In a move that signals a profound shift in how humans interact with software, OpenAI has introduced a sweeping update to its Codex platform that allows the AI agent to operate a user’s computer in the background—clicking, typing, and navigating applications without interrupting ongoing work.
The update, rolling out first to macOS users, transforms Codex from a powerful coding assistant into something closer to a fully autonomous digital operator—one that can actively use software rather than simply suggest how to use it. According to a report by The Verge, Codex can now operate desktop apps on macOS in the background without disrupting user activity, allowing developers to test and iterate while continuing their own work.
At the center of this release is what OpenAI calls “computer use,” a capability that allows Codex to interact with desktop applications directly. The system can see what is on screen, move a cursor, type inputs, and execute tasks across virtually any app installed on the machine. OpenAI notes that Codex can now use all of the apps on your computer by seeing, clicking, and typing with its own cursor.
Crucially, it does all of this in the background.
Unlike earlier generations of AI agents that required users to cede control of their device, Codex can now operate in parallel. Multiple agents can work simultaneously, enabling developers to delegate tasks such as running tests, reviewing code, or iterating on frontend changes without micromanagement. This parallel execution is particularly useful for workflows involving applications that lack traditional APIs.
This shift may seem incremental, but it marks a deeper philosophical leap. AI is no longer confined to chat windows or development environments. It is beginning to act directly within the operating system itself.
From assistant to operator
Codex was initially introduced as an AI system capable of writing and debugging code, helping developers accelerate routine tasks. But this update expands its role far beyond code generation.
With the new features, Codex can now manage entire workflows across the software development lifecycle—from reviewing pull requests and running terminal commands to testing front-end designs in a built-in browser. It also integrates with a growing ecosystem of tools, enabling it to gather context and act across multiple environments.
More strikingly, Codex is now able to run multiple agents simultaneously. Each agent can work independently on different tasks, allowing developers to supervise rather than execute. OpenAI describes the app as a “command center for agents,” reflecting a shift toward orchestration rather than direct interaction.
Automation that persists over time
The update also introduces a deeper layer of automation—one that extends beyond a single session.
Codex can now remember user preferences, workflows, and prior corrections, building a persistent understanding of how an individual works. It can resume tasks after interruptions and continue long-running processes with minimal input.
As AI systems become more embedded in everyday workflows, this marks a major leap in latest AI updates and automation tools shaping the future of software development.
Beyond coding: a “super app” vision
OpenAI’s ambitions for Codex appear to extend well beyond software engineering.
The platform now includes image generation capabilities, allowing users to create visual assets, mockups, and design elements directly within the same workflow. Combined with an in-app browser, Codex can interact with web content, comment on pages, and assist in building applications in real time.
Taken together, these features suggest a push toward a unified environment where coding, design, and automation converge. This reflects OpenAI’s expanding influence beyond software tools, as the company positions Codex as a central interface for digital work.
The competitive landscape is also intensifying. Codex operates in a growing market of AI coding tools, competing with platforms like Anthropic’s Claude Code. As the technology evolves, it is increasingly shaped by the global race for AI infrastructure and computing power, which continues to redefine the boundaries of what these systems can achieve.
The implications for work
The implications of this shift extend far beyond software development.
By enabling AI to directly interact with applications, OpenAI is effectively removing the boundary between suggestion and execution. Tasks that once required human input—opening programs, navigating interfaces, copying data—can now be delegated entirely.
For developers, this could mean faster iteration cycles and reduced cognitive load. For organizations, it could translate into significant productivity gains. But it also raises new questions about control, security, and trust in systems that can act independently.
OpenAI has emphasized that Codex operates within defined permissions and environments, but the broader implications of autonomous software agents are still unfolding.
A turning point for AI interfaces
If earlier waves of AI were defined by conversation, this one may be defined by action.
Codex’s new capabilities suggest that the future of AI is not just about generating answers, but about doing the work itself—quietly, continuously, and often invisibly in the background.
And in that future, the computer may no longer be something we use.
It may be something that works for us.
