Anthropic folds Cowork into Claude and adds Docs and Slides

What shipped

Cowork’s switcher will disappear as individual accounts are updated. Claude will decide when a conversation needs extended work, without asking users to move their files or open a separate session.

The underlying capabilities remain:

Cowork was available in Claude Desktop for paid users, on the web and mobile apps for Pro, Max and Team customers, and through connected Enterprise accounts. Desktop access also let Claude work directly with local files and the user’s computer browser.

Claude Code is not part of the merger. Anthropic still positions it as a separate product for developers working with codebases from terminals and development environments.

The more visible part of the release is a set of work surfaces built into Claude:

Docs and Slides are not simply export buttons at the end of a chat. They remain editable, can be opened through a shared link and work on phones. Claude Docs is unrelated to Anthropic’s developer documentation site, which was renamed from Anthropic Docs to Claude Docs in September 2025. Claude Slides is also separate from Claude for PowerPoint, Anthropic’s existing Microsoft PowerPoint add-in, which received a major update in March 2026.

From an agent mode to the product itself

Anthropic introduced Cowork as a research preview on January 12, 2026. It began as a Claude Max feature in the macOS desktop app, applying the architecture of Claude Code to tasks outside software development.

The original pitch was simple: developers were already using Claude Code for travel research, slide creation and email cleanup, so Anthropic exposed the same delegation model through a graphical interface. Cowork became “Claude Code for everyone else.”

That framing helped make Cowork more than a feature launch. It became one of the early reference points in the 2026 debate later called “SaaSpocalypse”: whether general-purpose AI agents could take over work previously handled inside specialized SaaS products. By the end of February, VentureBeat reported pressure around ServiceNow, Salesforce, Snowflake, Intuit and Thomson Reuters as investors worried that more capable agents could reduce the value of individual software products.

Cowork was not the sole cause of the later software selloff. The broader narrative also included subsequent Anthropic releases, specialized plugins, Claude Code capabilities and the possibility that AI-driven workflows could span several enterprise systems—or replace parts of them.

Now Anthropic is making a more consequential product decision than the disappearance of a navigation toggle. It is taking the agent architecture that users had to deliberately enter and making it the default layer of Claude.

The weekly-report example shows the intended workflow. A user can ask Claude to:

Claude can ask follow-up questions, continue working without the user and return a document and presentation from the same conversation. The task can then be scheduled to run again the following Monday.

That is a coherent product direction: the chat is no longer just where the answer appears. It becomes the place where a request turns into a report, a presentation, a design and a recurring process.

The control problem is still there

Anthropic says simplification of the interface will not remove user control. Claude asks for permission before taking an action by default. Users can configure it to continue independently and return only when a more careful review is needed. The unified product retains the confirmation modes Cowork used for connectors and tools.

That matters because the scope of Claude’s work is expanding across several systems:

My view is that removing the Cowork boundary is the right interface decision, but it also makes the cost of a bad decision less visible. A separate mode told users that they were handing Claude a longer, more autonomous job. In the unified design, that escalation can happen inside an ordinary conversation.

The announcement is quieter about the operational details that will determine whether companies trust this model: how users inspect work performed while they were away, how organizations audit actions across connectors, and how reliably a document and its five-slide summary stay aligned after repeated scheduled runs. Anthropic explains what Claude can produce; it says less about how teams will verify the process.

That gap is important. If Claude becomes the front door to files, services, documents and recurring tasks, the product is no longer competing only with chat interfaces. It is asking to sit above the systems where work is actually recorded.

The Cowork name will gradually disappear. The more significant disappearance is the assumption that users should decide in advance whether they want a chat or an agent. Anthropic is betting that Claude should make that choice—and that the interface can stay simple even as the work underneath becomes harder to supervise.