Six weeks ago I put three environments side by side in a table: Claude Code, Cowork and Dispatch, with a column each for what I actually used them for. That table stopped being accurate on 16 September. Not because I changed my mind, but because two of those three columns are no longer columns.
Anthropic has merged Cowork and chat into one Claude. You no longer decide up front where a task goes. You ask for something, and the model works out whether that becomes an answer, a document, a presentation or a job that runs for an hour. Claude Docs and Claude Slides launched the same day, both in beta, and Claude Design now works inside any conversation.
What was announced
| Part | What changes | Where it applies |
|---|---|---|
| Cowork | No longer a separate place; everything it did works from any conversation | Pro and Max first, Team and Free later |
| Claude Docs | New, in beta: documents you write with Claude and your colleagues | Pro, Max, Team, Enterprise. Not on Free |
| Claude Slides | New, in beta: decks you present from Claude or export | Paid plans |
| Claude Design | No longer only a standalone surface; callable anywhere | Paid plans, Enterprise via an admin |
| Permissions | Auto or Manual, set per conversation. Manual is the default | Everywhere in the new experience |
The rollout happens in stages. Two accounts on the same plan can get it on different days, and there is nothing to switch on. If your message box still shows “Chat” and “Cowork” buttons, you do not have it yet. Enterprise admins get at least 30 days of notice before anything changes.
One sentence in the help documentation is worth reading before you get excited: “Once your account has the new experience, you can’t switch back to separate ‘Chat’ and ‘Cowork’ options.” This is a one way door rather than a setting you can put back.
The mechanism: routing became the model’s job
Anthropic is honest about the reason. Cowork was built as a separate place for bigger work and Design for visual work, and what people found frustrating was deciding where a task belonged. On top of that, work you started in one place did not travel to the other.
That is a real problem and the fix is sensible. The choice does not go away though, it relocates. Where I used to decide before typing which environment got the task, the same model that executes the task now decides. That is a different kind of decision than it looks like, because a router that is also the worker has no independent view on whether it routed correctly.
You feel it in two places. There is no web search toggle any more, because Claude searches when it judges that searching helps. And research now sits behind /deep-research or behind the plus button, so what used to be a visible mode is a command you have to know about.
The permission setting is the one place where you still hold the reins. Manual, the default, asks before each action. Auto lets Claude keep working with automated safety checks running before it acts. That setting applies to the whole conversation, which makes it a decision you make once rather than per task.
The doors that close
This part was not in the blog post. It sits in the help documentation under a heading called “Current limitations”. Four of them read like temporary gaps in a new release. The first one below does not.
- Dispatch is not available to new users. Anyone already using it can continue for now. In July I wrote about starting a Claude Code session from your phone; that path is now shut for anyone who was not already through it.
- Search does not cover older Cowork tasks. You can still find them by name in Recents, but not through search.
- Branching a conversation from an earlier point is not available.
- Add from GitHub does not work.
- Incognito chats open in the previous experience, where Claude cannot create files or run code.
On that last one, you use incognito precisely for the things you do not want kept, and that is exactly where you get the reduced version. It is the same dividing line I ran into with the shared memory: whoever keeps work separate for good reasons also keeps the benefit outside the door.
Docs and Slides are new, and it shows
Claude Docs is a rich text document stored in your Claude account, with tabs, tables and comments. You can mention @Claude inside it, and it edits in place and explains what it changed. Every change is attributed to whoever made it, person or model, and Claude only acts with the permissions of the person who asked. Export goes to Word, PDF, Markdown and Google Docs.
The limitations list is longer than the feature list, and one item on it is something I would want to know before putting a client document in there:
Important: Deleting a doc is permanent. There’s no trash, and the doc becomes unavailable to everyone you shared it with after you delete it.
There is also no version history, no comment-only access level, and charts do not refresh themselves even when the data comes from a connected app. On Team and Enterprise plans a doc cannot be shared outside your organisation. And activity inside a doc, meaning edits and comments, does not reach the Compliance API yet.
Claude Docs is also unavailable to organisations using CMEK, zero data retention or a HIPAA-ready configuration. On 1 September Anthropic announced Enterprise Frontier Safeguards as the answer that pairs ZDR privacy with misuse detection, and that rolls out later this autumn. In the meantime, the customers who chose ZDR for exactly those reasons are the customers who do not get the new writing surface.
The number that is missing
This announcement contains no usage figure at all. Not how many people used Cowork, not what share of tasks started in the wrong place, not how large that frustration was that the post refers to. The hard numbers that are there cover the rollout rather than usage: 30 days of notice for Enterprise, and four export formats for a doc.
