Claude Cowork is being folded into ordinary Claude conversations, and with it Anthropic has removed the single decision its customers complained about most: which tab a piece of work belongs in. The announcement went up on 16 September 2026 under the blunt title Claude Cowork and chat are now one Claude, and the company’s own framing of the problem is unusually candid. “People used both,” Anthropic wrote, “and told us the frustrating part was deciding where a task belonged.”
The merge arrives with three collaborative surfaces in beta — Claude Docs, Claude Slides and Claude Design — and it quietly reduces Claude from three front doors to two. What survives alongside the merged experience is Claude Code, the terminal and IDE product, which Anthropic explicitly kept out of scope. That same week, however, Claude Code shipped three releases of its own carrying a batch of new opt-in capabilities, which is why the two stories keep being told together.
This article separates them. It covers what the Claude Cowork merge actually changes, which plans get it and when, what carries over and what does not, the beta status of Docs, Slides and Design, and then the specific features Claude Code gained in versions 2.1.271 through 2.1.274 — with the counts taken from the published changelog rather than from a press summary.
Table of contents
- What the Claude Cowork Merge Actually Changes
- From Three Modes to Two: Claude Cowork, Chat and Claude Code
- Claude Docs, Claude Slides and Design Arrive in Beta
- What Claude Cowork Users Keep — and What Is Missing
- Autonomy Settings in the Merged Claude Cowork Experience
- Who Gets the Claude Cowork Merge, and When
- Claude Code Is the Mode That Did Not Merge
- The Beta Features Claude Code Gained in the Same Week
- What the Claude Cowork Merge Means for Teams
- Limits, Caveats and Open Questions
- Frequently Asked Questions About the Claude Cowork Merge
- References
What the Claude Cowork Merge Actually Changes
The change is a removal, not an addition. The toggle goes away, and the routing decision moves from the user to the model.
One conversation, no mode picker
Before this week, Claude Cowork was a separate destination for larger pieces of work, while chat handled quick questions. Anthropic’s position now is that the separation was artificial: “You don’t have to choose where a task goes. Claude does more of the work.” A request typed into the ordinary message box is assessed by the model, which decides whether it warrants a short answer or a longer agentic run.
The context problem the split created
The deeper complaint was not the extra click. Work begun in one product did not carry into the other, so a user who started in chat and realised the job was bigger had to restate it. Anthropic’s summary of the fix is that “Claude can now figure out what a task needs, so what Cowork and Design can do is available from any conversation, with the context, skills and connectors you already have.”
Work that outlives the session
The capability Claude Cowork was built around survives intact. Longer jobs continue running server-side after the user disconnects: “More involved tasks keep running in the cloud even if you close your laptop or leave the page. Come back when the task is done, and the result is waiting in the conversation.” That is the part of the old product that justified its existence, and it is now reachable from any thread.
| Capability | Before the merge | After the merge |
|---|---|---|
| Choosing where work goes | User picks chat or Claude Cowork | Claude routes the request itself |
| Long-running agentic tasks | Claude Cowork only | Any conversation |
| Visual design work | Claude Design, separate surface | Inside conversations, standalone still available |
| Documents and decks | Artifacts and the PowerPoint add-in | Claude Docs and Claude Slides, in beta |
| Front doors to Claude | Three | Two |
| Developer terminal product | Claude Code, separate | Claude Code, still separate |
From Three Modes to Two: Claude Cowork, Chat and Claude Code
The arithmetic of the announcement matters more than the interface polish, because it tells you how Anthropic now segments its own users.
The three that existed
Until this week a customer faced chat for conversation, Claude Cowork for project work, and Claude Code for engineering against a codebase. Design sat alongside as a fourth visual surface for some users. Each had its own mental model, and in practice people learned one and ignored the rest.
The two that remain
The merge collapses the first two into a single product, leaving Claude and Claude Code. That is a cleaner split, and it is drawn along a real boundary: whether the work happens in a browser or against a repository. We covered an earlier stage of this consolidation when Anthropic added the chat and Cowork toggle to the sidebar in July, and again when the Cowork browser layer landed in Claude Desktop in August.
