Somewhere in your company's internal Claude guide, there's probably a sentence like this one:
"Select Cowork mode and turn on web search before you ask for a report."
It was right when somebody wrote it. The two buttons it points at just don't exist any more. On 16 September, Anthropic merged Claude Cowork and ordinary Claude chat into one app, and in the new version Claude picks the tools itself. The mode picker is gone. The web search toggle is gone.
This lands on people unevenly, and the dividing line isn't job title. It's habit. If anyone at your company has written down how to use Claude, in an onboarding page, a Loom video, a Notion doc or just a pinned message in a Teams channel, you're holding a document that describes a screen on its way out. If nobody has written anything down, you're spared this job and you have a different problem.
"You can't switch back". Anthropic's own sentence.
Anthropic says it plainly in the help centre article for the update: "Once your account has the new experience, you can't switch back to separate 'Chat' and 'Cowork' options."
Let me clear up two misreadings first, because both are easy to reach for.
Cowork isn't closing. This isn't a shutdown, and there's nothing you need to rescue before a deadline. Your chats, tasks, projects, connectors, skills, artifacts and files carry over, and that includes any Claude integration you have already connected. Anthropic's own wording is that your chats, tasks, projects and settings come with you when your account moves. Cowork tasks show up in Recents alongside the chats.
What doesn't carry over is the version of the app your material was written about.
On 16 September, Claude stopped asking which mode you wanted
The news itself is a merge. Claude Docs and Claude Slides arrive in beta, and Claude Design moves inside the conversation instead of sitting in its own workspace. Ask for a document and you and Claude write it in the conversation. Ask for a presentation and Claude drafts the slides, and you can edit directly, present straight from Claude, or download it as PowerPoint or PDF.
Underneath the headlines sits the change that hits written material. Anthropic has removed a set of places where a person used to make a choice:
- The choice between Chat and Cowork is gone. Claude picks the tools for the task.
- Web search has no toggle. Claude searches the web when it helps.
- Research sits somewhere new, behind the /deep-research command or the plus button.
- Global instructions have moved to Settings, General, Instructions for Claude.
And the thing somebody will ask about: Anthropic says nothing about data location, retention or training in this update. What they describe is the interface, and who makes the choice inside the app.
Permissions are still a setting you can set. By default Claude asks before it takes an action, and you can turn that off if you'd rather it kept working and checked in only when something needs a closer look.
The sentences in your guide that no longer hold
Here's the part you can go and check. The help centre names five things that don't work the way they did, and they aren't five of a kind. That difference decides what you do about each one:
- Add from GitHub isn't supported. That's a removal.
- Branching a conversation from an earlier point isn't available. Also a removal, and it's the one most non-developers will actually feel.
- Dispatch isn't available to new users. That's a restriction on new people, not a removal for those who already have it.
- Incognito chats still work, but they open in the previous experience, so Claude can't create files or run code in them. That's a behaviour change.
- Search doesn't include your older Cowork tasks. That's a gap in search, not a loss of content.
Read the last one twice. The content carries over, but the route into it doesn't. For a company that has been working in Cowork for months, the archive isn't what disappears. What disappears is the ability to find something in it by searching. Name the important work and move it into projects now, while you can still find it.
When a vendor tells you your data carries over, ask separately whether the index carries over. Those are two different promises.
For a few weeks your team runs two different versions of Claude
The rollout happens in stages. Pro and Max first, on web, desktop and mobile, with Team and Free to follow. Anthropic writes that even accounts on the same plan will see the changes at different times.
Read that as an operations notice rather than a technical footnote. Two colleagues at the same company, on the same plan, will spend a few weeks holding different versions of Claude. One of them has Docs and Slides. The other still sees a mode picker. Your guide is right for one and wrong for the other, and whoever answers internal questions gets both versions in their inbox in the same week.
You can't schedule around the window, because you don't get the date. Nothing states when an account moves, and there's no documented way to opt in, delay or reverse. Enterprise admins hear from Anthropic at least 30 days ahead. That's the only warning anyone has attached a number to. No other plan has been given one.
Rewrite the instruction so it names the outcome
This is the one thing I'd like you to leave with, and it takes an afternoon.
A guide that names a button has an expiry date. A guide that names an outcome keeps. Buttons move when a vendor merges two products. What the user is trying to achieve doesn't move.
Take the sentences you've actually written and turn them around:
| As it stands | Rewritten |
|---|---|
| "Select Cowork mode and turn on web search." | "Ask Claude to research it online and write up what it finds." |
| "Click the Research button in the toolbar." | "Ask for a deeper investigation when the topic needs more than one source." |
| "Put your standing instructions under Cowork settings." | "Put your standing instructions wherever Claude keeps account-wide settings." |
| "Branch the conversation if you want to try another angle." | "Start a new conversation from the version you want to build on." |
That last one isn't a rewrite at all. Branching is gone, so it's a different method, and that's exactly why it pays to walk the list: two of the five are real removals, and there you need a new route rather than new wording.
Two more things while you're in the document. Delete the screenshots, or put a date on them so the next reader knows what they're looking at. And write at the top which version of Claude the guide describes, because during the weeks when half of you have moved, that line decides whether the page helps or confuses.
It's also why I rarely spend a Claude workshop on buttons. Buttons are the part that goes stale first.
How I read this as an AI consultant
I don't want to leave this as a bill, because it isn't one.
The mode picker asked users to make a routing decision they were badly placed to make. Should this be a chat or a Cowork task? That depends on whether the job needs files, code, several steps or just an answer, and you usually find that out afterwards. Letting Claude pick the tools removes a choice that was guessed more often than made.
It isn't a change of direction either. Claude Design moved into the conversation back in June, and the direction has been visible ever since. 16 September is the end of a three-month move rather than a reversal.
What I'd do first with a client right now isn't teaching people the new interface. It's going through the written material they already have and taking the buttons out of it. It's the dullest part of any AI implementation, and it's also the cheapest to redo. The work gets done once and survives the next merge.
As an AI consultant I set that up with the team rather than handing over a slide deck: we take your own material, your own sentences, and rewrite them together. That's what a Claude workshop looks like with me.
This is the second time in a few months that Cowork has moved under people's feet, by the way. Last time Cowork went to web and mobile so tasks could run without your own computer. Same direction, bigger step.
One sentence you can put into the document today that will still hold after the next update:
"Ask Claude to research it online and write up what it finds."
FAQ
Frequently asked questions
No. Anthropic says so plainly: once your account has the new experience, you can't pick “Chat” and “Cowork” separately again. There's no setting for it and no way back.
Anthropic doesn't say. The rollout runs in stages, Pro and Max first, then Team and Free, and even accounts on the same plan move at different times. Enterprise admins get at least 30 days' notice. No other plan gets a number.
No. Chats, tasks, projects, connectors, skills, artifacts and files carry over, and tasks show up in Recents alongside the chats. The one gap is search, which doesn't cover older Cowork tasks.
Two new things in beta. Ask for a document and you and Claude write it in the conversation. Ask for a presentation and Claude drafts the slides, which you can edit, present straight from Claude, or download as PowerPoint or PDF.
There's no documented way to opt in, delay or reverse it. Enterprise admins get notice, not a button.
Anthropic says nothing about data location, retention or training in this update. What they describe is the interface, and who picks the tools inside the app. Anything that changes about data is documented somewhere else.
Go through it and take the buttons out. An instruction that names an outcome survives the update. One that names a mode or a toggle doesn't. Date the screenshots, and say at the top which version of Claude the page describes.
