Image: OpenAI / MacRumors OpenAI Merges Codex Into ChatGPT and Launches a New Work Agent
OpenAI unified its Codex and ChatGPT desktop apps into a single application on July 9, and launched ChatGPT Work — a GPT-5.6-powered agent that completes multi-step projects across your connected apps.
OpenAI shipped two things at once on July 9: a unified desktop app that absorbs Codex, and a new agent called ChatGPT Work that takes on multi-step knowledge projects across your connected tools.
The two are related but separate. Codex isn’t going away. It’s becoming a tab inside a bigger app.
What changed with the desktop app
If you had the standalone Codex app installed, you’ve seen (or will see) a prompt: “Codex is now the ChatGPT app.” Updating keeps your projects and settings intact and switches the app to include Chat, Work, and Codex as distinct modes within one interface.
The previous ChatGPT desktop app is now called “ChatGPT Classic” and stays available separately for anyone who prefers it.
Codex itself picks up a few improvements in the merge: PR review now lives in a sidebar panel, and you can add multiple repositories to a single project rather than switching between them.
ChatGPT Work
ChatGPT Work is the new addition. You give it an outcome, it pulls context from your connected apps, breaks the job into steps, and works through them. OpenAI says it can stay on complex projects for hours.
The agent includes a Plan mode where it shows you a step-by-step breakdown before doing anything, configurable check-ins so you can decide how often it surfaces for input, and action approvals for anything consequential. Computer Use handles cross-app automation for cases where no direct integration exists.
At launch there are 15+ integrations, directed by @-mention syntax. The confirmed list includes Google Calendar, Slack, Gmail, Microsoft Teams, Google Drive, SharePoint, Salesforce, GitHub, Canva, Dropbox, Zoom, LinkedIn, Adobe, and Outlook.
There’s also a Sites feature in public beta: the agent can produce interactive web apps from your data, intended for reports and dashboards that need to stay live rather than sitting in a static document.
Scheduled Tasks round it out — you can set Work to run at fixed times, on recurring schedules, in response to events, or when something in a connected source changes.
Model tiers and availability
ChatGPT Work runs on GPT-5.6. Pro and Enterprise users get Sol; free and Go users get Terra. Luna isn’t part of the Work experience yet.
Ultra effort, which uses subagents to break down harder tasks, is available to Pro users for Work and to Plus users within Codex.
As of July 9, Work is live for Pro, Enterprise, and Edu plans. Plus and Business subscribers get access in the coming days. Free and Go are excluded from the Work agent.
The picture for developers
Codex remains the coding-focused mode inside the new app, with the same repo-aware context, diff review, and terminal capabilities it had before. The practical framing OpenAI offered: use Work to gather distributed business context (emails, docs, Slack threads, CRM data) and produce deliverables from it, then hand the implementation piece to Codex.
Whether that boundary holds cleanly in practice is something developers will find out as they use it. The underlying model is the same either way.
Sources: OpenAI ChatGPT Work announcement · 9to5Mac · MacRumors