Claude Duet messages users first, keeps working in the cloud when their devices are off, and asks permission before it acts, as Anthropic joins Meta, OpenAI and others in the race for always-on AI agents.
Anthropic on Wednesday, October 21, opened a research preview of Claude Duet, a proactive AI agent built for professionals and small business owners that surfaces what needs attention, completes assigned tasks and keeps working in the background, CNBC reported.
The launch positions Anthropic against Meta’s Muse and OpenAI’s Dots in the fast-growing market for personal and workplace AI agents.
What Claude Duet does
Claude Duet sends users a short brief each morning drawn from their connected apps, then checks those apps every few hours and messages only when something needs action, MacRumors reported. All interactions stay in one ongoing thread rather than separate chats.
Users can hand it tasks ranging from routine administrative work to more involved projects. Examples cited in coverage include investigating unpaid invoices, drafting customer follow-ups, building financial forecasts and creating web apps.
The agent runs on Anthropic’s Claude Opus 5.5 model and operates on its own cloud-based virtual machine with a browser, so work can continue when the user’s phone or computer is closed or offline. It carries over memory, skills and connected apps from an existing Claude account, keeping setup short.
Claude Duet’s safety guardrails
Before sending a message, making a payment, signing a document or deleting anything, Claude Duet shows the planned action and waits for approval. Some actions, including trading securities and moving money between financial accounts, are blocked even with approval, The Decoder reported.
The agent never sees users’ passwords or login tokens directly. Purchases go through Stripe Link with a virtual card, so a user’s real card number is not shared. A new classifier screens messages to stop leaks of highly sensitive data such as payment card numbers and government ID details, alongside defenses against prompt injection, a type of attack that hides malicious instructions in web pages or messages.
Anthropic acknowledges the early version of Claude Duet has limits. It can misread a message thread, misjudge how urgent something is and miss items between its periodic checks. It is also weaker at personal errands, since the product is aimed at work.
Who can use Claude Duet
During the research preview, access is limited to users in the United States on Anthropic’s Pro plan ($20 a month) or Max plan (from $100 a month), through a waitlist. There is no separate charge, but Duet draws on the same usage limits as the underlying plan, and its background checks count against them, VentureBeat reported. Team and Enterprise plans are not included yet.
Claude Duet is available as a standalone app for iOS and Android and through claude.ai. It connects to more than 6,000 connectors and plugins through the Claude Marketplace, including Gmail, Slack, QuickBooks, Stripe, Gusto, Docusign, HubSpot, Canva and Shopify.
A crowded race for AI agents
Claude Duet arrives as rival agents gain traction. Meta’s Muse, launched September 8, has passed 10.4 million downloads, according to an estimate by BNP Paribas analyst Nick Jones based on third-party data, Let’s Data Science reported. Meta has not published its own figure.
OpenAI unveiled Dots, an always-on agent that users can reach through ChatGPT, Slack or Microsoft Teams, on September 29, alongside a new $500 paid tier, Bloomberg reported. Grok Bot, another always-on agent, opened a beta in August.
Whether Claude Duet gains real adoption will depend on how well its judgment, reliability and safety features hold up in daily use, and on how quickly Anthropic opens access beyond its U.S. Pro and Max waitlist. The launch adds another major lab to the contest over who can build agents that professionals will trust with their inboxes, calendars and business tools.
More AI coverage from Vincypowa News: Three Tech Chiefs Reportedly Talked Trump Out of an Industry-Funded AI Watchdog.
Disclosure: Vincypowa News uses AI tools, including Anthropic’s Claude, in its production workflow. Anthropic had no role in this story.
Leave a Reply