OpenAI has taken another step in the integration of artificial intelligence with daily computer use by completely revamping Codex, its desktop code agent. The tool, previously known primarily for its programming assistance, is now becoming an assistant capable of handling applications, remembering preferences, and managing projects almost seamlessly.
With this update, the company led by Sam Altman brings closer its plan to creating a A desktop “super app” that unifies ChatGPT, Codex, and your Atlas browserThe move aims to simplify the user experience, concentrate resources in a single workspace, and take on rivals like Anthropic, especially in the software development environment and advanced automation.
Codex learns to use the computer like any other user
One of the most striking changes is that Codex can now control the computer in the backgroundThe agent is able to see what is on the screen, move its own cursor, click, type text and react to what happens in the installed applications, without interrupting what the user is doing with the mouse or keyboard.
This capability allows multiple Codex agents to work together in parallel on the same teamThis involves performing tests, changing configurations, or executing repetitive tasks while the person continues with their normal activities. This opens the door to automating workflows that previously required constant supervision, such as testing an application after every code change or repeating interface actions.
OpenAI proposes this function as a way for the assistant to operate with tools that do not have an API or direct integrations, something common in many desktop programs. Codex, by "seeing" the screen and controlling the cursor, can interact with these applications just as a human user would, although for now in a controlled manner and with special attention to security.
According to the company, computer use is initially available in macOS via the Codex desktop application, linked to a ChatGPT account. The Windows version will arrive later, once performance has been tested and features adapted to the Microsoft environment.
In the case of Europe and the United Kingdom, OpenAI has confirmed that the computer control, advanced customization, contextual suggestions, and memory They will be activated later. The reason is the need to adjust these features to data protection regulations, a particularly demanding requirement in the European Union and the British market.
Integrated browser and direct work on web pages
The update also strengthens Codex's relationship with the web thanks to a browser integrated within the application itselfInstead of simply receiving generic text instructions, the agent can receive commands directly on specific elements of a page.
In frontend development, for example, the user can Select a web component and describe the change. The user wants to do something (like change the color, text size, or layout of a block), and Codex understands the instruction within that specific visual context. This avoids having to write lengthy prompts explaining which part of the site needs to be altered.
This embedded browser also allows Codex Analyze content, gather information, and use it as context For other tasks, such as documenting code, reviewing product specifications, or comparing technical documentation, everything is done within the same environment, without having to switch between multiple operating system windows.
Generating and editing images with gpt-image-1.5
Another pillar of the new version of Codex is the integration of the model gpt-image-1.5 for creating and modifying images From within the desktop application itself, the agent can generate charts, diagrams, mockups, or visual sketches from text, or by combining instructions with screenshots and code snippets.
This capability is aimed directly at the teams working in product design, user interfaces, or video game prototypesFor example, a developer can capture a screenshot of a half-finished interface, ask Codex to suggest a menu redesign or a new color scheme, and get generated images that fit that context, without changing tools.
With everything centralized in Codex, agents can chain together coding and design tasks: write the logic of a functionality, update the corresponding frontend and generate a visual mockup to present to the team, without leaving the workflow or having to open external graphic editing applications.
Memory, context, and long-term task automation
Beyond computer control and image generation, OpenAI's big bet with this version is to equip Codex with a persistent memory and much broader context managementThe tool can reuse previous conversations, preserve important information, and retrieve it when needed to continue a project.
The memory feature, which is being released in preview, allows Codex remember personal preferences, frequent corrections, or data that was difficult to collectThis includes everything from a team's typical technology stack to how folders are organized, the most commonly used report formats, and the tag typically used on GitHub for certain types of issues.
With that accumulated context, the agent can proactively propose tasksThis can be used to suggest resuming an unfinished project, recommend pending refactoring, or notify users about long-standing pull requests. Some companies are already using it to track conversations in Slack, Gmail, or Notion and react when certain events occur.
Codex can also plan work for the future and reactivate yourself independently to complete long-term tasks. For example, it can manage a queue of code reviews for days or weeks, addressing comments and updating documentation without the developer having to be constantly monitoring it.
Over 90 add-ons and improved developer support
To complete the change, OpenAI has added nearly 90 new Codex pluginsThese include integrations with Microsoft Suite, GitLab Issues, Neon by Databricks, Render, and Superpowers, as well as support for MCP (Model Context Protocol) servers. The goal is to give the agent more context sources and greater operational capabilities.
Thanks to these plugins, Codex can gather information from different services and act upon it From a single workspace, you can review issues in GitLab, query data in a Databricks environment, update Office documents, or interact with deployment services, without forcing the user to jump from platform to platform.
On a purely technical level, the update strengthens the typical developer workflowsNow Codex helps you review pull requests, manage multiple terminal tabs at once, connect to remote development environments via SSH, and open files with rich previews, whether they are PDFs, spreadsheets, presentations, or documents in other formats.
All of this is concentrated in a unified experience that, according to OpenAI itself, allows for faster progress in all stages of the software life cycleFrom writing the initial code to testing, deploying, documenting and maintaining it, with AI agents continuously collaborating within the same desktop environment.
Availability, focus on developers, and situation in Europe
The new version of Codex is already being deployed in the desktop application for macOS For those who log in with a ChatGPT account. Not all features require the same type of subscription, and some advanced features may depend on the plan purchased, as is the case with other OpenAI tools.
Thibault Sottiaux, director of Codex, explained that this update It is especially geared towards developersAlthough the intention is to expand it to a more general audience later. The idea is that technical teams will be the first to take advantage of the new agents and automated workflows and, over time, transfer that model to office tasks, document management, or personal productivity.
OpenAI emphasizes that the use of computing power and memory is implemented with a security and regulatory compliance approachThis is especially relevant in the European Union and the United Kingdom. For this reason, advanced customization features, contextual suggestions, persistent memory, and direct operating system control will take a little longer to become available in these markets.
In any case, the company maintains that this evolution of Codex opens the door to a desktop super-application focused on intelligent automationwhere AI agents not only help write code, but also coordinate tasks, adapt to the user's work style, and keep complex projects running for extended periods. For software developers in Europe, the message is clear: new capabilities are coming, but with the necessary adjustments to fit within the existing regulatory framework.