OpenAI takes artificial intelligence even further with new Codex version

Although OpenAI has not yet implemented its desktop super application plans, which were brought to the agenda last month, it reveals the first concrete signs of this goal with the comprehensive update it published for Codex. The company’s new version is a preliminary preparation for a broader experience that will bring ChatGPT, Codex and Atlas browser under one roof. Currently, a unified application is not offered directly to the end user; Despite this, the Codex’s capabilities have been significantly expanded. This approach shows that OpenAI prefers to progress the development process in phases.

The focus of the new version is on developers. Thibault Sottiaux, the head of the Codex team, states that the update aims to strengthen the developer experience before opening it to a wider user base. In this context, Codex now offers multi-purpose artificial intelligence agents that can work in a wider range of uses and act more proactively. On the usage side, this change is particularly embodied in the ability to perform operations directly on the computer.

Codex’s new agents can interact with other applications on the computer. Users can choose a specific application directly, or they can leave it to Codex to choose the most suitable software according to the task to be done. Although similar features are also seen in competing solutions, OpenAI aims to differentiate itself with its infrastructure that allows simultaneous operation without straining system resources. In addition, the launch of 111 new plugins increases Codex’s ability to both collect contextual information and integrate with the tools used by developers.

Codex update expands with new tools and memory features

One of the striking elements of the update is the built-in browser and editing tools. While working on web pages or web applications, Codex can make changes to certain fields directly through comments. For example, in cases where the axes do not appear correctly in a graphic component, the user can make the necessary adjustments with the command. This approach offers a time-saving structure, especially in interface development processes.

In addition, image generation capabilities have also been expanded. Thanks to the gpt-image-1.5 model, Codex can create product sketches, interface mockups and simple game assets. The ability to verify transactions by analyzing screenshots also helps reduce the system’s margin of error. When all these are evaluated together, it can be seen that Codex has become a more comprehensive production platform rather than just a tool that writes code.

OpenAI started testing two new memory features with this release. Codex can now respond faster and more consistently to new requests by remembering the context from previous missions. In addition, the system can provide proactive suggestions to the user in line with the data it obtains. Practical examples of this approach include suggestions such as a reminder to respond to a comment that should be made at the beginning of the day.

The new Codex version is gradually being rolled out to desktop application users. The computer usage feature, which is initially accessible on macOS, will come to other platforms in the future. Users in the European Union and the United Kingdom have to wait a little longer for some features. Memory functions will also be gradually rolled out regionally.

Codex’s latest update reveals the technical basis for OpenAI’s plans to bring together different tools under a single application. Although current features are still developer-focused, the system is expected to evolve to appeal to a wider range of users. In this process, performance, ease of use and depth of integration will be among the determining factors.