Claude Duet: Anthropic's proactive agent that messages you first

On October 21, 2026, Anthropic opened a research preview of Claude Duet — a standalone app where Claude operates as a proactive agent instead of waiting for prompts. It delivers a daily morning brief, monitors your connected apps (Gmail, Slack, Calendar, 6,000+ connectors), keeps working on its own cloud computer while you're away, and asks for approval before sending messages, making payments, or signing documents. Built on Claude Opus 5.5. US Pro/Max subscribers only, via waitlist.

For years, AI assistants have been reactive — they wait for you to open a chat, type a prompt, and hit enter. On October 21, 2026, Anthropic broke that pattern with Claude Duet: a standalone app where Claude reaches out to you first.

What Duet actually does

Duet is not a chatbot with a new coat of paint. It is a proactive agent with four core behaviours:

• Morning briefing: Each morning, Duet sends a concise summary of what needs your attention across connected apps — emails, calendar conflicts, Slack messages, task updates.
• Task execution: You message Duet with a request ("Build a financial forecast for Q4," "Draft responses to these emails," "Research and summarise this topic"). It works on its own cloud computer, using tools and browsing, and delivers the result.
• Always-on: Duet keeps working after you close the app or put your phone down. Its cloud computer runs independently of your devices.
• Asks before acting: Before sending a message, making a payment, signing a document, or deleting anything, Duet shows you what it's about to do and waits for approval.

The model: Opus 5.5

Duet runs on Claude Opus 5.5 — the same model Anthropic previously pitted against GPT-6 Astra on Terminal-Bench. There is no model picker; Opus 5.5 is the only engine. Anthropic says this model was chosen for its balance of reasoning, tool use, and instruction following in long-horizon tasks.

Integrations: 6,000+ connectors

Duet connects to the Claude Marketplace, which hosts over 6,000 connectors and plugins. Key categories:

• Communication: Gmail, Slack, Microsoft Teams, Outlook
• Calendar: Google Calendar, Outlook Calendar
• Productivity: Notion, Linear, Jira, Asana, Google Docs/Sheets
• Finance: Stripe Link, Shopify Shop Pay, QuickBooks, Xero
• Development: GitHub, GitLab, Vercel, AWS

This breadth is what makes the proactive model viable — Duet can actually do things across the tools you already use, not just summarise them.

Guardrails and the approval model

Anthropic's design centers on user approval for sensitive actions:

• Payments: Requires explicit approval via Stripe Link or Shopify Shop Pay.
• Messages/emails: Duet drafts but you send.
• Document signing: You review and confirm.
• Deletions: Explicit confirmation required.

Anthropic acknowledges Duet can misjudge urgency and remains less effective at personal tasks like booking tickets. The research preview is explicitly for learning these boundaries.

Market context: the agent race

Duet is the fourth major entrant in the "always-on proactive agent" category that formed between early September and late October 2026:

All four launched within a 6-week window. Each lab independently concluded a persistent, cross-app, proactive agent was the next product surface.

What Duet does not do (yet)

• No Team/Enterprise access: Anthropic gives no timeline.
• No trading/securities: Duet won't trade securities or move money between accounts.
• US only: Waitlist is US Pro/Max subscribers only. No international date announced.
• Shares your usage limits: Duet consumes your existing Claude Pro/Max token allowance. Heavy agent use could hit your cap.

What to watch

• Waitlist velocity: How fast Anthropic expands from the initial US Pro/Max cohort.
• Team/Enterprise roadmap: The biggest addressable market is B2B; no timeline means this is a consumer preview first.
• Opus 5.5 vs specialised agents: Whether a single general model beats the specialised-agent approach (OpenAI Dots, xAI Grok Bot) on real workflows.
• Cost at scale: Free during preview, but Opus 5.5 API pricing ($15/$75) suggests real compute cost. Anthropic hasn't disclosed post-preview pricing.

Bottom line: Anthropic has entered the proactive agent race with a well-designed product (morning brief + cloud computer + approval gates) on its strongest model (Opus 5.5). The waitlist and US-only constraints mean most people will read about it before they can try it. The real test is whether Duet completes tasks reliably enough that the morning brief becomes indispensable rather than noisy — and that answer requires the preview to run its course.

Prices and benchmark figures are standard published rates as of October 27, 2026 and can change without notice. Some outbound links may be affiliate links — see our disclosure. Found an error? Tell us and we'll correct it in public.