Anthropic is collapsing Claude Chat, Cowork and its other services into a single product. The choice that used to belong to the user — which environment to open for a given piece of work — now belongs to the model: Claude reads the demands of the task and picks the mode itself. The rollout has started with Pro and Max subscribers, with Team and Free to follow. Enterprise administrators get at least 30 days' notice before the change reaches them.
What the single surface contains is largely what was already shipped, now reachable without switching context. Claude Docs and Claude Slides create, edit and export documents and presentations inside the chat. Claude Design is part of the conversation. Tasks that have been started keep running in the cloud after the laptop is shut.
Source: the-decoder.com
The problem being solved is one Anthropic made for itself. Users complained that the split between Claude Chat and Cowork was awkward, and the interfaces overlapped enough that it was often unclear which one a task belonged in. The same confusion exists next door, between ChatGPT and ChatGPT Work. Two products that look alike and do overlapping things are not a menu; they are a quiz the user did not sign up for.
Source: the-decoder.com
This reads as a correction rather than an expansion. Nothing here is a new capability — the shipped thing is the removal of a decision, which is an admission that the decision should not have existed. The rollout order says the same in a different register: Pro and Max first, Free last, and Enterprise handled with a 30-day notice period. You do not give administrators a month of warning about a feature. You give it to them about a change to the thing their organisation already depends on.
The more interesting question is what happens when the routing is wrong. The announcement describes Claude determining the requirements of a task and selecting the appropriate mode; it says nothing about what the user does when the model's read of the task does not match their own. A visible mode switch was, whatever else it cost in confusion, a guarantee. An invisible one is a judgement call made on your behalf, and judgement calls have a failure rate.
Anthropic has bet that model-side routing is now reliable enough to be the product's front door. That is a bet on the model, not on the interface — and it is the kind of bet that only pays once, because there is no going back to a menu people have stopped looking for.