Thought Leadership
The agent checks connected apps, gets in touch when necessary, and continues working in the background. The research preview is launching for Pro and Max users in the US.
Anthropic has introduced Claude Duet, an app in which Claude operates as a proactive agent. The agent is designed to notify users of important events, complete tasks, continue working in the background, and ask for confirmation before taking actions on their behalf. According to Anthropic, the target audience is professionals and small business owners. A research preview is initially available via a waitlist for subscribers of the Pro and Max tiers in the US. Anthropic has not stated when Claude Duet will launch in other countries.
Claude Duet is available as a standalone app for iOS and Android as well as via claude.ai and is based on the Claude Opus 5.5 model. The app adopts memory, skills, and connected apps from the existing Claude account. During the preview, it is included in the Pro and Max plans and draws from their usage limits.
Morning Summary and Background Work
Every morning, Claude Duet sends a brief overview of the most important events across connected apps. Throughout the day, the agent checks connected accounts every few hours and reaches out if something requires attention. Users can assign tasks via text or voice, ranging from scheduling to dashboards, financial forecasts, or web apps. To achieve this, the agent operates in a virtual machine with its own browser, meaning laptops or smartphones do not need to remain switched on. More than 6,000 connectors and plugins can be integrated via the Claude Marketplace, including for Gmail, Slack, QuickBooks, Stripe, HubSpot, Docusign, and Shopify.
Approval Before Taking Action
Before Claude Duet sends a message, triggers a payment, signs a document, or deletes anything, the agent displays the planned action and waits for approval. Certain actions are excluded even with user consent, such as trading securities and transfers between accounts. For purchases, the agent uses Stripe’s digital wallet Link with virtual cards or Shopify’s Shop Pay. According to Anthropic, the agent does not directly see passwords or login tokens.
A dedicated classifier is designed to prevent confidential information from leaking. It rejects requests to share credit card numbers or identification data. Claude Duet only shares health or financial data if users explicitly instruct it to do so. To protect against prompt injections, Anthropic relies on methods from its research into browser control. Claude Duet was tested by internal red teams and an external company, and is part of the public bug bounty program.
“This is an early version of Claude Duet, and it might still misread a thread or misjudge what is urgent.”
Anthropic
Personal errands such as booking tickets are areas where the agent is currently less proficient. Background checks and larger tasks also consume more of the usage limit than standard conversations.
Several providers have recently unveiled continuously active AI agents, including Meta with Muse, OpenAI with dots, and SpaceX Super Intelligence with Grok Bot.
(Editorial Team)
