Claude Duet AI agent launches for professionals

Anthropic announced Claude Duet on October 21, 2026, a proactive AI agent designed to serve professionals and small business owners. The product competes with Meta’s Muse and OpenAI’s Dots, launching as a research preview available on iOS, Android, and through claude.ai.

What changed

Claude Duet is built on Claude Opus 5.5 and functions as a standalone app that integrates with your existing Claude account. The agent inherits your memory, skills, and connected applications from your main Claude setup.

The core functionality centers on proactive assistance. Each morning, Claude Duet sends a brief highlighting the most important items across your connected apps. Throughout the day, it monitors those apps and sends you messages only when something requires your attention. You can assign it tasks ranging from administrative work like scheduling meetings to complex projects such as building dashboards across multiple tools, creating financial and sales forecasts, prototyping, and building web applications.

A key distinction from passive assistants is that Claude Duet continues working even after you close the app. It runs on a virtual machine with its own browser, eliminating the need to keep your laptop or phone active. The company checks your connected accounts every few hours to balance responsiveness against usage plan limits.

Before taking sensitive actions, such as sending messages, making payments, signing documents, or deleting data, Claude Duet seeks your approval and shows exactly what it intends to do. Certain actions remain blocked even with user consent, including trading securities or moving money between accounts.

For access during the research preview, you must be a Pro or Max subscriber in the United States. Claude Duet uses the same usage limits as the rest of Claude. Interested users can join the waitlist. The company acknowledges the product is an early version that may still misread threads or miscalibrate urgency, and it is not yet as strong at personal errands like booking tickets.

Why it matters

Claude Duet directly addresses a pain point for knowledge workers and entrepreneurs: managing continuous flows of information and decisions across disconnected tools. Unlike AI assistants that wait for requests, this agent comes to you with prioritized information and takes action independently within defined guardrails.

The competitive landscape matters here. Meta Muse and OpenAI’s Dots already target this market. By launching Claude Duet as a research preview, Anthropic is staking a claim in the professional automation space while gathering real-world feedback to improve the product before full release.

For small business owners specifically, the ability to delegate routine monitoring and task execution across over 6,000 connected apps through the Claude Marketplace could reduce time spent on administrative overhead. Early testers included a real estate broker who appreciated the agent catching a scheduling error without being asked, a restaurant owner managing invoices across multiple locations, and a brand strategist who benefited from automated calendar analysis and proactive meeting preparation.

The safety architecture matters too. Anthropic implemented a dedicated classifier and safety system prompt designed to reject requests sharing the strictest tier of sensitive data like credit card numbers and government identification. Other sensitive information, such as health or personal finance details, passes through only with explicit instruction. This addresses a real concern about agents handling confidential business and personal information.

Claude Duet’s credential handling follows the design used for Claude Managed Agents, so the agent never directly sees your passwords or sign-in tokens. Actions are also approved by Anthropic’s auto mode classifier, the same safeguard that protects Claude and Claude Code from taking steps you did not ask for. The company ran adversarial tests including red team exercises and external penetration testing, and the product is covered by Anthropic’s public bug bounty program.

What to test

Before adopting Claude Duet, run through this practical checklist:

• Judgment calibration: Test whether the agent correctly identifies what actually needs your attention versus what can wait. Anthropic acknowledges this is still being tuned. Track false alarms and missed items over a week.

• App connectivity: Verify that all the tools your business relies on appear in the Claude Marketplace and connect properly. The company mentions over 6,000 connectors available, but your specific stack may have gaps.

• Approval workflow: Confirm that the approval process for sensitive actions feels natural rather than intrusive. Try it with payment scenarios, email sends, and document deletions to see if the balance matches your risk tolerance.

• Usage limits: Check your Pro or Max plan’s usage allowance. Since Claude Duet performs background monitoring every few hours, measure how quickly it depletes your monthly tokens.

• Personal errands: The company explicitly notes weakness here. If booking tickets, reservations, or travel arrangements is central to your workflow, test these thoroughly before relying on the agent.

• Safety classifier performance: Ask it to handle scenarios involving mixed sensitive and non-sensitive information. For example, ask it to draft an email that includes a client’s health information alongside normal business data. Verify the classifier enforces boundaries correctly.

• Integration with existing Claude features: If you already use Claude Code or Claude in Chrome, test how Claude Duet coexists with those tools and whether context transfers smoothly.

Common questions

How is Claude Duet different from regular Claude?

Regular Claude responds when you ask; Claude Duet proactively monitors your apps and initiates contact when it detects something needs attention. Claude Duet also runs continuously in the background with its own virtual browser, allowing it to complete work while you are away from your device.

Can Claude Duet access my passwords?

No. According to Anthropic, Claude Duet uses credential handling designed so the agent never directly sees your passwords or sign-in tokens. It follows the same security framework used for Claude Managed Agents.

What are the limitations?

The company identifies two main early limitations. First, Claude Duet “still might misread a thread or miscalibrate what’s urgent,” indicating the judgment system is still maturing. Second, it is “not as strong” at personal errands like booking tickets compared to administrative or analytical work. Additionally, certain financial actions like trading securities or moving money between accounts remain blocked.

The conclusion

Claude Duet represents Anthropic’s entry into the AI agent market for professionals. By building it on Opus 5.5 and integrating with Claude’s existing memory and app ecosystem, the company has minimized friction for existing users. The research preview approach lets Anthropic gather real feedback on judgment calibration and safety performance before full release.

The competitive threat to Meta’s Muse and OpenAI’s Dots is genuine. Claude Duet offers continuous background operation, integration with over 6,000 business apps, and safety guardrails specifically designed for sensitive information. Early user feedback from small business owners suggests utility for delegating routine monitoring and task execution.

The main unknowns are whether the judgment system will mature quickly enough to avoid alert fatigue, and whether usage limits remain reasonable as the agent monitors multiple apps throughout the day. Users should test these specifics before rolling Claude Duet into critical workflows. Watch for updates on safety classifier performance and personal errand capability as the research preview progresses.

Related reading

• GPT-6 vs Claude Opus 5.5: Frontier AI Models
• Voxtral 3 voice agent SDK: 50+ languages and emotional speech
• MiniMax M3.1 Flash: $0.15/M tokens cuts agent costs 70%