OpenAI is rolling out a major update to Codex, framing it as a partner for the more than 3 million developers who use it weekly across the software lifecycle—not only for typing code. The desktop app can now drive more of your machine and toolchain, generate images inside the same workflow, remember context, and keep recurring work moving with richer automations.
The through-line is familiar across the industry: coding assistants are becoming ambient agents that sit next to IDEs, browsers, terminals, and team tools. OpenAI’s pitch is that Codex should follow you from implementation to review, testing, and follow-ups without constant context switching.
Background Computer Use and an In-App Browser
With background computer use, Codex can see, click, and type in macOS apps using its own cursor while you keep working elsewhere. OpenAI says multiple agents can run in parallel on a Mac without fighting your foreground apps—useful for iterating on frontends, exercising local apps, or automating tools that lack APIs.
The app also gains an in-app browser where you can comment on pages to steer the agent precisely. OpenAI highlights frontend and game workflows today and notes plans to expand toward fuller browser control beyond localhost over time.
Images, Plugins, and the Long Tail of Integrations
Codex can use gpt-image-1.5 to generate and revise images alongside screenshots and code—pitched for product concepts, UI mockups, and game assets in one workspace.
OpenAI is also shipping more than 90 additional plugins that bundle skills, app integrations, and MCP servers. Named examples include Atlassian Rovo (Jira), CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon by Databricks, Remotion, Render, and Superpowers—the pattern is letting Codex pull context and take actions across the tools teams already live in.
Developer Workflow in the App
For day-to-day engineering, the desktop app adds GitHub pull-request review comment support, multiple terminal tabs, and alpha SSH access to remote devboxes. A sidebar opens files with previews for PDFs, spreadsheets, slides, and docs, plus a summary pane for agent plans, sources, and artifacts—aimed at tightening the loop between writing code, checking output, and collaborating with the agent.
Automations, Memory, and Proactive Suggestions
Automations can now reuse existing conversation threads so prior context carries forward. Codex can schedule its own future work and wake up to continue long-running tasks—OpenAI suggests timelines stretching across days or weeks for things like landing open PRs or following up across Slack, Gmail, and Notion.
A preview of memory lets Codex retain useful context from past sessions—preferences, corrections, and facts that were expensive to establish—so later tasks start closer to “already onboarded.” Separately, Codex can propose what to do next from projects, plugins, and memory (for example surfacing open Google Docs comments and pulling Slack, Notion, and repo context into a prioritized list).
Availability and Boundaries
Updates are rolling out to Codex desktop users signed in with ChatGPT. Personalization (context-aware suggestions and memory) is coming to Enterprise, Edu, and EU/UK users soon. Computer use launches first on macOS, with EU and UK to follow.
For web teams, the practical read is that OpenAI is compressing “terminal + editor + browser + ticket tracker” into one agent surface—competition for Cursor-style IDEs, Claude Code, and vendor-specific copilots, with the usual caveats around trust, permissions, and cost once agents touch production systems and company data.