Anthropic opened a research preview of Claude Duet, a new app where Claude works as a proactive agent: it comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf. CTO Rahul Patil announced it with "Just duet!", calling it "the first place I go to get work done." It targets professionals and small business owners with more on their plates than they can track.
Availability
Duet is a standalone app on iOS and Android and also on claude.ai. In the preview it is included in Pro and Max plans and uses the same usage limits as the rest of Claude, with access through a waitlist. Anthropic designed it around Claude Opus 5.5, built to decide what needs the user, what can wait and when to check first.
What it does
- Morning brief and watching: each morning it sends a short brief on the most important items across connected apps, then checks those accounts every few hours and messages only when something needs attention. Checking nonstop would use up plan limits much faster, and Anthropic is still tuning when it should speak up. Users can say what they want to hear about and how often.
- Tasks of any size: from scheduling meetings to building dashboards with dozens of tools, financial or sales forecasts, prototypes and web apps. It has a virtual machine with its own browser, so it does not need a laptop or phone on, and notifies the user when a task is done or needs review.
- One thread, your context: it carries over memory, skills and connected apps from a Claude account, and takes typed or voice messages in one ongoing thread.
- Connectors: over 6,000 connectors and plugins in the Claude Marketplace, with examples across Gmail and Slack follow-ups, QuickBooks and Stripe invoice reminders, Gusto payroll, LinkedIn job posts, Docusign agreements, HubSpot follow-ups, Canva campaigns and Shopify store setup, with a richer visual editor for storefronts coming. Purchases go through Link, Stripe's digital wallet, which asks the user to approve the spend and pays with a virtual card, or through Shopify's Shop Pay. Sign-in with saved 1Password logins is coming, with permission.
Safety and discretion
Duet reads, drafts and prepares work on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it will do and waits for approval. Trading securities and moving money between accounts are off-limits even with approval, and any connected app can be disconnected.
- Credentials: handled with the design used for Claude Managed Agents, so it never directly sees passwords or sign-in tokens.
- Actions: approved by Anthropic's auto mode classifier, the safeguard used in Claude Code, and the trustworthy-agents framework also guides Claude Code and Claude in Chrome.
- Injection defenses: drawn from Anthropic's browser-use research. Red teams and an external firm penetration tested Duet, and it is covered by the public bug bounty.
- Sensitive information: a new dedicated classifier and safety system prompt check each message. They reject requests to share the strictest tier, including credit card numbers and government ID, while health or personal-finance details go out only when the user has explicitly told Duet to send them. Anthropic says it will share more on Duet's safety design in the coming days.
Patil says that giving an AI the keys to Slack, mail and calendar "is a little scary," and that he trusts it because it checks before anything important, checks for private information before a message goes out and pays with a one-time card.
Early users and context
Boris Cherny, who leads Claude Code, talks through his mental to-do list with Duet on walks. Chelsea Larsson, who leads model and product behavior design, hands it her to-do list before leaving to be with her kids and gets an evening report. Karan Ahuja, who leads the Duet team, uses it to schedule user interviews, prototype and ship code. Small business owners had early access.
It lands alongside other always-on agents this blog has covered: Microsoft's Autopilot, a persistent personal agent on Copilot+ PCs, and the security design of Meta's Muse personal agent.