Anthropic's Claude Duet is an agent built to work while users are away

Anthropic has released a new AI agent called Claude Duet that can keep working while users are away, according to a report from The American Bazaar.

The pitch is narrow but pointed: the agent is positioned around work that continues after the person who assigned it has stepped away. That is a different proposition from an assistant that answers inside a single session and stops when the user leaves the screen.

The underlying shift is toward asynchronous agency. Instead of replying turn by turn, an agent takes a task, runs it in the background and hands back a result later. Once that is the model, product design moves away from conversational polish and toward task orchestration, state management and recovery from interruption.

Working unattended also raises the control question. An agent that continues without supervision needs explicit boundaries: which systems it can touch, how much it may spend, how failures are rolled back, and whether every step leaves an auditable trace. Those are the terms on which delegated work becomes acceptable, not features bolted onto a model.

The release also fits a broader repositioning across the industry, where vendors are moving from chat interfaces toward software that completes tasks on the user's behalf. Long-running agents are becoming the next contested surface, because they sit closer to actual work than a conversation window does.

Public detail is still thin. The report gives the product name and the away-from-keyboard positioning, but not the task types it supports, how it is priced, or what administrative controls exist for teams.

What to watch is scope and safety design. For agents that run for extended periods inside real workflows, permissions, auditability and cost control will decide adoption far more than a demo does.