One Claude to Do the Work: Anthropic Kills Cowork as a Separate App and Folds Docs, Slides, and Design Into Every Chat

On September 16, Anthropic quietly retired one of its own product names. Claude Cowork — the separate workspace it had built for “bigger work” like research reports, data analysis, and multi-step projects — is no longer a place you go. It is now simply part of Claude chat. Alongside the merge, the company launched two new artifact types, Claude Docs and Claude Slides, brought Claude Design into every conversation, and promised that what Claude produces can be edited, presented, and exported as PowerPoint or PDF without ever leaving the chat window.

The move is Anthropic’s clearest statement yet about where it thinks AI assistants are heading: not toward more apps and modes, but toward fewer. “You don’t have to choose where a task goes,” the announcement reads. “Claude does more of the work.”

What actually changed

Three things, concretely:

Cowork is absorbed into chat. Everything Cowork could do — long-running tasks, background work, file handling, connectors, skills — is now available from any regular conversation. Claude itself decides what a task needs, rather than the user picking the right surface in advance. Existing Cowork users lose nothing: chats, projects, artifacts, connectors, and skills all carry over in place.

Claude Docs and Claude Slides ship in beta. Ask for a document and Claude drafts it with you in a shared editing surface; ask for a presentation and it builds the deck. Slides can be edited directly element by element, presented straight from the Claude interface, or downloaded as a native PowerPoint file or PDF. Docs follow the same collaborative pattern — you write a line, Claude writes the next, and either of you can revise.

Claude Design works inside conversations. The visual design tool, previously its own destination, now operates inline, and anything created with Docs, Slides, or Design lives at a single shareable link that opens on a phone, where individual elements can still be selected, moved, or changed by asking.

All three artifact types are in beta on paid plans, with Enterprise administrators controlling when they activate for their organizations.

The workflow Anthropic is selling

The announcement walks through a scenario that reads like a summary of the entire agentic-AF pitch in miniature: a weekly report is due at noon. Before leaving, you ask Claude what moved in the pipeline last week, then add “write it the way we always do, flag anything that slipped, and put the highlights in five slides for the leadership meeting.” Claude asks if anything is unclear, then works after the laptop is closed. Progress is checkable from a phone. By the time you are at your desk, a doc is waiting with a teammate’s notes attached, and slides that already match the report are ready to open, adjust, and download. You fix a line yourself, leave a comment on a slide, and share it. Then you schedule it to repeat every Monday — and Claude starts the next one unprompted.

Two design decisions in that flow deserve attention. First, autonomy is opt-in and visible: by default Claude asks before taking actions, and users who prefer a more hands-off mode — “keep working and check in only when something needs a closer look” — must explicitly turn it on. Anthropic is keeping the “you keep the final say” framing even as it hands Claude the keys to long-running, scheduled work. Second, outputs are portable: PowerPoint and PDF export means the slides meet the formatting expectations of an enterprise that still runs on Microsoft Office, rather than trapping work inside a proprietary artifact format.

Why Anthropic merged them

The company’s own explanation is refreshingly blunt about the failure mode it was fixing. Cowork was built as a separate place for bigger work and Design for visual work — and people used both, but “the frustrating part was deciding where a task belonged.” Worse, “what they’d started in one also didn’t carry into the other.” Context did not survive the boundary between surfaces. The fix was not a better picker or a smarter menu but the removal of the boundary itself: Claude now infers whether a request is a quick question or a report due at noon, and routes its own capabilities accordingly, with the context, skills, and connectors from the existing conversation intact.

This is the same conclusion the rest of the industry is converging on. OpenAI’s ChatGPT has steadily absorbed document creation, code execution, and scheduled tasks; Google’s Gemini lives inside Docs, Slides, and Gmail natively. The Verge characterized Anthropic’s move as coming “for Gemini with its own take on Docs and Slides,” and the competitive read is correct — but the deeper story is that the standalone “AI workspace” category, which Cowork helped define, turned out to be a transitional form. When the assistant can decide the scope of a task, separate modes become navigation overhead.

Rollout and availability

The unified experience is rolling out to Pro and Max plans first, across the Claude app on web, desktop, and mobile, over the coming weeks — to both existing and new subscribers on those tiers, with nothing to enable. Team and Free plans follow “soon.” Enterprise administrators will get at least 30 days’ notice before anything changes for their organizations, and they choose when Docs, Slides, and Design turn on. Notably, Claude Code users are included in the artifacts push as well: Design, Slides, and Docs artifacts now work inside Claude Code, bringing editable outputs into terminal-based coding workflows.

The stakes

For Anthropic, this is both a simplification and a bet. The simplification is real — one interface, one mental model, fewer places to lose context. The bet is that Claude’s ability to classify a task’s scope automatically is good enough that users will trust it with work that runs in the background, on a schedule, without being asked. That is a higher bar than “pick the right mode from a dropdown,” because misclassification is now silent: if Claude treats a throwaway question as a project, or a deadline-driven report as a chat, the user finds out late.

The counterweight is the export story. PowerPoint, PDF, and shareable links mean that even when Claude misjudges a task, the output still lands in formats the non-AI world can consume. That pragmatic bridge — agentic autonomy on the inside, Office-compatible files on the outside — is the most telling detail in the announcement. Anthropic is not asking enterprises to change how they work; it is asking them to let Claude do more of it, one merged surface at a time.