OpenAI is reportedly expanding its Codex AI coding agent with new “computer use” capabilities that let it interact directly with macOS applications in a much more hands-on way.
What’s new in Codex
With the update, Codex can now:
-
Operate desktop Mac apps directly
-
Use its own on-screen cursor
-
See what’s displayed on the screen
-
Click, type, and navigate interfaces
-
Complete multi-step tasks inside native applications
What this means in practice
Instead of being limited to generating code or working in a terminal-style environment, Codex is moving toward being able to:
-
Interact with real software interfaces (like a human user would)
-
Perform end-to-end workflows inside macOS apps
-
Combine code generation with actual execution inside graphical tools
Bigger picture
This shift places Codex closer to a category of AI systems that act as autonomous desktop agents, where the model doesn’t just suggest actions—it can directly carry them out across apps and interfaces.
It also reflects a broader industry trend: AI tools increasingly moving from text-based assistants into full computer-controlling agents that can operate operating systems more like human users.
OpenAI upgrades Codex with desktop automation and memory features
OpenAI has significantly expanded Codex, its AI coding agent, adding deeper macOS integration, long-term memory, and multi-agent automation capabilities.
Key new capabilities
Desktop control (“computer use”)
Codex can now:
-
Run multiple agents in parallel on Mac
-
Operate without interfering with the user’s work
-
Use an on-screen cursor to:
-
See what’s on display
-
Click and type in apps
-
Navigate macOS software like a user
-
This makes Codex capable of interacting directly with real desktop environments, not just code or terminals.
Memory and personalization
Codex can now remember:
-
User preferences
-
Recurring workflows
-
Tech stacks
-
Project-specific context
It can also:
-
Resume paused work using existing chat threads
-
Schedule tasks across days or weeks
-
Propose work automatically based on project context, plugins, and stored memory
Automation and workflow improvements
New automation features allow Codex to:
-
Run tasks continuously over long periods
-
Coordinate multi-step development work
-
Handle asynchronous execution and task resumption
-
Suggest next steps based on project history
Browser and interaction upgrades
Codex now includes an in-app browser with:
-
Direct commenting on web pages for precise instructions
-
Upcoming full browser control for:
-
Opening websites
-
Navigating user flows
-
Taking screenshots
-
Inspecting UI outputs
-
Developer-focused enhancements
OpenAI says Codex is now more useful for:
-
App testing and QA workflows
-
Iterating on frontend design changes
-
Managing complex development pipelines
Additional updates include:
-
Multiple terminal tabs
-
Sidebar file previews (PDFs, spreadsheets, etc.)
-
GitHub review comment handling
-
Over 90 new plugins (skills, integrations, MCP servers)
New image generation support
Codex now uses gpt-image-1.5 to generate images, helping with:
-
Product concepts
-
UI mockups
-
Visual design prototypes
Availability
-
Rolling out to Codex desktop users signed in with ChatGPT
-
Not yet available for:
-
Enterprise (partially)
-
Education users
-
EU and UK users (computer use features also restricted)
-
Bottom line
Codex is evolving from a coding assistant into a full desktop automation system, capable of running parallel agents, interacting with real apps, and managing long-term development workflows with memory and scheduling.