OpenAI’s Codex is, as the name suggests, a tool built primarily for… developers. It’s designed to help you write code, fix bugs, review pull requests, build tools and automations, and manage repetitive engineering tasks that eat up deep focus time. competes with tools like antiGoogle’s AI Studio and Anti-gravityOpenCode etc.
While the coding vibe is no longer something limited to a niche developer community, and people who have never touched a line of code before are creating some really impressive tools, these tools have always been something you open with a specific intention: to build something. Whether you’re a seasoned developer or someone who just discovered vibe coding last week, you open Codex, Claude Code, or Cursor because you have a project in mind. The starting point is always the code. But OpenAI’s latest updates to Codex feel like a departure from it.
Codex can do more than just write code
OpenAI wants you to know that
A few days ago I wrote an article about a couple of roads that I went through using Claude Code for non-coding use cases. However, with Claude Code, I felt I was finding more creative solutions: pushing the coding tool into territory it wasn’t meant for. Codex, on the other hand, builds that area locally. If you take a quick look at the recent features the team has added to Codex (or even check out their posts on X), you’ll see that the push outside of coding really isn’t random.
For example, one of their most recent blog posts was titled Codex for almost everything. This article will focus on what was announced on that blog, but the title alone tells you all you need to know about where OpenAI is taking it. So it’s clear from the way they talk about Codex that OpenAI no longer sees it as just a coding tool.
Codex can now see and control your screen
Now it has its own cursor
The first major update that Codex is getting is computer usage in the background. Codex can now handle graphical user interfaces on macOS and perform tasks such as clicking buttons, typing in fields, navigating applications, and interacting with tools. Codex does this by using its own cursor, which works independently of yours. This feature is currently only available on macOS and is not available in the United Kingdom, Switzerland or the European Economic Area. OpenAI says it’s useful for iterating interface changes, testing apps, or working on tools that don’t expose an API, but it’s easy to see how it extends beyond developer workflows.
If Codex can run any application on a Mac, there’s nothing stopping it from managing spreadsheets, organizing files, or handling admin work that has nothing to do with code. You can also ask him to do several things at the same time. Multiple agents can run in parallel on your Mac, each performing a different task without interfering with what you’re doing. In fact, Sam Altman publicly advised people at X to try using computers for non-coding tasks. You can also use it for incredibly simple things like skipping to the next song on Spotify or copying information from a document to a spreadsheet right in front of you. Basically, you can make Codex do anything you can do yourself with a mouse and keyboard on your computer.
In the same post, OpenAI also announced that it has a built-in browser where you can open pages in Codex, directly comment on what you see, and ask Codex to make changes based on your feedback. Limited to local and public pages for now. That said, this feature is aimed more at developers, so I won’t go into it too much here. However, it’s worth noting that OpenAI has said that they plan to expand it beyond localhost over time, which will make it more interesting for general use.
Codex now has 90+ plugins and can schedule its own work
The advice I always give people is to start plugging AI tools into the apps and services they already use. Artificial intelligence labs take this idea very seriously, and OpenAI is no exception. The company uses Jira, GitLab, CircleCI, Microsoft Suite, and more. released over 90 new plugins with the above update, including tools like A more interesting update that the tool has received is the ability to schedule its own work.
Codex can now set future tasks for itself, wake up automatically, and continue working on something for days or weeks. It can also proactively suggest what you should work on next by extracting context from your codebase, Google Docs, Slack, Notion, and any related plugins. So instead of you opening the Codex and telling it what to do, the Codex can open and tell you what to do. This is quite a significant change.
Codex is also testing a feature that remembers what you’ve done
And you don’t need to tell him anything
Both of the features above are already in place, but what really caught my eye is something that OpenAI is testing within Codex called Chronicle. This feature is currently in preview for research access and is only available to ChatGPT Pro subscribers on macOS. I’m on the $20 Plus level, so I haven’t been able to try it myself, but based on what OpenAI has shared about it so far, it sounds like one of the most interesting (yet dubious) additions to the Codex yet. Chronicle actually monitors your screen in the background, periodically taking screenshots and turning what it sees into structured memories that are stored as native Markdown files.
The screenshots themselves are deleted after six hours, but the memories remain. That way, the next time you refer to the Codex, you already have the context you’re working on without having to explain anything. Now, the privacy this feature brings is something we’ll dive into another day. But it still shows that OpenAI is turning Codex into something that understands your workflow instead of performing tasks for you.
OpenAI explains that Chronicle is designed to reduce the amount of context you have to re-express when you start using Codex. The feature works by running sandboxed agents in the background to create memories from screenshots, and the company shares that the agents currently consume rate limits quickly.
The direction Codex is going in could not be clearer
If all of the above doesn’t make Codex and OpenAI’s direction clear, the company’s own words should. The bottom line in OpenAI’s blog post talks about it narrowing the gap between what people can imagine and what they can build.
