Anthropic introduced Claude Duet on October 21, a proactive AI assistant that periodically checks connected applications and prepares work without waiting for users to initiate every task. It can assemble morning briefings and draft follow-up messages, potentially reducing the time professionals spend reviewing accounts for unfinished work.
Duet still requires human approval before consequential actions, making the quality of its preparation important. If users must independently verify the information behind every recommendation, the time saved by delegating routine work may be limited.
Claude Duet checks connected accounts every few hours
Claude Duet is available through a research-preview waitlist for US Pro and Max subscribers, with access through dedicated mobile apps or the web. Team and Enterprise plans are not included. Powered by Opus 5.5, the assistant can use existing Claude memory and skills alongside connected applications, while a cloud-based virtual machine allows it to continue browser tasks even when the user's devices are offline.
Anthropic says Duet checks connected accounts every few hours rather than continuously. This schedule may be sufficient for reviewing overdue invoices or preparing daily summaries, but a last-minute meeting cancellation could go unnoticed until the next review. Professionals who depend on immediate alerts would still need separate notifications for time-sensitive developments.
How Duet prepares business tasks for human approval
Anthropic says Claude Duet can access more than 6,000 Marketplace connectors and plugins, with examples ranging from drafting Gmail replies to checking unpaid invoices in QuickBooks or Stripe. The company also describes payroll preparation through Gusto and customer follow-ups in HubSpot, although it has not published task-completion rates for these workflows.
For unpaid invoices, Duet could retrieve the outstanding balance and prepare a reminder with the customer's account information. The owner would then review the amount and recipient before sending it, potentially avoiding the separate work of locating the invoice and composing the message.
Payroll preparation requires a more extensive review because an omitted reimbursement could leave an otherwise complete-looking worksheet inaccurate. An owner would need access to the underlying payroll records and supporting entries to identify missing amounts before approving payment.
Anthropic requires approval before sending messages, making payments, signing agreements, or deleting documents. Securities trading and transfers between financial accounts remain prohibited even with permission.
Although these requirements keep consequential decisions with the user, frequent authorization requests could create another administrative workload if Duet does not provide the supporting information needed to assess each proposed action. Its time-saving potential depends on whether users can review recommendations without reopening every connected application to verify the underlying information.
Security safeguards do not guarantee accurate decisions
Anthropic says Duet uses credential protections and automated checks to detect potentially unsafe actions. Its safeguards also include defenses against prompt injection, where malicious instructions embedded in connected content attempt to redirect an assistant's behavior, and filters intended to prevent disclosure of sensitive financial or identity information. The company has not demonstrated how reliably these protections work across real-world business workflows.
Even when an assistant follows its access permissions, it can misinterpret information retrieved from connected applications. A customer might request a change through a recent conversation that Duet overlooks, resulting in a follow-up based on outdated instructions.
Anthropic acknowledges that Duet can misinterpret conversations or task urgency. Users also cannot currently inspect its browser activity, limiting their ability to reconstruct how the assistant reached a recommendation. The Verge's launch coverage identifies these limitations, though independent evaluations of Duet's task accuracy remain unavailable.
How Duet compares with Meta Muse and OpenAI Dots
Meta's Muse for Small Business also monitors connected services and can identify work without a direct request. Available in the US and Canada, it connects to applications including QuickBooks and Shopify, while requiring approval before publishing content, sending messages, or spending money.
Muse's similarities to Duet suggest that proactive monitoring and human authorization are becoming common features among business AI assistants. Differences in supported applications and task performance may ultimately matter more to businesses than the ability to initiate work without a prompt.
OpenAI Dots has begun rolling out to Pro and Business Premium users in eligible markets, with an admin-enabled Enterprise beta. Dots uses a cloud computer to perform ongoing work across connected applications, while its action-review process evaluates proposed steps against user instructions and safety requirements.
Dots' administrator-enabled Enterprise beta gives organizations a deployment option unavailable in Duet's individual-plan preview. Duet, meanwhile, may require less initial configuration for existing Claude users because it can draw on their account memory and skills. Choosing between the assistants would also require testing how accurately each handles the organization's existing business workflows.
Background tasks consume existing Claude plan limits
Claude Duet's background checks and delegated tasks count against the user's existing Claude plan limits. More frequent monitoring or complex assignments could reduce the available capacity for ordinary conversations and document work, although Anthropic has not published typical usage figures for a full working day.
Dots uses a different accounting model: conversations do not count toward ChatGPT limits, while tasks involving Codex or ChatGPT Work use their respective allowances. The distinction could affect operating costs for professionals who rely on ongoing monitoring alongside resource-intensive tasks.
Determining whether Claude Duet reduces admin work will require more than counting the tasks it prepares. Evaluations should measure how accurately it identifies pending work and how often users must correct its recommendations, while tracking missed urgent items and the Claude usage consumed during ordinary working weeks.
Duet's periodic checks and approval requirements make it better suited to preparing routine work than independently managing urgent or financially sensitive decisions. Its value will depend on whether the time saved gathering information exceeds the time users spend reviewing proposed actions.
