oversight.
• 🐛 The initial code generation was extensive, but the app immediately presented errors (e.g., "React is not defined") requiring debugging and explicit prompting to fix.

Iterative Development & Challenges

• 🔍 Early versions of the Wordpad clone had several functional bugs, such as the file menu not closing and incorrect text centering.
• ✅ After receiving specific feedback, Codex successfully resolved the reported issues, improving the application's functionality.
• 🧩 However, new problems emerged, including broken zoom and font selection, alongside a significant code structure issue where all components were in a single file.

Verdict on AI-Assisted Coding

• 📈 Codex is highly effective for generating proof of concepts (POCs) quickly, demonstrating significant advancements in AI-assisted coding.
• ⛔ Despite its utility, the tool cannot entirely replace human programmers, as it still requires substantial technical knowledge for debugging, error interpretation, and architectural decisions.
• ⚠️ The speaker cautions against organizations laying off engineers en masse to replace them with AI agents, warning of potential "spaghetti code" and maintenance issues.