Claude Duet is a proactive AI agent app built on Claude Opus 5.5 that preemptively surfaces and completes work across a user’s connected apps. Designed for professionals and small-business owners, it sends morning briefs, monitors inboxes and tools, executes background workflows in a virtual machine/browser, and notifies the user only when something needs attention. It handles tasks from scheduling and drafting messages to building dashboards, financial forecasts, prototypes, and web apps, and can access over 6,000 connectors (Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, DocuSign, HubSpot, Canva, Shopify, etc.). Actions that affect accounts or finances require explicit approval, and some operations (trading securities, moving money) remain off-limits. Early testers report time saved, fewer missed errors, and practical uses across scheduling, accounting, customer support, and marketing.
Safety and privacy are central: Claude Duet uses Anthropic’s trustworthy-agent framework, never directly sees passwords or tokens, relies on an auto-mode classifier before taking actions, and incorporates defenses against prompt injection; it underwent red-team and external penetration testing and is covered by a bug-bounty program. A dedicated classifier blocks sharing of the strictest sensitive data (credit card numbers, government IDs) and only allows other sensitive info when explicitly authorized. The product is in a US research preview for Pro and Max plans, shares plan usage limits (background checks count toward those limits), has known early limitations (occasional misreads, weaker personal-errand support, no browser view), and is available via waitlist.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.