OpenAI Gave You 3 Modes. Claude Just Deleted the Switch. | Edition 336

On Wednesday, Anthropic deleted a button. In the Claude app, the little toggle that made you declare — before you typed a single word — whether this was a Chat task or a Cowork task is going away. In its place: nothing. You type. Claude works out whether your request is a one-line answer or a four-hour job.

That sounds cosmetic. It is not. It is the opposite bet from the one OpenAI made ten weeks ago, and it quietly leaves you with exactly one decision — the only setting in the new Claude that can cost you real money, or real damage, if you get it wrong.

What actually changed

On September 16, Anthropic announced that Claude Cowork — the separate place where Claude took on long, multi-step jobs — is merging into ordinary chat. In the merged experience there is no mode to choose: you describe what you want, and Claude decides which tool the job needs, whether that is a quick answer or research, a report, a spreadsheet, or a presentation handed back as finished files.

The company gave a refreshingly unglamorous reason: people used both products, and the frustrating part was deciding where a task belonged — work started in one did not carry into the other.

Three things shipped alongside the merge. Claude Docs makes living documents you write together with Claude and your team. Claude Slides drafts presentations you can edit, present from Claude, or download as PowerPoint or PDF. Claude Design, previously its own thing, now works inside a normal conversation. All three are in beta on paid plans, and Enterprise admins decide when to switch them on.

The rollout starts with Pro and Max plans on web, desktop, and mobile over the coming weeks, with nothing for you to enable. Team and Free follow later. Enterprise admins are promised at least 30 days notice before anything changes for their organization.

OpenAI kept the switch. Anthropic threw it away.

This is the part worth sitting with, because both labs just solved the same problem and reached opposite answers.

On July 9, OpenAI folded its standalone Codex app into the ChatGPT desktop app. The result was one shell containing three distinct modes — Chat, Work, and Codex — with a switch at the top to move between them. Fewer apps, same question: which mode is this?

Anthropic removed the question. There is no picker in the new Claude, because the routing decision moved from you to the model.

For anyone who is not a developer, that difference matters more than it sounds. Picking a mode means guessing, before you start, how hard your own request is going to be — which is precisely the thing you do not know yet. Most people guess low, ask a big question in a small box, and get a smaller answer than the tool could have given them. Deleting the switch removes a whole category of user error. It also removes a whole category of user control, which is why the next section exists.

Three things to know before it reaches your account

1. There is no undo. Once your account has the new experience, you cannot switch back to separate Chat and Cowork options. The rollout is staged even within a plan, so you and a colleague paying the same amount may see different interfaces for a while. If your message box still shows Chat and Cowork, you have not moved yet.

2. The cloud does the work — until it needs your computer. More involved tasks keep running in the cloud even if you close your laptop or leave the page, and scheduled tasks run with your machine off. But anything touching your local files, the built-in browser, or computer use still needs the Claude Desktop app open. That is the single most common way a hand-over quietly fails: you close the lid on a task that needed your hard drive.

3. Auto mode costs more than Manual. The permission setting in the message box has two positions. Manual is the default and Claude asks before each action. Auto lets Claude keep working without stopping, running an automated safety check before each action instead. Anthropic states plainly that auto mode consumes more of your usage limit than the other modes, because that checking is work too. Manual is slower and cheaper. Auto is faster and burns your cap quicker — and for anything with real consequences, like messages sent as you or files that matter, Anthropic advises staying close regardless of setting.

What moved, and what is still missing

Moved: Cowork Global instructions are now part of Instructions for Claude under Settings > General, and they apply to every conversation — worth opening once to confirm they still say what you meant. Your old Cowork tasks now sit in Recents alongside your chats. There is no web search toggle any more; Claude searches when it helps. Research is now /deep-research, or the + button under the chat input.

Still missing, by Anthropic own list: Add from GitHub is not supported. Branching a conversation from an earlier point is unavailable. Incognito chats still work but open in the previous experience, so Claude cannot create files or run code in them. Search does not cover older Cowork tasks, though you can still find them by name in Recents. And Dispatch is closed to new users, with existing users grandfathered for now.

None of those is fatal. Each one is the kind of thing that ruins a Monday if your workflow happens to sit on top of it.

The skill this turns into

When the mode picker disappears, the quality of what you get back depends almost entirely on how you brief the thing. Anthropic own guidance is to write to Claude the way you would brief a colleague, and to include three items: the outcome you want to end up with, the format it should arrive in, and the inputs Claude will need — the files, links, or connected apps to work from.

The company own worked example is a weekly report due at noon: ask what moved in the pipeline last week, say to write it the usual way, flag anything that slipped, and put the highlights in five slides for the leadership meeting. You leave. The doc and the deck are waiting when you get back, both out of the same conversation, so the slides already match the report. Then you schedule it for every Monday and stop asking.

That is the actual product change. Not a new model, not a smarter answer — a shorter distance between a request and a finished file.

What to do this week

Three small things, in order. Open Settings > General and read your Instructions for Claude, because your old Cowork global instructions landed there and now apply to every conversation, including the quick ones. Leave the permission setting on Manual until you have watched Claude do something consequential and agreed with every step. Then pick the one deliverable you rebuild by hand every week and hand over the whole thing, outcome and format and inputs, instead of asking for the first paragraph of it.

The switch is gone either way. The only thing left to decide is how much you let Claude do before it checks in — and that one is still yours.