Franklin AI News Brief

Claude Duet previews a proactive assistant with approval gates for actions

Key Takeaways

  • The US waitlist preview connects Claude to ongoing personal tasks, using a separate virtual machine and shared Pro or Max usage limits.
  • Anthropic introduced Claude Duet on October 21 as an assistant that can follow ongoing tasks across connected apps.
  • Its research preview begins through a waitlist for US Pro and Max subscribers.
  • The distinction between a preview invitation and general access is important for anyone expecting to turn it on immediately.
  • Background work on a separate computer In Anthropic’s introduction, Duet uses a virtual machine with its own browser.

Anthropic introduced Claude Duet on October 21 as an assistant that can follow ongoing tasks across connected apps. Its research preview begins through a waitlist for US Pro and Max subscribers. The distinction between a preview invitation and general access is important for anyone expecting to turn it on immediately.

Background work on a separate computer

In Anthropic’s introduction, Duet uses a virtual machine with its own browser. Background work therefore does not require the user’s laptop or phone to remain switched on. The assistant can draw on existing Claude memory, skills and connected applications, with a continuing conversation accessible through text or voice.

Examples include preparing a morning briefing and following tasks that involve more than one application. These are the company’s descriptions of the intended workflow, rather than results from a Franklin hands-on test. Connecting an account gives the assistant a source of context; it does not establish that every task can be completed reliably.

Proactive does not mean constant monitoring

Duet checks connected accounts every few hours. That cadence is different from a live alerting service, so time-sensitive tasks need to be considered in light of the actual refresh interval.

During the research preview, Duet is included in Pro and Max plans and shares their usual usage limits. Background activity is consequently part of the same allowance as other Claude work. Users evaluating the preview should look at what it checks, how often it acts and how that affects the rest of their usage.

Actions still have approval boundaries

Anthropic says Duet shows the proposed action and waits for approval before sending messages, making payments, signing documents or deleting anything. Those checkpoints are a central part of the product’s design, particularly when a task crosses from gathering information into making a commitment.

An initial trial could start with a briefing or research task, then inspect the proposed next step before enabling more consequential work. The announcement explains the intended controls; actual access, connected-app coverage and task quality still depend on the preview.

Our read

Franklin AI Take

Duet’s separate virtual machine makes background work possible without leaving a personal device on. The more useful boundary to inspect is the handoff to approval: a prepared message or payment should remain visibly distinct from an executed one. Shared usage limits and checks every few hours also shape which tasks fit the preview.