US Pro/Max Users Face Faster Usage Limits With Anthropic's Proactive Claude Duet Agent
The new proactive AI assistant manages background tasks and connects to apps, but users may hit usage limits faster.
Key points
- Claude Duet is a research preview for US Pro and Max subscribers, accessed via a waitlist.
- The agent runs on the Opus 5.5 model and operates on a dedicated cloud computer.
- It integrates with Stripe, Link, and Shop Pay, requiring user approval for payments or deletions.
Anthropic has announced Claude Duet, a proactive artificial intelligence agent designed to manage tasks across connected applications for professional users. The service is currently available as a research preview to Pro and Max subscribers in the United States, with access granted through a waitlist rather than immediate universal rollout.
The agent is built on the Opus 5.5 model and functions as an always-on assistant that continues working even when the user closes the application. According to 9to5Mac, the feature is included at no additional cost but counts toward existing plan usage limits, which Anthropic warns could cause users to reach those caps sooner than with standard chat interactions.
What we know
Claude Duet is positioned as a competitor to OpenAI’s Dots and Meta’s Muse, targeting professionals and small business owners who need help tracking complex workflows. As reported by The Verge, the agent provides daily briefings on important items from connected apps and only messages users when specific attention is required. It can handle tasks ranging from scheduling meetings to building financial forecasts and web applications.
The system uses a virtual machine with a browser to complete work and pulls memory and skills from the user’s existing Claude account. According to Engadget, Duet checks with the user before taking significant actions such as sending payments, deleting files, or signing documents. It also features a dedicated safety classifier designed to reject requests involving sensitive data like credit card numbers.
Availability is currently restricted to the US market for individual subscribers. Anthropic stated that Team and Enterprise plans are not yet supported, though the company plans to expand access to those tiers in the future. The agent is available as a standalone application on iOS and Android, as well as through the web interface of Claude.
Defender's checklist
- Review connected application permissions to ensure the agent only accesses necessary data sources.
- Monitor usage limits closely, as background tasks may consume quota faster than expected.
- Verify that all payment and deletion actions require explicit user approval before execution.
- Check that the safety classifier is active to prevent the sharing of sensitive financial information.
The background in plain language
Proactive agents differ from standard chatbots by acting on user instructions without constant supervision. They maintain context across sessions and can operate in the background, allowing them to complete multi-step tasks over time. For organisations, this shift introduces new governance challenges, as these systems can access multiple data sources and perform actions autonomously. Teams usually need to define clear boundaries for what tasks the agent can perform independently versus what requires human oversight. This approach helps prevent
Background: Virtual machines
Virtual machines emulate full hardware and run separate operating systems, providing strong isolation. Containers share the host kernel and package only application code, offering faster startup. Choose virtual machines for strict security or mixed operating systems. Choose containers for rapid deployment and resource efficiency in consistent environments.
What to do and how to stay safe
- Data classification is the primary control mechanism for preventing leaks.
- Standard frameworks provide a free structure for risk assessment.
- Delegating technical checks to vendors reduces internal workload.
AI governance for small teams relies on clear data rules and consistent human oversight, not expensive software. Start by classifying your data and defining what can be shared with AI tools.
Step-by-step guide: AI Governance for Small Teams: Practical Controls and Cost
General security guidance from the TechWire Beat newsroom. It is not confirmed advice from the organisations named in this story.
