OpenAI has rolled out a major Codex update that pushes the tool beyond code generation into a full workflow assistant. Used by more than 3 million developers weekly, Codex now handles tasks across the software development lifecycle, from writing and reviewing code to managing ongoing work across apps and services.
This update shifts Codex closer to an always-on development partner. It can now interact with desktop apps, run multiple agents in parallel, and pull context from connected tools. The result is a system that reduces context switching and keeps projects moving without constant manual input.
One of the biggest changes is support for more than 90 plugins. These integrations connect Codex to tools like GitHub, Jira, Slack, and CI platforms, allowing it to gather context and take action across workflows. Developers can use this to triage pull requests, manage deployments, update documentation, and coordinate across teams without leaving the Codex environment.
Codex can now also operate your computer directly. On macOS, it can see what is on screen, click, and type using its own cursor. Multiple agents can run simultaneously without interrupting your work, which makes it useful for testing apps, iterating on frontend changes, or handling tools that do not offer APIs.
The app now includes an in-app browser designed for development workflows. You can preview web projects, leave comments directly on pages, and have Codex apply changes in real time. This tight loop between feedback and execution makes it easier to refine interfaces, games, and web apps without switching tools.
Image generation is now built in using GPT Image 1.5. Developers can create and iterate on visuals such as mockups, UI concepts, and game assets within the same workflow. Combined with screenshots and code context, this makes Codex useful beyond traditional programming tasks.
Workflow improvements extend to core developer features as well. Codex now supports reviewing GitHub comments, running multiple terminal tabs, connecting to remote environments over SSH, and previewing files like PDFs and spreadsheets in a sidebar. A new summary pane tracks agent actions, sources, and outputs, making it easier to follow what the system is doing.
Automation is another key part of this update. Codex can now reuse previous conversations, schedule tasks, and continue work over time. It can resume long-running projects across days or weeks and even suggest what to work on next based on project context, memory, and connected tools.
Memory is currently in preview and allows Codex to remember preferences, workflows, and past corrections. This reduces the need for repeated instructions and improves output quality over time. Alongside this, Codex can proactively surface tasks, such as unresolved comments or pending updates across tools like Slack, Notion, or shared documents.
These features are rolling out to Codex desktop users signed in with ChatGPT. Some capabilities, including memory and personalization, are expected to expand to enterprise and regional users in upcoming releases, while desktop-level control is currently limited to macOS.
(via OpenAI)