The least flattering sentence sits in the section about usage: “While the new experience rolls out, usage may be measured slightly differently for accounts that have it and accounts that don’t.” It does not say in which direction, and there is no percentage attached. Long agentic tasks that search, run code or create files consume more than a short question, according to that same page, and it is not quantified beyond that. Anyone who hits their limit harder after the rollout has no figure to compare against.
The caveat
The default shifts from choosing to being chosen for. A mode you switched on yourself was a moment of thinking about the size of what you were asking. That moment is gone. For a quick question that is pure gain, but the cases where you specifically do not want half an hour of work set in motion are exactly the cases where nobody asks you to confirm any more. Manual is the default and that is the right call, yet it is a per conversation setting, so you flip it once and forget that you did.
Responsibility for the result stays with you while you see less of it. Claude Docs attributes every change to a person or the model, which is carefully done. But without version history you cannot return to an earlier state, and without a trash can one wrong click is final for everyone you shared with. A writing surface with no undo across sessions asks for a discipline the tool itself does not support.
What drains away is the ability to go back. You cannot switch back to separate Chat and Cowork. Dispatch closes to new users. Incognito gets the older, reduced version. Each of those is defensible on its own, but together they mean the way you work today will not be available to return to in a few weeks, and you find that out after it has already happened.
What I am doing
I am leaving my August comparison table up, with a correction on top. The table is wrong now; the reasoning underneath it holds. Claude Code remains the only environment that sees your whole project instead of one task, and this merge does not change that. The Code tab is where it was.
What I am changing is my own automation. In my working directory I maintain a sync that pushes the same skill to three targets: chat, Cowork and Claude Code. That becomes two. It sounds like tidying up, and it is, but it is the kind of tidying you only get round to after something has quietly been written to the wrong place.
I will use Docs and Slides for what they are: drafts and internal pieces. Nothing whose loss would hurt while there is no version history. Real work stays in files in the repository, where git does keep a history and a deletion is not a final deletion.
And I am setting the permission control to Manual on purpose, even though it costs clicks. Auto is probably fine, but I want to keep making that distinction rather than clicking it away once.
Frequently asked questions
Can I go back to the old Chat and Cowork?
No. The help documentation is explicit: once your account has the new experience, you cannot switch back to separate Chat and Cowork options. As long as you still see both buttons in your message box, you do not have it yet and can keep working the way you are used to.
Will I lose my existing Cowork tasks?
No. Tasks, projects, connectors, skills, artifacts and files carry over, and old Cowork tasks open as they did before. Some things sit elsewhere now. Cowork’s Global instructions live under Instructions for Claude in Settings > General, so check that they say what you want. Search skips older Cowork tasks, which you can still find by name in Recents.
When do I get it?
The rollout is phased and starts with Pro and Max on web, desktop and mobile. Team and Free follow later, and Enterprise admins get at least 30 days of notice. Two accounts on the same plan can receive it at different moments, and there is nothing you can enable to move up the queue.
Is Claude Docs suitable for confidential material?
Not for now, and that is a product limitation rather than a judgement call. Claude Docs is unavailable to organisations using CMEK, zero data retention or a HIPAA-ready configuration. On top of that, activity inside a doc does not reach the Compliance API yet, and deletion is permanent with no trash can.
Sources
- Anthropic, “Claude Cowork and chat are now one Claude”, claude.com, 16 September 2026, accessed 17 September 2026.
- Anthropic Help Center, “Claude Cowork and chat are one Claude”, support.claude.com, accessed 17 September 2026.
- Anthropic Help Center, “Get started with Claude Docs”, support.claude.com, accessed 17 September 2026.
- Anthropic, “Developing Enterprise Frontier Safeguards with our customers”, anthropic.com, 1 September 2026, accessed 17 September 2026.
- Anthropic Help Center, Release notes, support.claude.com, accessed 17 September 2026.
Checked on 17 September 2026. Every fact and quotation above comes from Anthropic itself rather than from the press coverage that appeared the same day. Two things I could not verify. The new experience is not on my own account yet, so everything about how it behaves comes from the documentation and not from use. And several summaries report that Design, Slides and Docs can point at files and RFCs in your repository from inside Claude Code; the documentation only confirms that you can ask Claude Code for a doc that turns your session into a spec, runbook or readout, with a link handed to you in the terminal. The wider claim is therefore not in this piece. At the time of writing, the Help Center release notes did not mention the merge at all; the most recent entry there is dated 15 September.