OpenAI sent another earthquake rocking across the software engineering world today. 

This time, it was the X announcement that Codex, its AI coding agent, is now available for Plus tier users. In other words, you no longer have to spend$200/mo to get Codex's programming help. Instead, you can get it for$20/mo, as part of the ChatGPT Plus subscription.

(Disclosure: Ziff Davis, 's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.) 

The company says Codex "includes generous usage limits for a limited time, but during periods of high demand, we might set rate limits for Plus users so that Codex remains widely available." In other words, performance today might be rough.

Also: OpenAI upgrades ChatGPT with Codex - and I'm seriously impressed (so far)

That didn't stop me from digging in. There's a new menu item on the right of the ChatGPT screen:

Interestingly, the Mac app version of ChatGPT is far more limited. If you launch ChatGPT through the app, not only will you not get Codex, but Search Chats, Library, and Sora are also missing. That's kind of an unforced error, especially since the Mac app updated yesterday.

Also: The best AI for coding in 2025 (and what not to use)

In any case, let's see what happens when you click the Codex button in Plus. First, you'll get this exciting start screen.

Next is an animation that describes some features of the tool. In this first frame, the tool explains that it will draft multiple PRs in parallel. PR stands for Push Request, and that's GitHub-speak for pulling changes into the main repository, essentially integrating it into the main codebase.

By saying Codex drafts multiple PRs, it means it can work on multiple coding projects, but none get committed until you've done a review. This is roughly the equivalent of giving a bunch of projects to some subordinate programmers, letting them work on it, but not including their code until you examine it carefully.

You will examine Codex's code carefully, won't you? Hint. You better.

Next is a sample prompt. It's basically showing how you can prompt Codex to dig through the entire repo of a codebase to make changes and tweaks.

Next, Codex apparently can run Lint and tests. Lint, which was originally a C language tool that detects errors and bugs in code, works with many different languages. 

What Codex is saying here is that it's capable of running coding validation check, creating tests, and running those tests to confirm the code changes work as expected. This is a good thing.

Finally, one more panel describes the use of a new coding model specifically for software engineering. The company doesn't explicitly specify that model, but ChatGPT speculated that the model is a tuned version of GPT-4o, specifically trained on software engineering applications.

And then, I was finally in. The next form provided connection to a GitHub repository. This is very reminiscent of the process I went through with the Google Jules coding model last week. I'm telling you, these AI features are in lockstep among these competitors.

I was about to move forward and start playing around with Codex, when I noticed a message at the bottom of the screen.

This references a feature mentioned in the X posting by OpenAI this morning:

You can now give Codex access to the internet during task execution to install base dependencies, run tests that need external resources, upgrade or install packages needed to build new features, and more.

Stay tuned for hands-on testing

And that's when I decided it's time to stop. It's not even 7 a.m. here, I haven't even finished my first cup of coffee, and I want to get this news out to you as soon as possible. But I'm not comfortable, foggy morning head and all, letting Codex loose on my main repo, or thinking through the implications of giving it internet access.

Also: Google's Jules AI coding agent built a new feature I could actually ship - while I made coffee

That said, I will share with you one interesting discovery I made while the caffeine was starting to boost my brain. Apparently, Codex is stateless. This means that it doesn't remember anything from session to session. So you need to build clear and complete prompts, and make sure those prompts are repeated in new sessions if they set up ground rules or otherwise are designed to calibrate Codex on its work.

And with that, I'll wrap up this quick hit intro article. I'll be back soon with some hands-on tests to see what this beastie can do. Stay tuned.

Are you planning on trying out new Codex coding agent in ChatGPT Plus? What are your thoughts on letting connect to the internet to install dependencies? Do you feel comfortable giving it access to your codebase, or do you have concerns about automation at this level? What features are you most excited or nervous about? Let us know in the comments below.

Want more stories about AI? Sign up for Innovation, our weekly newsletter.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.

Featured

How much energy does AI really use? The answer is surprising - and a little complicated

6 ways I save money on TV streaming without losing the shows I love

This 16-inch Dell laptop is a worthy rival to the MacBook Pro - and it's better in a few ways

I changed 6 settings on my Roku TV to instantly improve the user experience

• How much energy does AI really use? The answer is surprising - and a little complicated
• 6 ways I save money on TV streaming without losing the shows I love
• This 16-inch Dell laptop is a worthy rival to the MacBook Pro - and it's better in a few ways
• I changed 6 settings on my Roku TV to instantly improve the user experience