Why the Claude Cowork name disappears
Anthropic launched Cowork in January 2026 as a place for bigger work, and spent August syncing memory between it and chat so context could cross the boundary. Reading those two moves together, the merge looks less like a reversal and more like the end of a staged migration: first make the two halves share state, then delete the wall. The Claude Cowork brand is retired; the behaviour is not.
Claude Docs, Claude Slides and Design Arrive in Beta
Three collaborative surfaces ship with the merge, and all three are labelled beta on paid plans. This is the part of the announcement with the clearest competitive intent.
Claude Docs
Claude Docs produces a living document written jointly by the user, Claude and colleagues. Anthropic describes Claude as drafting, asking clarifying questions up front, and leaving comments explaining its choices. Documents are private by default, with Viewer and Editor access levels, and export to Word, PDF, Markdown and Google Docs. On Team and Enterprise plans they can be shared with named people, named groups or the whole organisation; on Pro and Max they are shared by link.
Claude Slides
Claude Slides drafts a deck from an ordinary request. Individual slides can be edited, the deck can be presented directly from Claude, and it downloads as PowerPoint or PDF. An existing document can be converted into a presentation in place. This is a distinct product from the older Claude for PowerPoint add-in, which is worth stating plainly because the two are easy to conflate.
Claude Design inside conversations
Claude Design is not new, but its location is. It now runs inside a conversation rather than as a separate destination, while the standalone version remains available for people already using it. The Next Web reports that Design is built on Canva’s design engine, integrated in April 2026; that detail comes from one outlet rather than the announcement itself, so treat it as reported rather than confirmed.
What beta means here in practice
Beta is not decoration. Claude Docs is available on Pro, Max, Team and Enterprise plans and is not available on Free. It is on by default on Pro, Max and Team, and off by default on Enterprise until an owner turns it on. Individuals can disable it under Settings, then Capabilities. It counts against the plan’s usage limits like any other work, with larger requests consuming more.
| Surface | What it produces | Export | Status |
|---|---|---|---|
| Claude Docs | Collaborative living document | Word, PDF, Markdown, Google Docs | Beta, paid plans only |
| Claude Slides | Editable, presentable deck | PowerPoint, PDF | Beta, paid plans |
| Claude Design | Visual assets and prototypes | Not specified in the announcement | Beta in chat, standalone continues |
| Free plan access | None of the three | — | Docs explicitly excluded |
What Claude Cowork Users Keep — and What Is Missing
Migration promises are the part of any consolidation worth reading twice, and Anthropic’s is more specific than most.
Everything carries over
The Help Centre wording is unambiguous: “Your existing Cowork tasks, projects, connectors, skills, artifacts, and files carry over, and those Cowork tasks open as they did before, so you can continue them.” A half-finished Claude Cowork job is not stranded by the merge — it opens in the new surface and resumes.
The gaps Anthropic names itself
The same document lists what does not work yet, and the list is short but real. Adding context from GitHub is not supported. Conversation branching is unavailable. Search does not cover older Claude Cowork tasks at first. Dispatch is limited to people already using it, and incognito chats open in the legacy experience. None of these are catastrophic; all of them are the kind of thing that ruins a Monday if you depend on one.
The migration is one-way
Once an account has moved, there is no toggle back to separate chat and Claude Cowork modes. Rollout is staged within a tier as well as across tiers, so two colleagues on identical plans can see different interfaces for a period. For a team that documents its own internal processes with screenshots, that is worth planning around.
Autonomy Settings in the Merged Claude Cowork Experience
Handing routing to the model raises an obvious question about control, and the answer is a permission setting that predates the merge.
Manual is the default
Two settings govern how far Claude proceeds unattended. Manual, the default, requires approval before Claude takes an action, and the user allows each one. Auto lets Claude work continuously with automated safety checks, checking in when a decision genuinely needs review. The setting can be changed mid-conversation.
Why the default matters more after the merge
When Claude Cowork was a separate destination, entering it was itself a signal of intent — you went there because you wanted a long job done. With that signal gone, the permission setting is the only remaining brake. Organisations that were relaxed about autonomy inside a dedicated workspace should re-examine the setting now that the same behaviour can start from any thread, which is the same governance question that arises whenever intelligent automation moves from a pilot into everyday tooling.
