OpenAI has released a major update for Codex, the AI assistant used by developers. With this update, the tool’s role shifts from a programming aid to a broader digital assistant that supports a larger part of the software development process.
The new features are being rolled out in phases. For example, in the updated version, Codex can perform actions on the user’s computer, such as clicking and typing, but this feature is currently only available on macOS and not yet in Europe and the United Kingdom. Features such as personalization and memory are also expected to be added there later. At the same time, OpenAI is stepping up its competition with Anthropic, whose tool Claude Code is considered a market leader by many companies, according to SiliconANGLE.
Whereas Codex was previously used primarily for generating and analyzing code, the software can now also perform tasks on the user’s own computer. The application can take over actions such as clicking and typing, operating in the background while the user continues their own work. According to OpenAI, this makes it possible to test applications or implement changes to interfaces, even when no direct API connections are available. Multiple agents can be active simultaneously without disrupting other applications.
In addition, Codex will see stronger integration with other software tools. The update introduces support for services commonly used within development teams, such as code management, project management, and cloud infrastructure. This should enable the AI to better gather context and perform actions across different platforms.
Expansion of tooling and work environment
The capabilities have also been expanded within the application’s own environment. Developers can view multiple files and terminal windows simultaneously, review pull requests, and connect to external development environments via SSH. A built-in browser should make it easier to develop and test front-end applications and games directly. In doing so, Codex is following similar steps to its competitor Claude Code, which recently gained comparable functionality.
Another addition is the ability to work with visual content. Codex can generate and modify images within the same workflow used for writing code. This should accelerate the development of visual concepts and interfaces.
The update explicitly focuses on longer-term and recurring work. Users can schedule tasks that Codex will tackle independently at a later time. In doing so, the software can build on previous conversations and retain context. Through integrations with external tools, Codex can analyze messages from Slack and calendar appointments. Based on this, it can compile to-do lists, so users spend less time on planning.
In a subsequent phase, the company will also introduce memory functionality. This will allow Codex to remember preferences and previously collected information. According to OpenAI, this should ensure that future tasks are executed faster and more consistently, without the need for extensive instructions each time.
The updated version of Codex can also make suggestions for follow-up actions on its own. By combining information from projects, linked tools, and previous interactions, the software can make suggestions regarding priorities or pending tasks.