Anthropic launches AI agent Claude Duet, based on Claude Opus 5.5
Anthropic releases AI agent Claude Duet, benchmarked against Meta's Muse and OpenAI's Dots, launching as a research preview via standalone iOS and Android apps and the web version of Claude. The agent is based on Claude Opus 5.5, can call on Claude's memory, skills and connected apps, and comes with a virtual machine with a browser to complete tasks; before key operations such as payments and deletions, it first sends a message requesting user approval. Anthropic says it comes with new dedicated classifiers and safety system prompts to refuse to share information at the strictest tier, such as credit card numbers, while stating that this is still an early version that may misread conversations or misjudge urgency, and is not yet strong enough for personal matters such as booking tickets.
Why read this
SelectedAI score 76
Anthropic positions Claude Duet as an agent for professional users, and provides an approval mechanism and safety classifiers, allowing comparison of trade-offs against similar products.
Topics
Body
Anthropic just announced an AI agent competitor to Meta’s Muse and OpenAI’s Dots: Claude Duet. The new agent, designed for professionals and small business owners, is launching as a “research preview,” according to Anthropic , and it’s available as a standalone app on iOS and Android and through Claude on the web.
The AI agent is built around Claude Opus 5.5. Here’s Anthropic’s description of what it can do:
Claude Duet pulls in memory, skills, and connected apps from Claude, and it has a virtual machine with a browser that it can use for completing work. Before any big step, like sending a payment or deleting something, Claude Duet messages you for approval. Given that Duet may be taking action on sensitive information, it has a new “new dedicated classifier and safety system prompt” that are “designed to reject requests to share the strictest tier of information,” such as credit card numbers.
Anthropic says Claude Duet is currently an “early version” and that it “still might misread a thread or miscalibrate what’s urgent.” It’s also not “as strong” at personal errands like booking tickets.