Anthropic on October 21 opened a research preview of Claude8.3 out of 10 Duet, a separate app in which Claude acts as a proactive agent instead of waiting for a prompt. According to Anthropic, Duet surfaces what needs a user’s attention, completes tasks it is given, keeps working when the user steps away, and checks in before acting on their behalf. CNBC framed the launch as a challenge to OpenAI’s Dots and Meta’s Muse.
What Duet does
Anthropic’s announcement page describes a morning brief of priorities drawn from connected apps, plus a single ongoing thread that accepts typed or voice messages. Duet runs on its own virtual machine with its own browser, so a laptop or phone does not need to stay on, and it sends a notification when a task finishes or needs review. It checks connected accounts every few hours, and users can set what they are told about and how often.
The page lists scheduling, dashboards, financial and sales forecasts, prototyping and web-app building among its capabilities. It carries over the memory, skills and connected apps from a user’s Claude account and can use connectors from the Claude Marketplace, including Gmail, Slack, QuickBooks, Stripe, HubSpot and Shopify. For purchases it can use Link, Stripe’s digital wallet, with a virtual card, after the user confirms the item and Link approves the spend; it can also check out with Shop Pay.
Availability and controls
Duet is built on Claude Opus 5.5 and is available on iOS, Android and claude.ai. Access is by waitlist, starting with US Pro and Max subscribers, and it is included in those plans. It shares the plan’s usage limits, and background checks count toward them.
Anthropic says Duet reads, drafts and prepares work on its own, but shows the exact action and waits for approval before it sends messages, makes payments, signs documents or deletes anything. Trading securities and moving money between accounts are off-limits even with approval. Credentials are handled as in Claude Managed Agents, so Duet does not directly see passwords or sign-in tokens, and requests to share the strictest data tier, such as card numbers or government IDs, are rejected. Anthropic says it uses an action classifier, defenses against prompt injection, red teams, an external penetration test and a public bug bounty, and that more safety design details will follow.
Limitations
Anthropic acknowledges that Duet may misread a thread or misjudge urgency, that personal errands such as booking tickets are weaker than work tasks, and that it cannot yet show users its browser. Saved 1Password sign-ins and a visual Shopify storefront editor are listed as coming soon.