Connectors widen the blast radius
The merged experience reaches the connectors an account already has — Google Workspace, Microsoft 365, Slack and others — from ordinary conversations rather than from a dedicated workspace. Anthropic’s own enterprise groundwork here, including managed authentication for Claude MCP connectors, is what makes that defensible at organisational scale.
Who Gets the Claude Cowork Merge, and When
Nothing about the rollout is instant, and the sequencing tells you who Anthropic considers its most forgiving users.
Pro and Max first
The merged experience reaches Claude Pro and Max subscribers first, across web, desktop and mobile, “over the coming weeks”. It is automatic: there is nothing to install or enable, and the Cowork toggle simply disappears as the change reaches an account.
Team and Free next
Team and Free plans follow afterwards, with no date given. Free-plan users should note that arriving at the merged interface does not bring the beta document tools with it, since Claude Docs is excluded from Free entirely.
Enterprise last, with 30 days’ notice
Enterprise organisations receive at least 30 days’ notice to administrators before anything changes, and admins control enablement. On Enterprise, Docs is off until an owner turns it on, and can be granted to specific groups through custom roles rather than the whole organisation. It is also unavailable to organisations running CMEK, zero data retention or HIPAA configurations — a limitation that will decide the matter for a number of regulated buyers.
Claude Code Is the Mode That Did Not Merge
This is the point most coverage of the Claude Cowork story gets slightly wrong, so it is worth stating precisely.
Explicitly out of scope
Claude Code is not part of the consolidation. Anthropic’s position is that it is intended for developers working from terminals and IDEs against codebases, and it remains a separate product with its own release train. The merge reduced three consumer-facing modes to two; Claude Code is the second of those two, not a casualty of the first.
A different release rhythm
Claude Code ships continuously and publishes a changelog, whereas the Claude Cowork merge was a single announcement with a staged rollout. The two things happened in the same week and are frequently reported as one event, but they come from different teams on different cadences. We looked at the desktop side of that boundary recently in our piece on the sessions hub for Claude Code management.
Where the boundary now sits
The practical dividing line after the merge is the artefact. If the output is a document, a deck, a design or a research answer, it belongs in the merged Claude. If the output is a commit against a repository, it belongs in Claude Code. That is a far easier rule to teach a new starter than the old three-way split.
The Beta Features Claude Code Gained in the Same Week
Claude Code published three substantive releases in mid-September 2026, and their contents are public. The counts below come from the project’s own changelog, not from a summary.
Counting what actually shipped
Version 2.1.274 carries 108 entries, of which 13 are new additions. Version 2.1.271 carries 96 entries with 10 additions. Version 2.1.273 carries 64 entries with 4 additions. Two further releases in the window, 2.1.270 and 2.1.272, are single-line fix releases. The ratio is the story: these are maintenance-heavy releases with a small, deliberate set of new capabilities each.
The opt-in capabilities worth knowing
Several additions are gated behind a flag or a setting, which is what makes them beta-shaped in practice. Fast mode arrived in Claude Code Remote sessions on cloud and self-hosted runners. Gateway hint headers must be switched on with an environment variable. Sessions started with remote control can now be forked, with the fork running as a background session on the user’s own machine.
Controls aimed at administrators
The other cluster is governance. Sandboxed commands gained per-command allowed domains, so the hosts a single command needs are reviewed with it and opened for that command alone. Custom and plugin subagents can be told to run without user, project and local instruction files, while managed policy files still load. Plugin installs can accept one exact command by hash instead of a blanket approval.
Reading the ratio honestly
Roughly nine in ten changelog lines are fixes, improvements or behavioural changes. A product that publishes that ratio openly is telling you where its engineering effort actually goes, and it is a useful corrective to the impression left by a headline about beta features.
| Release | Entries | New additions | Headline capability |
|---|---|---|---|
| 2.1.274 | 108 | 13 | Critical-memory warning; MCP startup wait control |
| 2.1.273 | 64 | 4 | Forking a remote-control session; gateway hint headers |
| 2.1.272 | 1 | 0 | Fixes only |
| 2.1.271 | 96 | 10 | Fast mode in Remote sessions; per-command allowed domains |
| 2.1.270 | 1 | 0 | Fixes only |
What the Claude Cowork Merge Means for Teams
The interface change is small. The operational consequences are not, and they land on whoever owns tooling policy.
