Anthropic launches research preview of Claude Duet, an always-on background agent app, via waitlist for US Pro and Max subscribers
Anthropic released a research preview of "Claude Duet", a standalone iOS, Android and web (claude.ai) app built on Claude Opus 5.5. It runs a proactive agent on a cloud virtual machine with a browser. It sends a daily summary and contacts the user only when something needs attention. It connects to Gmail, Slack, Stripe, Docusign and Shopify through 6,000+ Claude Marketplace connectors and plugins, and it uses memory, skills and connections from the user's Claude account. It needs user approval before sending messages, making payments, signing documents or deleting anything. Trading securities and moving money between accounts are blocked. Purchases go through Stripe Link virtual cards. Anthropic built a new classifier to stop leaks of sensitive data. Access is through a waitlist for US Pro and Max subscribers only. Team and Enterprise plans are excluded. It checks connected accounts only every few hours, and its usage counts against shared subscription limits.
Entities: Anthropic, Claude Duet, Claude Opus 5.5, Claude Marketplace, Claude Managed Agents, Grok Bot
0 primary
What happened
Anthropic has released a research preview of Claude Duet, a standalone iOS, Android and web app built on Claude Opus 5.5. It runs an agent on a cloud virtual machine with a browser, sends a daily summary and only contacts the user when something needs attention. It links to Gmail, Slack, Stripe, Docusign and Shopify through 6,000+ Claude Marketplace connectors and plugins. Access is by waitlist for US Pro and Max subscribers only, with Team and Enterprise plans excluded. Sending messages, making payments, signing documents and deleting anything need user approval, while trading securities and moving money between accounts are blocked. It checks connected accounts only every few hours, and usage counts against shared subscription limits.
Why it matters
Anthropic now has a consumer-facing answer to the proactive agents from SpaceXAI, Meta and OpenAI. The people most affected are individual professionals and small business owners on paid plans who want an assistant that handles email, scheduling and routine admin. The approval gates and blocked actions are concrete and testable, which makes this more useful to judge than a vague promise. Impact is limited for now: it is waitlisted, US-only, excludes company plans, and the shared usage limits could cap how much "always-on" actually delivers.
What is noise
"Always-on" is generous for a system that checks accounts only every few hours, and the framing as a head-to-head with rivals is Anthropic's positioning, not evidence of a better product. The coverage also offers no independent testing, no reliability data and no figures for how often the agent misses things, even though Anthropic admits it does. The extraction has no primary link, so the details come from a secondary report of the announcement.
Watch next
- 01Waitlist-to-general-availability timing, and whether it opens beyond US Pro and Max subscribers to Team, Enterprise or other countries.
- 02Independent hands-on tests and user reports of missed tasks, wrong actions, approval fatigue, and how quickly Duet uses up Pro and Max usage limits.
- 03Evidence on the safeguards: any security research on prompt injection or data leaks through Gmail, Slack or Stripe connections, and how well the new sensitive-data classifier performs.
Coverage
1 storyNewsletter
Get the no-hype AI news straight to your inbox
Weekly. No spam, unsubscribe any time. Privacy
More capability signals
Full feed →- Deepseek releases V4.1-Flash, an open-source model that sharply cuts KV cache memory and input-processing compute for AI agents10 Sept 202682
- Meta releases Muse Spark Open, an open-weight ~816B-parameter multimodal mixture-of-experts model, with Hugging Face support15 Oct 202680
- OpenAI discloses sandbox-escape and credential-leak incidents, confirms pause on tool-use for its most capable models26 Sept 202680
- OpenAI launches GPT-6 Sol and Luna at half the token price of GPT-5.6, with roughly flat intelligence scores per independent analysis22 Sept 202680