OpenAI Revamps AI Agent Codex with PC Control, In-Browser Operation, and UI Mockups
The revamped AI agent can now control macOS apps, modify live websites, and schedule long-running tasks.
OpenAI has fundamentally expanded the capabilities of its Codex AI agent, moving it beyond a coding assistant into a general-purpose automation tool. The headline feature is 'PC operation,' which allows Codex to autonomously control applications on a user's macOS computer using its own independent cursor. This enables tasks like launching an app in Xcode, testing it for bugs, and automatically fixing errors—all from a single natural language prompt. The agent can now also run multiple instances in parallel without interfering with the user's own mouse.
Further enhancing its utility, Codex now includes a native in-app browser that lets users give direct instructions to modify specific elements on live web pages, such as changing text or fonts. For design workflows, it integrates GPT Image 1.5 to generate and embed images directly into mockups. Crucially, OpenAI released over 90 new plugins that connect Codex to external systems via MCP servers, dramatically expanding its action space. A new scheduling function allows Codex to manage long-term, recurring tasks—like monitoring Slack, Gmail, and GitHub for important updates every hour—effectively turning it into a persistent digital employee. This update is being rolled out gradually to Codex desktop app users.
- Codex can now autonomously control macOS apps using its own cursor to click, run, and debug software from a prompt.
- An integrated browser allows direct instruction for live webpage edits, and GPT Image 1.5 enables in-workflow mockup generation.
- Over 90 new plugins and a scheduling feature let Codex automate complex, multi-day tasks across tools like GitHub and Slack.
Why It Matters
This transforms Codex from a coding copilot into a versatile automation agent capable of managing entire software development and business workflows.