OpenAI Codex Explained for Normal People (“Claude Cowork Killer”)

Summary

The video presents OpenAI Codex as a desktop app for knowledge work as well as coding. Unlike ChatGPT, which the presenter describes as keeping created content in the cloud, Codex can work with files in folders on a user’s computer. The presenter compares its workflow to Claude Cowork and explains why some people call Codex a “Claude Cowork killer,” while focusing on practical beginner use rather than a direct product review.

Beginner walkthrough and features

• Setup: Download the Codex app, sign in with a ChatGPT account, and choose a work category during setup. The presenter says Codex is included with a ChatGPT subscription.
• Projects: A Codex project is essentially a computer folder that Codex can use as its working location. The video demonstrates selecting a folder of travel receipts and asking Codex to create an Excel workbook that categorizes expenses and totals trip costs. The workbook can be previewed in Codex or opened from its folder.
• Creating files from prompts: Codex can produce files beyond spreadsheets. The presenter dictates a request to analyze his website’s colors, fonts, and styling, then create a Markdown brand guide.
• Desktop pet and multitasking: A slash command activates a small desktop pet that signals when a task needs attention or has finished, allowing the user to work elsewhere while Codex runs.
• Skills: Skills are reusable instruction sets. In the example, the presenter asks Codex to create an “on brand” skill that applies the brand guide to future work, then invokes it while requesting an infographic comparing Codex and Claude Cowork. The presenter says this uses OpenAI’s Images 2.0 model.
• Plugins and email triage: Plugins connect Codex to third-party services. After installing the Gmail plugin, the presenter asks Codex to scan email, draft replies in his voice, and adapt based on past responses. He then asks it to run the triage automatically each morning at 9 a.m. Automations and plugins can be managed in the app.
• Interactive dashboards: The final example asks Codex to create an interactive dashboard from a project folder, illustrating that its output need not be a conventional document.

Key takeaway

The tutorial’s central advice is to describe the outcome you want and let Codex determine the steps, rather than prescribing how to perform every task. The presenter emphasizes using projects to keep work organized, and skills and plugins to reuse instructions and connect Codex to external services.

Main speaker and sources

• Presenter: Paul J. Lipsky, explaining and demonstrating Codex.
• Products and services discussed: OpenAI Codex, ChatGPT, Gmail, and Claude Cowork. Canva, Remotion, Google Drive, and Google Calendar are mentioned as examples of other available integrations.