OpenAI folds Sites, a built-in browser, and Computer Use into ChatGPT Work

Generated interfaces become a first-class output

The headline capability of ChatGPT Work is that it produces finished artifacts — the announcement lists sheets, slides, docs, and web apps — rather than returning text for a person to assemble. That reframes ChatGPT's output from a chat transcript into deliverables that follow your templates and reference files, powered by the newly released GPT-5.6.

For frontend work, the most concrete addition is Sites, launching in public beta. OpenAI describes it as a way to turn work or ideas into an interactive site or web app, shareable by URL, with examples including live dashboards, project trackers, launch calendars, prototypes, internal portals, and interactive reports.

Two details matter here. First, you can test the Sites you build inside ChatGPT. Second, ChatGPT can update them as the underlying information changes — pairing the generated UI with the Scheduled Tasks that keep data fresh. That combination is closer to a living internal tool than a one-off artifact.

The browser moves inside the app

OpenAI is putting a built-in browser directly in the desktop app so ChatGPT can research a market, compare sources, pull information from websites, and open files from Google Workspace and Microsoft 365 in one place. On desktop, Computer Use adds the ability to click, type, and move files across apps, tools, and the browser on the user's behalf.

The Chrome extension is also being updated so ChatGPT runs in Chrome's sidebar. The clearest signal, though, is the retirement of a product: OpenAI says it will begin sunsetting the standalone Atlas browser and help users transition to ChatGPT.

These capabilities build on what we learned from Atlas and from the users who helped us understand how agentic tools can make browser-based work more useful.Montana Labs

Rather than maintain a separate browser as the surface for agentic web interaction, OpenAI is absorbing that function into the ChatGPT desktop app and the Chrome sidebar. The web-interaction layer is being consolidated, not shipped as its own destination.

Consolidation on desktop: Chat, Work, and Codex in one shell

The Codex app is merging into the new ChatGPT desktop app, which now hosts Chat, Work, and Codex on every plan, including Free, on Mac and Windows. The previous desktop app is renamed ChatGPT Classic, and existing Codex users update in place.

OpenAI positions Codex technology as the engine behind ChatGPT Work's ability to get real work done, citing more than 5 million weekly Codex users and more than 1 million using it outside software development. Codex-specific frontend features also ship here: inline editing within diffs, pull request review in the side panel, faster computer use, and multi-repository projects.

For developers, the practical effect is that the coding agent and the general work agent now share a window and a governance model — desktop ChatGPT Work builds on Codex's enterprise governance and admin controls for local files, apps, browsers, and tools.

What a consolidated ChatGPT surface means for building internal tools

The specific implication for teams that build interfaces is that OpenAI is offering a path where the dashboard, the browser interaction that feeds it, and the schedule that keeps it current all live inside one product. Sites plus Scheduled Tasks plus Computer Use is, functionally, a way to stand up a self-updating internal tool without a separate frontend stack.

That is worth weighing against the constraints stated in the announcement. Sites is public beta. ChatGPT Work uses a Codex-style usage structure where more complex tasks consume more of a plan, and Enterprise and Edu admins manage capacity through workspace defaults, group limits, and per-user overrides. Auto-review gates important actions on connected tools before they run.

The bet OpenAI is making is that fewer people will build a small internal dashboard from scratch when ChatGPT can generate, host, and refresh one from connected data. Whether generated Sites hold up as durable interfaces — versioned, testable, and reliable enough to depend on — is the open question the beta will answer.