Claude merges chat and Cowork, with no return to the old mode

On September 16, 2026, Anthropic announced the merger of Claude chat and Cowork into a single conversation. In this new experience, the assistant chooses the tools needed for the request, whether it involves a short answer or more complex work.

Cowork previously had a separate entry point for substantial tasks. Anthropic explains that its users were unsure where to start and that work begun in one space did not automatically carry over to the other. A discussion can now serve as the starting point for a report, then for its presentation.

The transition applies to every account, gradually

The migration is taking place over several weeks, starting with Pro and Max on the web, desktop and mobile. Team and Free are expected to follow, with no specific date announced. Enterprise administrators will receive at least 30 days’ notice before any change to their organization.

The documentation for the unified experience states that accounts on the same plan may migrate at different times. An account that still displays the Chat and Cowork options in the input area retains both entry points. Once it has migrated, returning to those separate options becomes impossible. Conversations, tasks, projects and settings carry over with the transition, with no activation required.

The betas have their own availability

The migration schedule must be distinguished from access to the new tools. Docs and Slides are launching in beta on paid plans. The Claude Docs documentation details availability on Pro, Max, Team and Enterprise, with the feature enabled by default on the first three plans. On Enterprise, Docs is disabled until an owner enables it.

Docs lets users draft a document, edit it directly or ask Claude to make changes. Collaborators with editing access can work on the same document. Slides lets users prepare and edit a presentation, present it from Claude and export it to PowerPoint or PDF.

Docs availability is subject to restrictions for organizations using customer-managed encryption keys (CMEK), zero data retention (ZDR) or a HIPAA-ready configuration. On mobile, users can request and view a document; templates, editing and sharing settings are managed on the web or desktop.

Choose access permissions and the authorization mode

Manual is the default setting. Claude requests authorization before taking actions, and the user decides whether to approve them. With Auto, steps continue without a new request each time, following automated safety checks.

The setting applies to the entire conversation and can be changed. Users can also interrupt or redirect the work. For a first trial, we recommend granting access only to the files and applications needed for the task. In Docs, Claude acts with the permissions of the person requesting the change; read-only access does not give it the right to edit.

Monitor usage and know where the task runs

All work is counted toward subscription limits. A lengthy task that conducts research, runs code or creates files generally uses more than a quick question. We recommend checking Settings > Usage and starting with a clearly defined deliverable, specifying the files to use and the expected format.

Tasks running in the cloud can continue when the computer is closed. Those using local files or applications require Claude Desktop to remain open. Whether users can step away therefore depends on the resources needed for the requested work.

Authorization to act and content approval are separate

Auto’s safety checks guarantee neither the absence of errors nor the quality of the result. Authorizing an action and approving the content of a document are two different decisions. Charts and diagrams in Docs also need to be checked for freshness, because updating them requires a new request to Claude, even when they use a connected application.

For organizations, our assessment is that this autonomy could require clarifying who grants access and who checks deliverables before distribution. This question of oversight echoes, from another angle, our article Anthropic promises external oversight at the heart of its AI systems. The lab’s external evaluators provide no guarantee regarding documents produced for a user.

Anthropic promises fewer manual steps between conversation, document and presentation. No quantified productivity gain has been established at this stage. The next evidence to watch for would be an independent trial comparing Manual and Auto on equivalent tasks, measuring deliverable quality and quota usage, as well as errors and the human interventions required.