←News detail

MacRumors Apple IntelligenceUpdated Oct 25 00:40

Anthropic launches proactive AI agent Claude Duet

AI summary

Anthropic has launched Claude Duet, a proactive AI agent that can send reminders and complete tasks before the user asks. Duet is based on Opus 5.5 and sends a briefing of important matters every morning, after which it checks connected apps every few hours and issues reminders when needed; it can run in a cloud virtual machine with its own browser, and obtains user approval before sending messages, making payments, signing documents, or deleting files. Duet reuses the memory, skills, and connected apps of the Claude account, with interactions concentrated in a single continuous thread; it does not see account passwords, and is equipped with dedicated classifiers and safety system prompts that check messages and prevent the leakage of sensitive information.

Why read this

AI score 73

The original text explains Duet's proactive reminder mechanism, execution in cloud virtual machines, and the requirement for user approval before sensitive operations, which can be used to assess the permission boundaries of proactive AI agents.

Topics

Tags

Body

Anthropic today introduced Claude Duet, its answer to Meta's Muse agent and OpenAI's Dots agent. Like Dots and Muse, it is a proactive agent that "reaches out before you ask."

Claude Duet connects to your apps and suggests things that need your attention. It completes tasks and can work in the cloud without local access to a computer. As with Muse and Dots, Duet has a virtual machine with its own browser to complete tasks.

Anthropic says it built Duet for professionals and small business owners with too much to track, as those are the people most likely to find it helpful.

Duet is built around Opus 5.5. It sends a short brief each morning on important items from connected apps, then checks apps every few hours and sends an alert when something needs attention. It can complete simple administrative tasks or more complex tasks like building web apps. While the agent works on its own, it will get user approval before it takes actions like sending a message, making a payment, signing a document, or deleting a file.

Introducing Claude DuetClaudeWatch on YouTube It incorporates memory, skills, and connected apps from a Claude account, so there isn't much setup involved. Interactions happen in one ongoing thread instead of across different conversations.

Claude Duet does not see passwords or logins when given access to accounts, and it has a dedicated classifier and safety system prompt that checks each message and guards against sensitive information leaks.

Source

MacRumors Apple Intelligence