Crmgo Blog

Anthropic launched Claude Duet — a proactive AI agent that works for you in the background

Anthropic launched Claude Duet — a proactive AI agent that works for you in the background

Anthropic has opened early access to Claude Duet: an agent on Claude Opus 5.5 that sends a daily briefing, monitors email and calendar, executes tasks on its own cloud virtual machine, and asks for permission before payments or sending emails.

Anthropic launched Claude Duet — a proactive AI agent that works for you in the background

Anthropic has opened research preview access to Claude Duet — a new app in which Claude works as a proactive agent. It lets you know when it needs your attention, carries out tasks in the background, and asks for permission before important actions.

What Claude Duet is

According to Anthropic, Claude Duet was created for professionals and small business owners who have more to do than they can keep track of. Unlike a regular chat with Claude, where the user opens the conversation themselves, Duet "comes to you first": every morning it sends a short brief on the most important items in connected apps, such as email and calendar, and during the day it writes only when something genuinely needs a response.

It works even when you're offline

The agent has its own virtual machine in the cloud with a separate browser, so it doesn't need the user's laptop or phone to be turned on. All you have to do is describe a task in a short message — by text or voice — and Duet will determine which apps and steps are needed. For sites without ready-made connectors, it uses its own cloud browser. The results of large tasks are returned as files: analytical memos, financial models, presentations, or dashboards.

Important decisions remain with the human

Duet can independently read, prepare drafts, and prepare document templates. But before sending a message, making a payment, signing a document, or deleting something, it shows exactly what it is about to do and waits for confirmation. Some actions are prohibited even with the user's consent — in particular, trading securities and transferring money between accounts.

What it runs on and who can try it

The agent is built on the Claude Opus 5.5 model; no other model can be selected. It carries over memory, skills, and connected apps from the Claude account. Access is being opened in waves via a waitlist — first for Pro and Max subscribers in the US. Duet is available as a separate app for iOS and Android and on claude.ai, while Team and Enterprise plans are not yet supported.

Limitations of the preview version

Anthropic explicitly warns that this is an early version: the agent may misunderstand correspondence or misjudge urgency. It checks connected accounts every few hours so as not to exhaust plan limits, so something urgent may arrive between checks. Background checks and large tasks count toward the same limits as regular chats. Also, for now you cannot watch the agent work in the browser, and Microsoft 365 is not yet supported.

The launch of Claude Duet is part of a broader race for "personal agents" that don't just answer questions but take on routine work. CNBC notes that Anthropic is thus competing with similar products from other companies.

Sources: Anthropic, CNBC, TechCrunch.

Note: This post may be AI-assisted and then human-edited. Editorial responsibility: Dmytro Lomaka, Kaltenkirchen.