Getting Started with Codex from Zero: A Complete Hands-On Guide with Vibe Coding Tutorial

For most users who have never touched programming, AI tools often seem stuck at the level of "chat and Q&A." OpenAI's Codex is pushing AI beyond "talking" into "doing" — it doesn't just answer questions; it actually executes tasks on your computer and phone.

This article is based on a hands-on tutorial by a humanities-background content creator on Bilibili, distilling a clear onboarding path that anyone can follow — even with zero technical background.

What Is Codex and Why Does It Matter?

Unlike purely conversational AI tools like Doubao or DeepSeek, Codex's core value lies in its ability to take action. On top of conversation, it can genuinely help you get things done on your phone and computer — not just hand you a wall of text.

Compared to similar AI agents (such as some coding workbench products), the creator's takeaway was: Codex is smarter, more cost-effective (around the $20/month tier), and offers a Chinese-language interface that significantly lowers the barrier to entry. These three factors — execution capability, value for money, and language accessibility — explain its rapid rise in popularity.

One key thing to understand: Codex is already integrated into the new version of ChatGPT — both come from the same company. You don't need to hunt down a separate app called "Codex"; you access it by upgrading to the new ChatGPT.

Installing and Setting Up Codex

Download and Installation

The installation process is quite smooth. Search for and download ChatGPT on your computer — the installer automatically detects whether you're running Windows or Mac. Once installed, you'll notice something interesting: the old ChatGPT and the new Codex-integrated version can run side by side, independently, without interfering with each other.

Worth mentioning: if you've used other AI coding tools before (like Claude Code), Codex will automatically migrate your existing configuration during installation — and the migration experience is seamless.

After logging in, you'll land on the main interface. The top-left corner lets you toggle between ChatGPT and Codex modes. If you can't find this toggle, you're probably on the old version — note that simply clicking "Update" in the old version may not get you to the Codex-enabled new version. You may need to download and reinstall fresh.

Four Key Settings to Configure

Before diving in, it's worth tuning these basic settings:

• Request Approval (bottom-left of the input box): Set it to "Auto-approve for me" so Codex can run tasks without you having to manually confirm every step — you can step away and let it work.
• Model and Token Quota (bottom-right of the input box): The default model is fine for everyday tasks. Newer models perform better but consume more tokens.
• Planning Mode (found in the "+" menu at the bottom-left of the input box): This is a "think before you act" mode that helps complete most everyday tasks while conserving tokens.
• Remaining Usage (bottom-left corner): Check how many tokens you have left at any time.

Extending Codex's Capabilities with Plugins

Codex's capabilities can be significantly expanded through plugins. The tutorial highlights two commonly used categories:

• Computer Use: Gives your computer "hands" to operate the interface
• GitHub: Think of it as a repository of skills — find and pull whatever functionality you need

Here's how to install the GitHub plugin as an example: Go to "Plugins" in the top-left menu — most popular plugins are findable under "Featured." Select GitHub, then click "Add → Select → Continue," authorize access, configure your repository, choose the installation scope, then exit and restart the plugin. After restarting, you'll see the newly installed plugin under "Installed."

This "pull skills from a repository" design fundamentally lowers the barrier for non-technical users — you don't need to understand how a plugin works under the hood; you just need to know what problem it solves.

Vibe Coding in Practice: A Complete Loop from Idea to Finished Product

The most valuable part of the tutorial is the creator's hands-on demonstration of Vibe Coding — coding by feel — from a humanities background. She summarizes a closed-loop workflow that mirrors how a product manager thinks:

1. Identify a need
2. Describe the need to the AI
3. Let the AI write code to implement it
4. Review the result
5. Ask the AI to keep refining when issues come up

Step 1: How to Find Your Need

Many people get stuck at step one — not knowing what to ask the AI to do. Here's a practical tip: think about the things you have to do every single day, but absolutely hate doing. Your pain points are your requirements.

The creator used herself as an example: she dreads editing videos before each upload, especially adding animations — a huge burden for someone without a design background. So she turned "having AI automatically add motion effects to videos" into her use case.

Step 2: How to Describe Your Need

There's an efficient shortcut for describing what you want: if someone online has already built a skill for it, just search and install it directly. The creator found a ready-made skill online that could automatically add motion effects to videos, and installed it right from within the plugin interface.

Step 3: Execute and Review

After installing the skill, she opened the Codex chat window and described what she wanted in "coded" terms — she sent both her script and reference video effects to the AI, then let it run.

The AI took 17 minutes to complete the task and delivered a finished product. Her verdict: the quality was solid, but the pace was a bit slow. She simply said, "Please keep refining this and speed up the animation," and the AI produced a new version. That's the full "review and optimize" loop in action.

No Ready-Made Skill Available? Build Your Own

If no existing skill on the market solves your problem, Codex lets you create your own skill. Click "Create" in the top-right corner of the skill marketplace, and Codex will automatically invoke the skill creator tool. From there, just describe the functionality you want in plain language — and it will package it into a custom skill just for you.

The Barrier to AI Execution Is Disappearing

The biggest takeaway from this tutorial isn't the step-by-step instructions — it's what they prove: the shift from AI as a "conversation tool" to AI as an "execution tool" is already delivering real benefits to non-technical users. A creator with a humanities background was able to independently complete the full loop — identifying a need, describing it, executing it, and refining the result — not through coding skills, but through clear communication and an understanding of the tool's workflow.

For everyday users, instead of agonizing over "should I learn to code," a better starting point is to look at your most tedious, repetitive tasks and hand those pain points to Codex to try solving. The barrier is already low enough. All that's left is to act.