OpenAI Ships Codex Desktop Overhaul: Background Computer Use with Parallel Agents, 90+ Plugins

OpenAI just turned Codex from a coding assistant into a desktop operating system for AI agents.

The April 16 update adds background computer use. Multiple agents can click, type, and navigate macOS apps via their own cursors without interrupting your workflow. You keep working. They keep working. In parallel. On different apps.

That's a capability jump, not an incremental update. The previous model was "ask the agent, wait for the agent, review the agent." Now it's "dispatch three agents to three different tasks and check on them when you're ready."

90+ new plugins shipped alongside it: Atlassian Rovo (JIRA), CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon by Databricks, Remotion, and Render. A new in-app browser lets you annotate web pages to guide the agent. And GPT-5.3-Codex-Spark runs at 1,000+ tokens per second for rapid iteration loops.

3 million weekly developers use Codex. This update changes what "using a coding agent" means for all of them. It's not pair programming anymore. It's delegation.

I keep thinking about the convergence here. Anthropic ships Ultraplan (plan in cloud, review in browser, execute locally). GitHub ships cloud-based coding agents. Now OpenAI ships desktop agents that operate independently. The shared pattern: the terminal becomes a dispatch interface, not the execution environment. Planning and execution are decoupling. Builders who architect for this split will move faster than those coupling everything to one shell session.

The plugin ecosystem is the real platform play. When your coding agent can create JIRA tickets, trigger CI pipelines, open GitLab issues, and deploy to Render without leaving the conversation, you're not using a tool. You're orchestrating a workflow. Not available in EEA/UK/Switzerland at launch, which is a nontrivial limitation for distributed teams.