AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Anthropic takes on Grok Bot, Meta's Muse, and OpenAI's Dots with its always-on agent "Claude Duet"

Collected Oct 22, 2026

Anthropic has opened a research preview of Claude Duet, a standalone app in which Claude operates as a proactive agent that keeps working on tasks while the user is away. The app is pitched at professionals and small business owners who, in Anthropic's framing, have more on their plates than they can track. It runs on iOS, Android and the web at claude.ai, and is built on Claude Opus 5.5. Access is via a waitlist for Pro and Max subscribers in the US; Team and Enterprise plans are not part of the preview yet.

The product arrives into a crowded field. SpaceXAI opened the Grok Bot beta in August, Meta launched Muse in September, and OpenAI launched Dots the same month. All four are chasing the same idea: a personal agent that acts on its own initiative rather than waiting for a prompt. Anthropic's answer to that field is a preview, not general availability, so the comparison is currently between shipped betas and a limited waitlist.

The stated difference from the regular Claude app is the direction of the interaction. In the Claude app, a user opens a chat and starts a conversation. Duet pushes information outward instead: every morning it sends a short summary of the most important items from connected apps, and after that it only reaches out when something needs attention. It inherits memory, skills and app connections from the user's existing Claude account, so a user does not start from a blank agent.

Tasks are assigned by text or voice message in a single ongoing thread. Anthropic says this covers scheduling, dashboards, financial forecasts, prototypes and web apps. The agent runs on a virtual machine, its own cloud computer with a browser, which means the user's laptop or phone does not have to stay on for work to continue. That architecture is the core of the always-on claim: the agent's session lives on Anthropic's infrastructure rather than on the user's device.

Duet reaches external services through more than 6,000 connectors and plugins drawn from the Claude Marketplace. Named integrations include Gmail, Slack, Stripe, Docusign and Shopify. The illustrative workflows Anthropic describes include tracking down unpaid invoices and drafting payment reminders, preparing payroll, and other billing-adjacent chores. For purchases, the agent uses Stripe Link with a virtual card, so the user's real card number is never shared.

A permission layer sits between the agent and anything consequential. Anthropic says Duet can read, draft and prepare on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it plans to do and waits for approval. Two categories are hard-blocked even with approval: trading securities and moving money between accounts. That is a notable line for a product aimed at small businesses, since it keeps Duet adjacent to payments and invoicing without letting it become a treasury tool.

Several technical safeguards underpin the design. Anthropic says the agent never sees passwords or login tokens directly, following the credential-handling design of Claude Managed Agents. Against prompt injection, the class of attack in which hidden instructions in fetched content try to hijack an agent, Anthropic says it applies defenses from its browser-use research. Red teams and an outside firm also ran attack tests against Duet. A new classifier works alongside special safety instructions in the system prompt to check every message for leaking sensitive data; the pair is meant to refuse requests to share credit card numbers or ID details, and health or personal finance information passes only when the user explicitly says so.

Anthropic is candid about the preview's shortcomings. It says the early version can misread a message thread, misjudge urgency, and miss items that arrive between two checks. It is weaker at personal errands such as booking tickets, because the core features are built around work. Frequency is a deliberate trade-off: Duet only checks connected accounts every few hours, which Anthropic attributes to usage. Constant polling would exhaust subscription limits much faster. Duet shares those limits with other Claude products, and the background checks count against them too. In practice, this suggests the agent's responsiveness is bounded by economics as much as by capability, and users should expect hours-level latency on connected services rather than live monitoring.

Two questions remain open by Anthropic's own account. It plans to use tester feedback to tune when the agent should reach out, which implies notification behavior is still being calibrated. And how Duet fits with the main Claude app is unresolved; for now it runs separately, and early users' experience will shape how the two eventually work together.

Why it matters: Teams evaluating agentic products now have a fourth comparable option alongside Grok Bot, Muse and Dots, and the differentiating feature set is less about model quality than about permission design and connector breadth. The approval gate plus hard blocks on securities trading and account transfers sets a conservative baseline that regulated or finance-adjacent businesses may find easier to justify internally. The shared usage limits and hours-level polling are likely to matter most in practice, since they constrain how autonomous the agent can feel, at least in this preview.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Anthropic has launched Claude Duet, a proactive AI agent that handles tasks on its own, even when the user is offline. The app is built on Claude Opus 5.5 and connects to services like Gmail, Slack, and Stripe through more than 6,000 connectors. It asks for approval before sensitive actions like payments. With Duet, Anthropic is responding to similar products from xAI, Meta, and OpenAI. The article Anthropic takes on Grok Bot, Meta's Muse, and OpenAI's Dots with its always-on agent "Claude Duet" appeared first on T