Anthropic has introduced Claude Duet, a proactive artificial intelligence agent designed to monitor connected applications, prioritize tasks, and execute work in the background. According to Inc42, the tool operates independently of user prompts, sending daily briefings and updates when decisions are required. The launch places Anthropic in direct competition with OpenAI’s Dots and Meta’s Muse, as major technology companies expand their offerings into agents that perform tasks on behalf of users.
The research preview is currently available via an exclusive queue for American users holding Pro or Max plans. Inc42 reported that the agent is powered by Claude Opus 5.5 and targets professionals and small business owners. Users can access the tool via standalone mobile applications or through Claude’s website. It integrates with over six thousand utilities from the Claude Marketplace, including Gmail, Slack, QuickBooks, and Shopify. This connectivity allows the agent to identify unpaid invoices, draft payment reminders, and manage online stores.
A key feature of Claude Duet is its ability to run on a remote virtual environment equipped with a web browser. This setup enables the software to persist in its operations even after users terminate the local application or power down their hardware. Inc42 noted that the system polls linked services at intermittent intervals instead of running a continuous stream, a design choice intended to conserve subscription allowances. Complex assignments consume more usage limits than simple background checks, and the tool shares these limits with other Claude products.
To manage autonomy, Anthropic implemented safeguards that require user approval for sensitive actions such as sending messages, making payments, or signing documents. The agent uses a virtual card via Link, Stripe’s digital wallet, to avoid sharing actual card numbers, and plans to integrate with 1Password for saved logins. Inc42 explained that the system includes automated assessments to prevent prompt injection and controls that keep passwords out of the agent’s direct view. However, the company acknowledged that the early version may misjudge urgency or miss information arriving between periodic checks.
The launch adds to the growing scrutiny of always-on AI agents that operate across business systems with limited human intervention. Inc42 highlighted that as these tools gain access to more organizational data, their permissions and accountability are becoming focal points for security experts. The initial release excludes Team and Enterprise plan holders, and Anthropic stated that input from initial testers will guide subsequent iterations of the platform.