OpenAI ships Codex desktop plugins, recasting the assistant as a workspace

OpenAI didn’t just add a few integrations to Codex. It changed what Codex is supposed to be. The April 16 update turns the desktop app from a coding sidekick into a workspace that can see your tools, move across them, and keep long-running jobs alive over time. That matters because the bottleneck in AI coding is no longer just “can the model write code?” It’s whether the assistant can stay grounded in the mess of real work — repos, terminals, tickets, docs, browsers, and approvals. (openai.com) ### What actually shipped? OpenAI added background computer use, an in-app browser, richer developer controls, image generation, memory, automations, and more than 90 new plugins to the Codex app for macOS and Windows. The company also pushed deeper workflow support — multiple terminal tabs, PR review handling, file previews, a summary pane, and SSH connections to remote devboxes in alpha. (openai.com) ### Why are plugins the real story? Because plugins change Codex from “answer my prompt” into “go do the job.” In OpenAI’s setup, a plugin can bundle instructions, app connections, and MCP servers into a reusable workflow. That means Codex is no longer limited to the files in front of it. It can pull context from Slack, Drive, GitHub, Jira-like systems, Microsoft tools, and other external services, then act inside them. (developers.openai.com) ### Why does desktop control matter so much? A lot of software work still lives outside clean APIs. Frontend testing, bug reproduction, local dev servers, internal tools, and random desktop apps all break the neat “model plus repo” fantasy. OpenAI’s answer is background computer use — Codex can see, click, and type with its own cursor while you keep working (developers.openai.com)ion usually dies. (openai.com) ### What’s the browser doing here? The browser is less about web search and more about closing the loop between code and interface. Codex can open pages inside the app, and you can comment directly on the page to steer the agent. For frontend work, that’s a big deal — instead of describing a visual bug in text, you can point at the thing that looks wrong. Ope(openai.com)or local tasks. (openai.com) ### Is this still a coding product? Yes, but it’s clearly stretching past that box. OpenAI’s own framing is that developers are now supervising multiple agents across longer jobs, and existing IDEs or terminal tools don’t fit that pattern very well. So Codex becomes a command center — one place to run parallel threads, manage context, inspect outputs, and han(openai.com)ategory from autocomplete with chat attached. (openai.com) ### What changes for teams buying this stuff? Model quality still matters, but turns out it’s no longer the whole buying decision. Once an assistant can touch files, terminals, browsers, tickets, and third-party apps, the hard questions shift to access boundaries. Which plugins are enabled? What data gets sent to outside services? When does the agent need appro(openai.com)own auth, privacy, and data-sharing rules, which means governance now sits in the middle of the product experience. (developers.openai.com) ### Why now? OpenAI says more than 3 million developers already use Codex weekly, and the company has been building toward this since the Codex app launch in February and its Windows expansion in March. The update looks like a response to a broader market truth — coding agents are converging on the same raw capability, so the next fight is over workflow owne(developers.openai.com)ch stronger moat than whoever just writes the nicest snippet. (openai.com) ### Bottom line? The important shift is conceptual. Codex is no longer being pitched as a smarter chat box for code. It’s being pitched as the layer that sits on top of your tools, pulls in live context, and orchestrates work across them. If that sticks, the winning assistant won’t just be the one with the best answers — it’ll be the one that becomes your workspace. (openai.com)