claude.com via Hacker News

Anthropic Opens Claude Duet, an Opus 5.5 Agent, to US Pro and Max

6 sources tracking this story
Anthropic Agents ai-business

TL;DR

  • Claude Duet connects to 6,000-plus plugins via Opus 5.5 but launches US-only via waitlist, with Team and Enterprise plans explicitly excluded.
  • Background agent activity counts against the same plan limits as standard Claude chat, creating a usage paradox for the users Duet is designed to help most.
  • Anthropic gates consequential actions including payments, signatures, and deletions behind explicit approval, a deliberate design contrast to Meta Muse's autonomy-first model.

"Get back to what needs you," reads the top of the new Claude Duet product page. Anthropic describes it as "a proactive agent that understands what needs your attention. It takes on tasks of all sizes, keeps working while you're away, and waits for your ok when it matters."

The product is powered by Opus 5.5 and is being rolled out as a research preview on a waitlist for Pro and Max subscribers in the US. Team and Enterprise plans are not included yet. The page says Duet "works with more than 6,000 connectors and plugins," with logos for Stripe, 1Password, Shopify, Gmail, Google Calendar, Slack, Intuit QuickBooks, DocuSign, Google Drive, HubSpot, Gusto, Microsoft 365, Notion, Canva and Ramp.

Instead of waiting to be prompted, the page says, "Claude Duet comes to you. It tells you what needs your attention each morning, takes tasks you hand it by message, and reaches out when one is done or needs your approval." Example jobs on offer include "finding unpaid invoices and drafting the reminders, preparing payroll for you to review, and getting contracts ready to send," plus putting together "a brief, deck, or dashboard."

Sensitive actions are gated. "Claude Duet asks you before sensitive operations like sending messages or paying," the page says, and "approvals, logins, and safety checks run in your own cloud computer, so sensitive details stay private."

The caveats are stated plainly. "Claude Duet is a research preview. It may misread a thread or miscalibrate what's urgent, and it can miss something that arrives between checking in with you." Background work is not free either: "its background checks count toward those limits," meaning a Duet left to monitor an inbox draws from the same Pro or Max cap as a chat window. It joins a run of agent-shaped Claude releases from Anthropic this month.

What others are reporting

Coverage cluster as of 2h after publish

  1. 9to5Mac Read →

    Consumer workflow framing: details the morning-brief feature and flags that background activity burns plan limits faster than chat use alone, a practical constraint for heavy users.

    Claude Duet is a research preview of Claude that you message the way you'd message a person.
  2. Crypto Briefing Read →

    Focuses on the usage-limit paradox: the more helpful Duet becomes at background tasks, the faster it consumes the plan capacity users need for foreground work.

    Duet watches the apps where your work piles up, picks up tasks on its own, and checks with you before doing anything consequential.
  3. Tech Insider Read →

    Frames approval-gating as a genuine design philosophy split, not a feature gap, and notes neither company has published independent failure-rate data to validate their approach.

    An agent that pauses for approval is safer by construction, but only if users actually read what it is asking them to approve.
  4. Roos Newsletter Read →

    Audits help-center docs against the product page, surfaces the Microsoft 365 gap and the email-matching requirement, and flags the name collision with an unrelated npm package.

    Claude Duet is a messaging-style AI agent that comes to you instead of waiting for a prompt.
  5. Shattered.io Read →

    Provides four-lab competitive mapping and reads the staged US-only subscriber-gated rollout as risk management over speed, noting the Enterprise exclusion is a deliberate sequencing choice.

    It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.