Anthropic launches Claude Duet to automate multi-step workflows

Anthropic has released Claude Duet, expanding its product line into persistent agent tooling designed to automate multi-step workflows.

The release represents a shift from turn-based chat interfaces toward persistent, task-oriented execution. Claude Duet operates alongside users as an interactive collaborator, managing multi-stage technical and administrative procedures across connected environments.

Under the hood, the system relies on structured tool orchestration and continuous context management. Rather than treating prompts as isolated interactions, Duet maintains a working memory across extended sessions. This architecture allows the agent to decompose high-level user instructions into subtasks, plan sequential operations, and invoke APIs or local utilities to complete work asynchronously.

The interface centres on a collaborative workspace. Users provide initial instructions and parameters in an input console, while Duet maps the operational graph and executes tasks in a dedicated runtime window. As the model navigates files, gathers external information, and prepares code or documents, it provides a real-time audit trail of each action taken.

Anthropic has integrated explicit boundary controls into the execution loop to mitigate the risks associated with autonomous model operation. Claude Duet employs a human-in-the-loop framework for sensitive actions. System modifications, critical file writes, and external API requests require manual user confirmation before the model proceeds. This friction aims to prevent unintended side effects caused by model drift or brittle tool integrations.

The tool is positioned directly against emerging developer agents and enterprise productivity platforms. While earlier autonomous agents struggled with context decay and infinite error loops, Anthropic is relying on improved reasoning capabilities and runtime constraints to deliver predictable outputs.

Organisations deploying the agent can integrate custom tool suites via standardised protocols, enabling Duet to interface with internal databases, version control platforms, and communication software. Demonstrations indicate the system can draft code changes, reconcile data across formats, and synthesise distributed research materials into structured briefs.

Whether enterprise administrators will deem these safeguard layers robust enough to grant persistent agents broader access to sensitive production systems remains an open question.