Anthropic Merges Claude Chat with Cowork
Anthropic announced the merger of Claude Cowork and dialogue into a single Claude: users can ask questions or submit reports at any time, and the model continues to work even after closing the computer; if unclear, users can ask questions, with final decision-making power remaining with them.
The product also launches Claude Docs and Claude Slides, with Claude Design available for use within conversations. Users can collaboratively write documents, draft slides, edit directly, present in Claude, or download as PowerPoint and PDF. All three are in testing under paid plans, with enterprise administrators deciding when to enable them.
The reason for the merger is that users are tired of determining which entry point to use for tasks, and the context from one end does not carry over to the other. From now on, what Cowork can do can be called from any conversation, with tasks, projects, connectors, skills, products, and documents all migrated together. Conversations can run on Anthropic's cloud, not tied to local machines.
Initially rolling out to Pro and Max users on web, desktop, and mobile without the need for switches; teams and free users will follow, with enterprises notified at least 30 days in advance. The product form consolidates into two main paths from "chat / Cowork / Design" and adds Claude Code.
In market mechanics, this is a subscription retention battle driven by product line consolidation: reducing friction in mode selection and keeping long task computing power within the same session billing. Beneficiaries are paid Pro/Max users who stack workflows in Claude and Anthropic's cloud computing; those under pressure are competitors relying on independent Agent workspaces for differentiation. Funding and usage concentrate on "one box to finish" subscriptions rather than more buttons.
Source: Public Information
ABAB AI Insight
Cowork was originally intended to extend agents from coding to files, calendars, emails, and browsers as a second entry point, with the official claim that 90% of usage was not software development. Its retraction after a year indicates that the entry tax is higher than the capability tax: users are paying for judgment costs. At the same time, Anthropic integrated Claude Tag into organizational identities on Slack, retained Claude Code at the terminal, and pushed non-code agents back into chat on the consumer side—each of the three interfaces consolidating rather than proliferating brands.
The capital path is to replace local resident sessions with cloud conversations: billing continues even when the computer is closed, and documents and slides become products within the package, locking Office-style outputs into the same account. This aligns with the logic of removing system prompts and turning off Plan Mode—strengthening the model by eliminating patch interfaces designed for weaker models. The computing budget shifts from "opening a Cowork session" to "any conversation could be elevated to a task."
Analogous to ChatGPT integrating Advanced Data Analysis back into the main dialogue, Gemini merging the Workspace sidebar into chat, and Microsoft embedding Copilot from an independent application into Word: agent brands do not survive beyond a product cycle, leaving the default input box. The industry is in a phase of controlling interfaces and reducing modes.
Structural judgment belongs to technological substitution. The mechanism is that once the model can judge task scale independently, human mode selection shifts from assistance to friction; whoever eliminates choice first turns switching costs into competitors' customer acquisition taxes. Interface merging does not mean fewer functions, but consolidates the measurement rights of long tasks onto the same bill.
ABAB News · Cognitive Laws
With many entry points, users first pay taxes for classification.
Once the model can divert, mode switches should retire.
Billing rights leaving the local area only count when the computer is still running.