Retraining is mostly unlearning
Teams that built internal guidance around “use Claude Cowork for X, chat for Y” now have documentation that describes a product that no longer exists. The replacement rule is shorter, which is a genuine win, but someone has to rewrite the wiki page and the onboarding deck before a new starter follows the old one into a dead end.
Document sprawl is the new risk
Claude Docs and Claude Slides create artefacts that live inside Claude, share by link, and export to four formats. That is convenient and it is also a new place for a half-finished strategy deck to sit outside the sanctioned document store. Decide where the canonical copy lives before the tool decides for you.
Check the autonomy setting before the rollout lands
Because the merge reaches accounts silently, the sensible sequence is to confirm the permission setting first and audit connectors second. Anthropic’s own segmentation of autonomy levels is a reasonable model for how organisations deploying autonomous AI agents should think about the same trade-off in their own stack.
Regulated environments need a separate answer
If an organisation runs CMEK, zero data retention or HIPAA configurations, Claude Docs is unavailable, and the answer to “can we use the new document tools” is simply no for now. That is better than an ambiguous maybe, and it should be written into the internal policy rather than discovered by a frustrated employee.
Limits, Caveats and Open Questions
Several things are genuinely unresolved, and it is worth being explicit about which parts of this story are confirmed.
No numbers were published
Anthropic disclosed no user counts, no growth figures, no revenue and no pricing changes alongside the Claude Cowork merge. Any commentary attaching adoption numbers to this announcement is inferring them. The pricing page is unchanged in substance for the plans named.
“Over the coming weeks” is not a date
The rollout window is deliberately vague for every tier except Enterprise, where the 30-day notice is the only firm commitment in the announcement. Teams planning a communication of their own should not promise a date Anthropic has not given.
Routing quality is the real variable
The entire premise rests on Claude correctly judging whether a request needs a sentence or an hour of agentic work. Anthropic has published no benchmark for that routing decision, and the model reading intent from a sentence of ordinary language is an exercise in natural language processing, not a solved classification problem. Early users will find the edges before any evaluation does.
Design’s engine is single-sourced
The claim that Claude Design runs on Canva’s engine comes from one publication’s reporting rather than from Anthropic’s announcement. It is plausible and consistent with the April 2026 integration, but it is not confirmed in the primary source, and we have flagged it as such above.
Frequently Asked Questions About the Claude Cowork Merge
Is Claude Cowork being shut down?
The separate product is going away, but the capability is not. Claude Cowork’s long-running agentic tasks, projects, connectors, skills, artifacts and files carry over into ordinary conversations, and existing tasks open and continue as before.
Do I have to do anything to get the merge?
No. The rollout is automatic for Pro and Max plans across web, desktop and mobile, and the Cowork toggle disappears when it reaches your account. Enterprise administrators are notified at least 30 days ahead.
Can I switch back to the old separate modes?
No. Once an account has been migrated there is no way to return to separate chat and Claude Cowork experiences.
Are Claude Docs and Claude Slides free?
No. Both are in beta on paid plans. Claude Docs is available on Pro, Max, Team and Enterprise and is not available on the Free plan, and usage counts against the plan’s normal limits.
Did Claude Code get merged into Claude?
No. Claude Code stayed separate and is now one of the two remaining modes. It did ship new opt-in features in versions 2.1.271, 2.1.273 and 2.1.274 during the same week, including fast mode in Remote sessions and forking a remote-control session, but those are separate releases from the Claude Cowork announcement.
What does not work yet in the merged experience?
Anthropic lists adding context from GitHub, conversation branching, search across older Claude Cowork tasks, and Dispatch for new users. Incognito chats still open in the legacy experience.
References
Claude Cowork and chat are now one Claude
Claude Cowork and chat are one Claude, Claude Help Center
Get started with Claude Cowork
Anthropic merges Claude chat and Cowork in one interface
Anthropic folds Cowork into Claude and launches Docs and Slides
Anthropic merging Claude Cowork with chat
Anthropic folds Cowork into a single Claude experience across plans
Claude Cowork and chat are now one Claude, Simon Willison
Anthropic consolidates specialized apps into one unified Claude platform
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.