Anthropic targets business workflows with Claude Duet

The artificial intelligence landscape is becoming increasingly crowded with new assistants and productivity tools launching almost weekly. Anthropic is now joining this competitive race by introducing a new application designed to capture user attention. The company has unveiled Claude Duet, an agent that positions itself against rivals such as Meta Muse, OpenAI Dots, Hark, Instinct, Grok Bot, and Fo. This move signals a strategic shift as the market evolves beyond simple chatbots toward systems that can actively work on behalf of their users.

Anthropic has clearly defined the target audience for this new tool. The company states that Claude Duet is built specifically for professionals and small businesses looking to automate and manage their workflows. There is a deliberate distinction being made between this new product and personal agents like Instinct. Anthropic notes that Duet is not optimized for handling personal errands or booking tickets. Furthermore, it cannot provide a live view of its browser usage as tasks are executed, a feature found in some other competing products.

Instead of managing personal life logistics, Duet focuses on generating daily briefings, creating data dashboards, and suggesting actions based on user information. The agent operates similarly to other tools in this category by completing tasks in the background using its own virtual machine. This allows users to continue with their other duties while Duet works. The system will return to the user only when it requires specific approval to proceed with a task.

This launch coincides with a broader industry trend where many of Anthropic's competitors are focusing on business use cases. For example, while Meta's Muse is primarily a personal assistant, the company also released a version tailored for small businesses with connections to various applications. OpenAI's Dots also has a professional orientation, featuring integrations with specific professional tools. The market is clearly shifting toward agents that can handle complex, multi-step business processes rather than just answering questions.

Anthropic claims that Duet proactively delivers recommendations by monitoring ongoing tasks. The application is built on the Claude Opus 5.5 model, which provides the underlying intelligence for its operations. To facilitate its work, Duet can connect to more than 6,000 connectors and plug-ins available in the Claude Marketplace. This extensive connectivity allows the agent to interact with a wide variety of external services and data sources without needing custom code for every integration.

One of the key features of Duet is its ability to make online purchases on behalf of users. The agent can utilize Stripe’s Link digital wallet or Shopify’s Shop Pay to complete transactions. However, the company emphasizes that sensitive steps, particularly those involving payments, will require explicit user approval. This safeguards the user's financial data and ensures they remain in control of their spending. Despite its purchasing capabilities, Duet will not trade securities or move money between different accounts, limiting its scope to specific transaction types.

Security is a primary concern for any agent that handles user data, and Anthropic has introduced a new classifier to address this. This tool checks each message to prevent sensitive information from leaking out during interactions. By default, the agent is programmed to turn down requests that ask for credit card numbers or government identification documents. This layer of protection aims to make the system more trustworthy for business users who are concerned about data privacy.

Claude Duet is currently available on iOS and Android devices, as well as on the web through Claude.ai. The service is accessible to Pro and Max customers who are part of the research preview program. Because it is still in preview mode, users must join a waitlist to gain access to the application. This staged rollout allows Anthropic to gather feedback and refine the agent before a wider public release, ensuring the tool meets the specific needs of its professional user base.