Particle.news

Anthropic Releases Claude Duet, an Always‑On Proactive AI Agent

Running inside its own virtual machine and browser to act in the background, Claude Duet connects to more than 6,000 services, requires explicit user approval for sensitive actions, and uses virtual cards for payments.

Overview

  • Anthropic announced a research-preview launch of Claude Duet as a standalone app and web agent that can run continuously in its own virtual machine and browser to perform background tasks.
  • Users can give text or voice instructions for scheduling, email triage, invoice tracking, forecasting, dashboard building, and simple web development, and the agent notifies users or requests review when tasks complete.
  • Claude Duet taps over 6,000 connectors from the Claude Marketplace to act across services such as Gmail, Slack, QuickBooks, Stripe, Shopify, and DocuSign and can complete online payments using virtual cards through Link or Shop Pay.
  • Anthropic enforces safety limits by blocking requests that would leak high‑sensitivity data, forbidding direct access to passwords or login tokens, and requiring user approval for messages, payments, signatures, or deletions.
  • The release is a limited preview with shared usage quotas and known early‑version flaws, such as misreading conversations, misjudging urgency, limited personal-task handling, and the inability to show its browser actions, and it positions Anthropic alongside Meta’s Muse and OpenAI’s Dots in the push toward proactive, always‑on assistants.