Claude Duet: A Proactive AI Agent That Works While You’re Away
Anthropic has introduced Claude Duet, a proactive AI assistant designed to manage tasks independently and surface only items that require human attention. Powered by Opus 5.5, the agent operates across iOS, Android, and web platforms, allowing users to delegate work without continuous oversight.
Early coverage of the launch and its autonomous capabilities is available in The New York Times.
To understand how this new assistant fits into the broader AI landscape, let’s explore its underlying mechanisms.
How Claude Duet Works: Memory, Skills, and Cloud Automation
Claude Duet runs through standalone iOS and Android applications while processing tasks in virtual machines hosted in the cloud. The agent connects to a user’s Claude account to access stored memory, specialized skills, and third-party connectors, allowing it to retrieve context and execute workflows across services without requiring the user to remain online.
When an action involves sensitive data, such as sending a message or initiating a payment, the system pauses and requests explicit user approval before proceeding. This human-in-the-loop design ensures that autonomous execution remains bounded by direct consent for high-impact operations.
Additional reporting on the agent’s architecture and competitive positioning can be found in Engadget’s coverage, while technical specifications are detailed in Shattered’s analysis.
With that technical foundation in mind, here are the key facts you need to know.
Key Facts About Claude Duet
- Pricing tiers include a free research preview, a standard monthly subscription, and an enterprise plan.
- Access is currently limited to a waitlist, with no public launch date announced.
- Usage limits apply during the research preview, including caps on autonomous task executions and cloud processing time.
- Known limitations include occasional context drift in long-running workflows and restricted third-party connector availability.
- The research preview does not support offline execution or local data storage.
- Human approval remains required for sensitive actions such as payments and outgoing messages.
These details set the stage for understanding the rollout strategy and usage constraints.
Availability, Plans, and Usage Limits
Claude Duet is rolling out as a research preview for Claude Pro and Max subscribers in the United States. Team and Enterprise plans are excluded from this initial release.
Background tasks executed by Claude Duet share the same plan usage limits as standard Claude conversations, meaning autonomous activity counts toward existing monthly caps.
Further details on Anthropic’s latest model releases are available in coverage of Claude Opus 4.5.
Looking ahead, privacy and safety considerations remain central to the agent’s design.
Privacy, Safety, and What Comes Next
Claude Duet requires explicit user approval before executing sensitive operations, such as payments or outgoing messages. Processing occurs on local cloud infrastructure rather than on-device, and users retain visibility into agent activity through session logs and real-time monitoring.
Current limitations include occasional context drift in long-running workflows and restricted third-party connector availability. Enterprise applications may include banking automation and internal workflow orchestration, as explored in Goldman Sachs’ Claude AI banking automation.
As Anthropic continues to refine Claude Duet, users can expect tighter integration, expanded connector support, and more robust safeguards.
Frequently Asked Questions
How does Claude Duet handle memory across sessions and what are its limits?
Claude Duet stores context in the user’s Claude account, allowing the agent to retrieve past interactions and maintain continuity across tasks. However, the research preview does not support offline storage, and long‑running workflows may experience occasional context drift, meaning memory may become less accurate over extended periods.
What are the pricing tiers for Claude Duet and how do usage limits differ between them?
Claude Duet is offered in a free research preview, a standard monthly subscription, and an enterprise plan, with access currently limited to a waitlist. Autonomous task executions and cloud processing time count toward the same monthly caps as regular Claude conversations, so higher‑tier plans provide larger caps and fewer restrictions on autonomous activity.
How does Claude Duet ensure privacy and safety when performing sensitive actions?
Before executing high‑impact operations such as payments or outgoing messages, Claude Duet pauses and requires explicit user approval, keeping the user in the loop. All processing occurs on Anthropic’s local cloud infrastructure, and users can monitor activity through session logs and real‑time monitoring to verify that no unauthorized actions are taken.
Last Updated on octubre 22, 2026 6:20 am by Laszlo Szabo / NowadAIs | Published on octubre 22, 2026 by Laszlo Szabo / NowadAIs
