The battle between the two leading AI developers seems to never stop. The newest chapter: OpenAI has released a major update to its Codex platform, repositioning the tool from a coding assistant into an automation layer operating across a developer’s environment.
The release is clearly an effort to keep pace with Anthropic, whose multi-faceted Claude Code is gaining big traction in the lucrative enterprise sector.
The most important improvement is Codex’s ability to act directly within a user’s operating system. The platform can now interact with applications by issuing commands that simulate user behavior, including opening programs and typing inputs. On macOS systems, multiple AI agents can execute tasks in the background while the user works, a true advance from the interactions typical of earlier AI tools.
“Codex gaining computer use, persistent memory, and autonomous scheduling moves coding agents past the IDE boundary,” said Mitch Ashley, VP Practice Lead at the Futurum Group. “The center of gravity shifts from code generation to system operation, with agent actions persisting across sessions and days.”
Built-In Browser
The update also introduces a built-in browser, allowing Codex to interact with web-based environments. Developers can markup elements on a page and direct the agent to modify or test them. This capability targets frontend and game development use cases, where rapid iteration across visual elements is essential.
More than 90 plugins connect Codex to services such as project management platforms and CI/CD pipelines. These integrations allow Codex to gather context across workflows and take action within those tools. For example, Codex can review messages and assemble prioritized worklists.
Codex can now schedule work, resume incomplete processes, and maintain continuity across sessions. This provides ongoing agency, where the system operates over extended periods rather than responding to continued prompts. Early use cases include monitoring collaboration tools for updates and automatically advancing software tasks like pull request management.
Supporting this continuity is a new memory function, currently in preview. The system retains user preferences and accumulated context, including prior corrections. Over time, this reduces the need for repeated instructions and allows the agent to refine its output based on past interactions.
Codex also incorporates image generation into the development workflow. By integrating OpenAI’s image model, the platform produces visual assets such as mockups and design elements alongside code. This unifies tasks that typically require separate tools.
Super App on the Way
OpenAI has indicated that this approach represents an incremental path toward a more comprehensive so-called super app, whose release date is not certain. The super app is expected to combine development, communication, and browsing into a single environment.
Anthropic has introduced similar capabilities, including remote system control and persistent workflows. However, OpenAI’s emphasis on simultaneous background operation differentiates its approach, particularly for users managing parallel tasks.
For dev teams and company executives, the advances of Codex and Claude Code offer a new world of productivity boosts, where agentic tools manage workflows end-to-end. However, these autonomous systems raise deep concerns about oversight and governance as AI agents become enmeshed in more layers of business applications.
