Anthropic folds Cowork into Claude and launches Docs and Slides: the race for the AI office speeds up

By TechRepublic · September 17, 2026.

Anthropic announced on Wednesday that it is merging Claude Cowork —its more autonomous mode, designed for long, multi-step tasks— with the standard Claude chat. From now on, users describe what they need and Claude itself decides which capabilities to activate, without forcing them to choose in advance between "conversation mode" and "work mode". The company is also adding two new tools, in beta for paying users: Claude Docs, for creating and editing documents within the same conversation, and Claude Slides, which generates editable, presentable presentations that can be downloaded as PowerPoint or PDF.

Both tools are designed for collaborative work: documents can be exported to Google Docs or Microsoft Word, and projects are shared via access and editing links from desktop or mobile. Anthropic illustrates the workflow with a telling example: asking Claude to research and draft a weekly report in an already established format, flag the missing points and, without leaving that same conversation, turn the conclusions into a presentation that inherits the report's context. The rollout begins with Pro and Max subscribers on web, desktop and mobile over the coming weeks; Team and Free plans will follow, and Enterprise administrators will receive at least 30 days' notice before the change. Anthropic says Cowork users will lose nothing: chats, projects, artifacts, connectors and skills will remain available as they are.

As industry context, the move is not a product whim: it is yet another sign that the fight over AI is no longer waged solely on model benchmarks, but over who controls the surface where people produce real work. Anthropic is now shifting its bet to the product layer where Microsoft 365 and Google Workspace live. The article itself recalls that OpenAI made an equivalent move recently by merging ChatGPT with its coding tool Codex: the two companies applying the most pressure at the model frontier have reached the same conclusion — that winning on the model is not enough if the end user still has to copy and paste the result into another application.

Our reading is that this accelerates, rather than initiates, a shift we have already been flagging in our sector coverage of employment: the more capable an assistant is of completing the entire cycle —research, draft, format, present— without intermediate human intervention, the more exposed the administrative and back-office work that today lives precisely on that shuttling between applications becomes. This is not an abstract threat: it is the concrete task of "take this report and turn it into a PowerPoint for the boss" that Anthropic has just automated out of the box. In the short term, it is bad news for anyone building their professional value on repetitive tasks of that kind; in the medium term, the relevant question for teams stops being "what does Claude do" and becomes what judgment, business context or human relationship the person still contributes once the draft is already done.

The article also points to the uncomfortable flip side of this convenience: the more tasks that run through a single conversation —uploaded files, connectors to other services, sensitive company information— the more an organization's security depends on the permissions each employee enables. Anthropic says Claude asks for confirmation before acting by default, and that users can choose to let it work more autonomously and only flag matters that require attention. That checkbox, seemingly minor, is the kind of governance decision IT teams will have to weigh carefully. That the rollout to Enterprise accounts comes with 30 days' advance notice is not merely commercial courtesy: it acknowledges that IT departments need that margin to decide which connectors to allow and with what level of autonomy, before the convenience of "everything in one conversation" turns into a risk surface that nobody audits in time.