Claude no longer separates conversation from delegated work

Anthropic is merging Claude Chat and Cowork into a single interface, capable of running background tasks, creating documents and presentations, and then editing them with the user.

A quick question can now turn into a report and then a presentation without leaving the conversation or selecting another mode. Anthropic is bringing Claude Chat and Claude Cowork together so its assistant can determine whether a request calls for a simple response or several stages of work.

The new experience is gradually removing the buttons that separated Chat from Cowork. Users describe the desired outcome, the sources to consult, and the delivery format. Claude can then search for information, read files, run code, use connected services, and produce the deliverable within the same conversation.

The change primarily concerns the organization of the product. Anthropic is not introducing a new model or additional reasoning capabilities measured through a new evaluation. The company is making the agentic functions previously confined to Cowork accessible from chat.

Chat was mainly used for immediate responses, while Cowork handled longer assignments capable of using files and tools. This distinction required users to choose the right environment before they even knew how far the request might develop.

Work started in one environment did not flow naturally into the other either. Research conducted in Cowork could lead to a separate conversation for rewriting the findings, followed by another workspace for preparing the visual material.

The merger turns the conversation into a shared entry point. A request can remain brief or expand as Claude identifies the operations it requires. The system can also ask questions when a deadline, source, or format is missing.

A company could, for example, ask what changed in its sales pipeline during the week, have delays flagged, produce a report in its usual style, and then convert the main findings into five slides. The document and presentation share the same context instead of being prepared from separate instructions.

Claude can run several tasks at once. A conversation devoted to a report should therefore not prevent another assignment from progressing in parallel. This capability remains subject to the plan’s usage limits and the resources required by each task.

Longer operations can continue in the cloud after a tab is closed or the computer is turned off. Users can later find the result in the conversation and monitor progress from the mobile app.

This continuity comes with an important limitation. A task based solely on the web, files saved to the account, and remote connectors can continue without the computer. As soon as it needs to reach a local folder, operate an installed application, or use the machine’s browser, Claude Desktop must remain open.

The claim that Claude continues working after the computer has been closed therefore does not apply to every operation. A report drawing from Slack or Google Drive can run remotely. Sorting a folder stored on the user’s Mac must wait until the computer and desktop app are available.

Recurring tasks also benefit from remote execution. Claude can prepare a summary of received messages every Monday, compile ongoing monitoring, query a calendar, or produce a report from connected services.

Each automation retains its instructions, schedule, and required permissions. Past and upcoming runs appear in a Scheduled section, where the task can be edited, paused, run manually, or deleted.

An automation running in the cloud cannot, however, be linked to a folder that exists only on the computer. If it depends on local files or an installed application, it must run locally and is subject to the same availability constraints as other desktop tasks.

The merger also introduces two new creation formats. Claude Docs is designed for evolving documents, while Claude Slides produces presentations. Claude Design, which was already available separately, is also moving into conversations.

A document can be requested using previous messages, a project, memory, files, or connected services. Claude asks a few questions before preparing it, writes inside the interface, and leaves comments to explain certain choices or request clarification.

Users can write directly in the document while Claude is working. Changes are saved automatically, and anyone with editing permission can contribute in real time.

A passage can also receive a comment mentioning `@Claude`. The assistant replies in the thread, applies the requested edit, and explains what it changed. This brings Claude Docs closer to a collaborative workspace than a file simply generated at the end of a conversation.

Documents can include rich text, tables, multiple tabs, charts, timelines, and diagrams. Visuals created from a connected source do not update automatically, however. Claude must be asked again to retrieve the latest data.

Creations are private by default. On Pro and Max plans, they can be opened by anyone with the link. On Team and Enterprise plans, the owner selects members or groups within the organization, but external sharing is not yet supported.

Viewing a shared document requires a Claude account. There is also no comment-only permission level: a person is either a viewer or an editor. Only the owner can rename or delete the document, and deletion is permanent, with no trash folder.

Claude Docs can export results to Word, PDF, Markdown, or Google Docs. A document can also be converted into a presentation with Claude Slides.

Slides builds the presentation from the context already available in the conversation. The result can be edited in Claude, presented directly, shared through a link, or downloaded as a PowerPoint or PDF file.

The value is not limited to generating a sequence of slides. A change made to the report can be carried over to the presentation without rebuilding the context. Users can also select an element, move it, or request an adjustment in natural language.

Claude Design covers mockups, visual summaries, interactive prototypes, and one-page materials. It combines a conversation with a creation area where text, dimensions, alignment, colors, and components