Anthropic launches Claude Duet, an always-on AI agent that works while users are offline
Anthropic has introduced Claude Duet, a proactive AI agent built on Claude Opus 5.5 that completes tasks autonomously, connects to services such as Gmail, Slack and Stripe through more than 6,000 connectors, and requests approval before sensitive actions like payments.
Anthropic has launched Claude Duet, an AI agent designed to carry out tasks on its own even when the user is offline. The product marks the company's entry into a growing field of always-on assistants, placing it in direct competition with similar offerings from xAI, Meta and OpenAI.
Claude Duet runs on Claude Opus 5.5, Anthropic's latest model, and is built to act proactively rather than waiting for instructions. It connects to external services including Gmail, Slack and Stripe through a network of more than 6,000 connectors, allowing it to move between communication, workplace and payment tools within a single workflow.
Because the agent can operate without a user present, Anthropic has built in a safeguard: Duet asks for approval before carrying out sensitive actions such as payments. The design reflects a central tension in the emerging market for autonomous assistants, where the value of uninterrupted task completion must be weighed against the risk of actions taken without direct oversight.
The launch positions Anthropic against a cluster of rival products. xAI offers Grok Bot, Meta has developed Muse, and OpenAI is preparing Dots. Each company is betting that users will want assistants that do not simply answer questions but continue working in the background, handling routine and multi-step tasks across the software people already use.
For Anthropic, Duet extends the company's strategy beyond conversational models into agentic tools that act on a user's behalf. The breadth of the connector network is central to that pitch: an agent that can reach email, team messaging and payment infrastructure can complete tasks that would otherwise require a person to switch between applications and make decisions at each step.
The approval requirement for payments suggests Anthropic is trying to draw a line between routine automation and actions with financial consequences. How that boundary is enforced in practice, and how users respond to an assistant that acts while they are away, will shape whether always-on agents become a standard part of daily work or remain a niche tool for early adopters.
The competitive field is still forming. With xAI, Meta and OpenAI all pursuing similar products, the coming months are likely to bring rapid iteration on features, safeguards and the range of services that agents can access. Anthropic's emphasis on connectors and on pre-approval for sensitive steps sets the terms of its own offer, while the wider contest will turn on trust as much as capability.
13
