Anthropic’s 6,000-Connector Bet: Why Claude Duet Is Not Your Personal Errand Runner

6,000. That is the sheer volume of connectors and plug-ins available to Claude Duet, the new agentic application launching this week from Anthropic. The figure underscores the company’s strategy: rather than building a general-purpose companion, Anthropic is leveraging its marketplace to create a dense, professional-grade automation layer. The app is currently in research preview, accessible only via waitlist for Pro and Max subscribers on iOS, Android, and the web.

The central claim from Anthropic is that Duet is designed specifically for professionals and small businesses, not casual users. The company draws a sharp line between its product and the wave of personal assistants like Instinct. Duet is not intended to book tickets or handle errands. Furthermore, it lacks the visual transparency of some competitors, meaning users cannot watch the agent navigate a browser in real-time. Instead, the focus is on operational efficiency: building daily briefings, generating dashboards, and suggesting actions based on proprietary user data.

Under the hood, Duet runs on the Claude Opus 5.5 model. It operates by executing tasks in the background within its own virtual machine. This allows the user to continue their day while the agent works, returning only when a specific approval is required. This "background worker" model aligns with a broader industry shift away from conversational assistants toward agents that act on a user's behalf. Rivals like Meta and OpenAI are making similar moves, with Meta’s Muse offering a small-business edition and OpenAI’s Dots integrating deeply with professional tools.

However, the automation capabilities come with significant caveats regarding scope. While Duet can make online purchases using Stripe’s Link or Shopify’s Shop Pay, it is strictly limited in its financial autonomy. It will not trade securities or move money between accounts. For sensitive steps, such as payments, the agent is programmed to pause and request explicit user approval. This suggests a cautious approach to trust, acknowledging that autonomous financial decisions remain a friction point for many users.

Security is another area where Anthropic has introduced new mechanisms. A specialized classifier now scans every message to prevent the leakage of sensitive information. By default, the agent is configured to reject requests that ask for credit card numbers or government identification. This proactive filtering is a response to the inherent risks of giving an AI access to personal and professional data streams, yet it raises questions about how well such classifiers handle edge cases in complex business workflows.

The launch positions Anthropic directly against a crowded field including Meta Muse, OpenAI Dots, Hark, Instinct, Grok Bot, and Fo. While the market is saturated with new assistants launching weekly, Duet’s value proposition hinges on its depth of integration rather than its personality. The success of this bet will likely depend on whether professionals find the 6,000 available connectors sufficient to replace manual workflows, or if the lack of personal errand-handling capabilities feels like a significant limitation. For now, the app remains behind a waitlist, suggesting Anthropic is still calibrating the experience before a wider rollout.