When OpenAI opened applications for the Codex Ambassador program this year, I was lucky to be selected in the first wave.
I was excited to share the news with friends, but I quickly realized that almost everyone outside engineering had the same question: “Huh? What is Codex?”
After I tried to explain, the next question was usually: “How is that different from ChatGPT? Aren’t they the same thing?”
That made me realize that, if you do not write code, the two can look very similar.
One of my biggest goals this year is to help my friends start using AI. So I decided to start writing, and this is the first question I need to answer.
Is Codex only for engineers?
Codex has “code” right in the name, so it’s easy to assume it’s only for engineers writing software.
That was true at the beginning. Codex was originally designed so AI could edit code files directly inside a project. But people soon realized that “editing files” is not only useful for engineers.
Most jobs that involve a computer also involve files: Excel, Word, PDFs, presentations, Photoshop, or internal company tools. Codex can help with those too. The idea is similar to editing code: it can work with files and apps on your computer.
Over time, more people outside engineering, especially knowledge workers, started using Codex. It has gradually become a general productivity tool, or what people often call an “agent” that can work for you.
An agent is AI that can plan and take action on its own. Once you get used to the idea that “AI can figure out what to do to finish a task,” you stop saying “let me ask AI” and start saying “I’ll ask my agent to research this.”
If that still does not make sense yet, no problem. For now, you can think of it as another way to refer to AI.
The main difference
The main difference is where they run:
- ChatGPT: runs on its website. It can see the conversations you have had with it and the projects you created there.
- Codex: runs on your computer. It can see the files and applications on your computer.
The difference is a bit like the difference between outsourcing a problem and hiring someone to work inside your company.
- Using ChatGPT is like working with a contractor. You need to send over the context and give them access before they can help. When they are done, they give you suggestions or files and say, “Here is what I recommend: …” or “You can download or copy this file to use it.”
- Using Codex is like bringing that person into the company. After you set up permissions once, it can find the information and tools it needs, finish the task directly, and then tell you, “I’ve updated the file and messaged the right people. You can review it here.”
The biggest day-to-day difference is whether you need to be the person moving content around.
Because Codex is on your computer, it can find the content and tools it needs by itself. That means it can more often complete the whole task for you. This is also why we usually call it an agent.
Should I start using Codex?
Codex may be able to take more off your plate, but ChatGPT is already very convenient for a lot of everyday tasks.
It can help you reply to email from your inbox, generate Google Slides, or edit Google Sheets. It already handles many cloud-based tasks. So depending on how you work, you might not need Codex yet.
I think the key question is this: do you often feel that “I’m tired of copying and pasting,” or that “switching between windows feels overwhelming”? If you often have that feeling, then I think you should start using Codex.
Are there reasons not to use Codex?
One reason: there is a learning curve at the beginning.
As mentioned earlier, Codex can see files on your computer and use apps on your computer. But you probably do not want it to have access to everything all the time. There may be files you do not want it to read, or apps you do not want it to use.
Because of this, Codex lets you decide what the agent can see and do for each project. That means the interface includes things like project, permissions, and sandbox settings.
Also, because Codex started as a software development tool, some parts of the interface still make more sense to engineers, such as worktrees, git commands, and the terminal.
These two things make the interface look like it has a lot of buttons and settings. Some people may open it and think: this looks complicated, and I do not know where to start.
This is where the first bit of friction comes from: when you open Codex, the UI asks you to think about a few things before you start. ChatGPT is very easy and intuitive: you see an input box asking what you want to do today, and you type your question. With Codex, the UI puts one extra question in front of you: where should Codex work? (In other words, what content can it read and use?)
This is not actually difficult. Even if you don’t configure anything, the default setup is still safe: when Codex needs access, it asks first. When the agent needs to do something, it will ask questions like: “Can I look at this file?” or “Can I edit this file?” You can decide then.
But that approach interrupts your workflow, because you will often need to go back to the Codex window to approve permissions.
So to use Codex efficiently, it is worth spending a little time at the beginning learning how to set up projects and permissions. Maybe about ten minutes. If that sounds annoying or complicated right now, then I think it is totally fine to keep using ChatGPT.