Follow ZDNET: Add us as a preferred source on Google.
ZDNET's key takeaways
- OpenAI launches a Mac-only Codex app as an agent command center.
- Sandbox controls limit folder writes and network access for safer use.
- Switching between IDE, terminal, and app keeps context across tools.
OpenAI today announced a new Mac app dedicated to working with its Codex AI coding agent. This is different from the general-purpose ChatGPT app that OpenAI has been shipping for a while. The new coding app is intended to be something of a command center, not only for directing coding agents, but also for managing multiple coding agents across projects and tasks that run for long periods of time.
Also: I got 4 years of product development done in 4 days for $200, and I'm still stunned
In a briefing I attended with OpenAI executives a few days before the launch, OpenAI cofounder and CEO Sam Altman said, “There's obviously been a huge shift in software agents over the last few months as the models have crossed a threshold of real utility. 5.2, in particular, is a model that many of us have found can do extremely complex things, and we realized we started to feel limited by the interface.”
The new Mac app is intended to help mitigate that limitation.
OpenAI described Codex as expanding from single-agent coding operations to being at the core of multi-agent software lifecycles. Rather than pairing with a developer on a single edit, OpenAI sees programmers as coordinating teams of agents across all of the design, build, ship, and ongoing maintenance stages of work.
(Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
Usage growth and customer uptake
The latest version of Codex, GPT-5.2-Codex, only launched in mid-December. Yet, according to the company, Codex usage has nearly doubled since that time, with more than a million developers using it in the last month.
I used it. In fact, with just the $20-per-month ChatGPT Plus plan, IÂ found a mystery bug. Subsequently, I also used GPT-5.2-Codex on that same Plus plan to add two fairly significant features to my security product and ship a major update.
Also: I used GPT-5.2-Codex to find a mystery bug and hosting nightmare – it was beyond fast
In the briefing, Altman stated that GPT-5.2-Codex “is the fastest adopted model that we have ever made.” He also reported on an extreme level of momentum using this tool, with usage growing more than 20 times since last August. In addition to independent developers like me, the company reported that major customers include Cisco, Ramp, Virgin Atlantic, Vanta, Duolingo, and Gap.
Inside the Codex app: parallel agents, skills, and automations
The new app is designed to run multiple tasks in parallel, and help users supervise agents across longer end-to-end work. The company is also introducing skills, similar to those in Claude Code. These are meant to allow developers to define repeatable workflows like fetching logs and fixing tests, along with coding-adjacent tasks like summarizing threads and completing tickets.
Right now, the app is Mac only. However, I expect that there will be a Windows app eventually. OpenAI shipped the ChatGPT app in Mac-only form, but some months later launched a Windows version.
In addition to the app itself, OpenAI announced a new plan mode for Codex that allows for a read-only review (meaning the AI won't muck with your code) and selectable personalities. Personally, I've had just about enough personality from the human programmers I've managed, so I'd prefer a nice, personality-free personality in my coding agent.
Recently, OpenAI also announced that Codex has an IDE extension for use in the JetBrains IDEs. Readers may recall that back in June I moved off of PhpStorm, my favorite JetBrains development environment. I moved to VS Code simply because the AI tools were more available for that environment. It's nice to see JetBrains IDE availability for those of us who prefer it over VS Code.
During the briefing, I asked about the relationship between the various coding options now offered with Codex. We can code in our favorite IDE, in the terminal, and now in this new app. I wanted to know whether it would be possible to get the same richness of experience in the app that we have in the IDE, with our entire codebase in front of us.
Sam told me, “I was astonished by this, but I did a fairly big project in a few days earlier this week and over the weekend.” He said, “I did not open an IDE during the process. Not a single time. I did look at some code, but I was not doing it the old-fashioned way.”
He further told me, speaking about this level of coding ability from the AI, “I did not think that was going to be happening by now. I never had that experience before.”
I was given access to a preview build of the Mac app. I found I was easily able to work in my codebase. In fact, it knew about my most recent work in the IDE. It just automatically picked up from where I last left off. It seems that we'll be able to switch between whatever coding tools we want, and Codex will maintain context.
Security model and control on local machines
Codex does have a cloud tool, which lives in GitHub. In fact, that's how Codex initially launched. When I first tested it in that context, I didn't like it. But once I started working with Codex on my own computer using the IDE extensions, I found that my productivity skyrocketed.
The problem with running an AI on a local computer, especially with write access, is that it can be a security concern. In the briefing, one of OpenAI's Codex developers posed the worry from a design perspective, saying, “If the agent runs on the user's computer, how do we make that as safe and secure as possible?”
Also: How to learn ChatGPT in an hour – for free
The key is limiting sandbox access to only approved folders. In my case, that's my project folder. The tool also has permissioned network access and remembers approvals over time. The new Mac app adds a sandbox mode and lets developers set approval levels, including Untrusted, On failure, On request, and Never (meaning the app is never permitted to ask for elevated permissions).
What OpenAI expects next for coding
The OpenAI development team described its current coding agents as now reliable enough to let it push entire workflows into the new app. A number of the OpenAI developers said that, in just the last few weeks, they've started working almost entirely in the Mac app. They haven't even needed to use their IDEs or terminal windows.
OpenAI currently makes Codex available with limited usage in the $20-per-month ChatGPT Plus plan. It offers much more extensive resource availability in its $200-per-month ChatGPT Pro plan. While I ran into roadblocks building new products (full days of coding) in the $20-per-month plan, I never ran into a roadblock when I was using the $200-per-month plan.
But companies have asked OpenAI for something more. It's not just about how many tokens can be allocated. For those companies, it's about speed of computation. You can wait quite a few minutes after each major AI command. OpenAI said that it's looking at creating higher compute-intensive, speed-intensive tiers for companies willing to pay for pure speed. These would allow for very long context coding work and include high-speed infrastructure options.
Are you vibe coding?
To encourage users to give Codex a try, the company announced, “For a limited time we're including Codex with ChatGPT Free and Go, and we're doubling the rate limits on Plus, Pro, Business, Enterprise, and Edu plans.” So feel free to give Codex a try.
Also: 10 ChatGPT Codex secrets I only learned after 60 hours of pair programming with it
What about you? Are you planning to try the new Codex app when it launches, or do you prefer sticking with an IDE extension like VS Code or JetBrains? Have you used GPT-5.2-Codex (or other coding agents) for real features and bug hunts yet? Did AI tools actually save you time, or did they create cleanup and tech-debt headaches later? Comment below.
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.
