OpenAI Turned Codex Desktop Into a PC Agent With Browser, Memory, and Plugins

OpenAI has essentially rebuilt Codex Desktop from a code-writing tool into a management layer for your work computer. After a major update, the application can do more than just help with development—it can see the interface, click, type, work with a browser inside the app, and continue long tasks without constant user involvement. OpenAI presented the update on April 16, 2026.

According to the company, Codex is already used by over 3 million developers per week, and the new release is clearly designed to keep them within a single workspace. Instead of juggling an IDE, terminal, browser, task tracker, and separate automations, OpenAI offers a unified hub where developers can delegate not only code but also related actions to an agent. ZDNet describes this as an attempt to turn Codex from a code tool into a full-fledged work hub.

The desktop application is already available on Mac and Windows, but the most ambitious features are rolling out gradually. The main innovation is the background computer control mode. Now Codex can work with applications on your device using its own cursor: see the screen, click buttons, and type text.

OpenAI specifically emphasizes that multiple agents can work in parallel on Mac without interfering with what the user is doing at that moment. The practical scenario here is clear: a frontend developer edits the interface, while the agent simultaneously opens the application, checks the changes, repeats test actions, and returns the result. This is particularly important for programs that lack proper APIs and where automation previously hit a dead end due to manual work.

The second notable update is the built-in browser. Inside Codex, you can now open pages and leave comments directly on interface elements to more precisely explain to the agent what needs fixing. For now, OpenAI is positioning this primarily for frontend, applications, and games, especially when you need to quickly cycle through "look at screen—give instruction—get new version."

Additionally, Codex learned to work with image generation through gpt-image-1.5, so in one workflow you can assemble a mockup, write code, and immediately generate the needed visuals. The third layer of changes concerns the developer workflow.

Codex Desktop now knows how to parse comments from GitHub reviews, manage multiple terminal tabs, connect to remote devbox environments via SSH in alpha mode, and show previews of PDFs, spreadsheets, slides, and documents right in the side panel. A summary panel has appeared where you can track the agent's plan, sources used, and artifacts. Separately, OpenAI added over 90 new plugins that combine skills, application integrations, and MCP servers.

This isn't for catalog aesthetics, but so the agent can pull context and perform actions in JIRA, CI/CD, documents, cloud services, and other work tools without constantly switching between windows. No less important part of the release is automations and memory. Now automations can be tied to existing threads so Codex picks up accumulated context rather than starting the task over each time it runs.

The agent is also allowed to plan work for the future: it can "wake up" later and continue a long task days or even weeks later. A memory feature appeared in preview mode that saves user preferences, corrections, and information that previously took time to collect. On top of this, Codex started proactively suggesting what to do next: for example, point out unresolved comments, pull related context from connected tools, and provide a list of next actions.

That said, OpenAI clarifies that computer control mode is initially only available on macOS, and some personalized features are still rolling out to enterprise users and certain regions. What does this mean in practice? Codex Desktop doesn't yet look like a direct replacement for traditional software.

It doesn't replace IDEs, browsers, task trackers, and specialized applications—rather, it begins to act as a coordination layer above them. But the direction is clear: OpenAI is moving the product from "an assistant that writes code" to "an agent that manages workflow." If computer control proves reliable and memory and plugins don't break on real tasks, Codex could become not just another AI tool in the stack, but the interface through which that entire stack is used.