A quiet but intense competition is unfolding between OpenAI and Anthropic, with both companies striving to develop more convenient and powerful AI programming tools. Currently, Anthropic appears to hold an edge, as Claude Code has been designated the preferred tool by numerous enterprises. However, OpenAI is not conceding defeat.

This week, OpenAI announced a comprehensive update to its automated programming tool, Codex, introducing a suite of new features designed to significantly expand its capabilities and permissions.

On Thursday, the company detailed numerous new characteristics and upgrades. Perhaps the most notable is that Codex can now operate in the background on a computer—capable of opening any application on the desktop and executing actions by simulating mouse clicks and keyboard inputs.

In a blog post, OpenAI stated that, functionally, this update allows Codex to deploy multiple AI agents. All of these agents can "run in parallel on a user's Mac without interfering with the user's work in other applications." In other words, thanks to the background operation mode, users can continue to use their computers normally even while these intelligent agents are performing tasks. The company described these agents as programming assistants that can handle auxiliary work concurrently while the user focuses on a core project. OpenAI listed typical application scenarios for such intelligent assistance, including iterative front-end modifications, application testing, and operating within applications that lack open API interfaces.

Overall, this upgrade to intelligent agents, along with other new features, underscores OpenAI's ambition: not only to make Codex a competitive programming assistant but also to transform it into a versatile tool deeply integrated into various enterprise workflows.

Observers of the AI programming space will note that some of the new functionalities added to Codex bear a resemblance to features previously launched by Anthropic for Claude Code. Last month, Anthropic announced that Claude, paired with its Cowork tool, could remotely control a user's Mac device and desktop when the user is away from their computer.

Beyond the intelligent agent tool, the new version of OpenAI Codex also includes a built-in in-app browser. Users can issue commands to the intelligent agent to perform specific operations within designated web applications. OpenAI indicated that this feature is particularly useful for front-end development and game development, with plans to further expand its capabilities to allow Codex to "fully control the browser, rather than being limited to web apps on the local machine."

Codex has received additional updates. A new preview feature called "Memory" enables Codex to retain information from past work sessions, generating key contextual information tailored to the user's work habits. The intelligent agent also gains image generation capabilities, which OpenAI says can be used to create product concept images, presentation visuals, design prototypes, placeholder graphics, and other business-related materials. Finally, to further enhance Codex's execution efficiency, OpenAI announced integration with 111 plugins, covering applications like CodeRabbit and GitLab Issues, enabling Codex to leverage these tools to complete relevant tasks.

OpenAI stated that these plugins allow Codex to handle simple administrative tasks and assist users in organizing their work. For example, if a user wants Codex to review a Slack channel and a Google Calendar to generate a to-do list for the day, the tool can now accomplish this.

OpenAI has also introduced a new on-demand pricing plan for Codex, available to ChatGPT Enterprise and Team users, clearly intended to offer greater flexibility in using this programming tool.

OpenAI, once the undisputed leader in the field, has seen its competition with Anthropic intensify in recent months. The company's focus has shifted towards enterprise-grade features, while development on consumer-facing tools like the social video application Sora 2 has slowed. Additionally, the company has been embroiled in several controversies recently, including lawsuits alleging that ChatGPT has negatively impacted the mental health of some users.