代理應用
Anthropic Merges Claude Chat and Cowork, Enabling Automatic Selection of Agent Tools and Execution Environments Within a Single Conversation
Claude can now answer questions or launch long-running agentic tasks within the same conversation, while bringing documents, slides, and design tools into a shared context. With mode selection removed, tool routing and permission prompts become the new security boundary, while access to local files and computer use still depend on the desktop app.
Anthropic announced on September 16 that it is merging Claude Chat and Cowork. Users will no longer choose between chat and agent modes in advance; instead, Claude will determine whether to search the web, read files, execute code, or generate editable artifacts based on the request. The new version is initially rolling out in phases to Pro and Max users across the web, desktop, and mobile apps. Once an account has been migrated, it cannot switch back to the previous two separate entry points.
In practice, this change replaces “mode switching” with runtime routing. Short questions can still be answered directly, while longer tasks can continue running in the cloud. Even after closing the page or shutting down their computer, users can check progress, answer questions, or redirect the task from their phone. Multiple tasks can run in parallel, and scheduled tasks can also be triggered in the cloud. However, workflows that require local files, desktop applications, or computer use still require Claude Desktop to remain open.
The new Claude Docs and Claude Slides, along with the existing Design tool, are also being integrated into regular conversations. From the same context, the agent can create documents, spreadsheets with formulas, and PowerPoint presentations, which users can then edit, comment on, share, or export directly. Existing projects, memories, connectors, Skills, and Cowork artifacts will be migrated as well. Actions require confirmation by default, although users can change this setting so they are interrupted only when the system determines that more careful review is needed.
For enterprise and agent-platform engineers, the significance goes beyond a simplified interface. When the system independently decides to escalate from text generation to browsing, code execution, or actions through external connectors, permissions must be evaluated for each action according to the tool, data source, and execution location; authorization can no longer be granted only once upon entering Cowork. Anthropic has not yet disclosed the router architecture, false-activation rate, additional token costs, or safety evaluations. The features also remain in a phased beta, and enterprise administrators will receive at least 30 days’ notice before the change.