Anthropic has collapsed Claude chat, Cowork, Artifacts and Claude Design into a single window, and handed the choice between them to the model. Users no longer pick a tab or switch windows for a given kind of work; Claude decides where the request goes. The company's stated aim is one interface as the entry point to every part of the app. Pro and Max subscribers get the change over the coming weeks on the web, desktop and mobile, with the free and team plans to follow later.
The single window now contains four things: Claude chat, Cowork, Artifacts — Claude's interactive workspace — and Claude Design, the feature for building websites and prototypes that shipped in April. Until now each had its own place in the interface, and the user had to know which one a task belonged to before starting it.
Alongside the merge, Anthropic added dedicated handling for presentations and documents. Claude can create slides, edit an existing presentation, present it, and export the result as PDF or PowerPoint. Docs lets a user work on a document with Claude directly — building it out section by section, asking questions about it, leaving comments on finished parts. Generated documents and decks can be shared by link and then edited from a phone. A task can be started on a computer and its progress checked in the mobile app.
The routing is the actual product here, and it is a confession as much as a feature. Having to choose a tab before you know what you are doing was a real problem, and the honest way to describe this launch is that Anthropic accumulated four surfaces faster than it worked out how they fit together, and is now papering over the seam with a classifier. That is the right fix. It is also the kind of fix that only works when it is almost always right.
Which is the thing the announcement is quiet about. Nothing has been said about what happens when the router picks wrong — whether a user who wanted a document and got a chat reply, or wanted a quick answer and got an agent session, can send it back the other way, and how much of a task is already done by the time the mistake is visible. Removing the tabs removes the user's ability to be explicit. Every product that has replaced an explicit control with an inferred one has discovered that the 5% of cases where the inference fails generate most of the complaints, because in those cases the user now has no lever at all.
The document and slide work points somewhere more consequential. Exporting to PDF and PowerPoint, sharing by link, editing from a phone — none of that is about making the model better. It is about making the output land in the file formats and the workflow people already use to send work to each other, which is where the switching costs actually live. The same logic explains the cross-device handoff: start on the desktop, check it on the phone. The model is being wrapped in the ordinary machinery of office work.
This also follows an improvement to Cowork's memory layer, which carries user context between chats. Memory and routing solve the same problem from opposite ends: one removes the need to re-explain yourself, the other removes the need to declare what kind of work you are doing. Put together, they describe a product that intends to be addressed as a single thing rather than operated as a set of tools. ChatGPT, for what it is worth, still asks its users to manage that themselves across platforms, and would be better for a similar consolidation.
The tradeoff is now visible. Anthropic has bet that a model good enough to route your intent is more valuable than a menu that lets you state it. If the classifier is right often enough, nobody will remember the tabs. If it is not, the company has removed the only way to tell it what you wanted.