Anthropic Launches Claude Duet AI Agent for Professionals

Anthropic is expanding beyond conversational AI with Claude Duet, a standalone agent designed to identify what needs a user's attention and carry out work across connected applications without waiting for a prompt at every step.

Announced on October 21, 2026, the product is aimed primarily at professionals and small-business owners who need help managing communications, schedules, administrative tasks and more complex projects.

Unlike a conventional chatbot, Claude Duet is designed to continue working after a user steps away.

It can review connected applications, prepare a morning briefing, monitor for developments and notify users when something requires attention.

Anthropic says the system is built around Claude Opus 5.5 and is available through a research preview, rather than a general public release.

What Claude Duet does differently

Traditional AI assistants generally respond when users submit questions or instructions.

Claude Duet is designed to take a more proactive role: it can identify relevant information, organize work and return with updates without requiring users to initiate every interaction.

Anthropic says the agent prepares a short briefing each morning based on connected applications.

During the day, it checks those applications every few hours and sends a message when something appears to need the user's attention.

Users can also assign tasks directly, ranging from scheduling meetings to creating financial forecasts, dashboards, prototypes and web applications.

The goal is to reduce the effort involved in coordinating work across multiple tools.

Instead of manually checking email, calendars and business applications one by one, a user can ask the agent to bring relevant information together and prepare the next steps.

That does not mean every task can be completed without supervision.

Claude Duet is still in an early research preview, and Anthropic acknowledges that it can misjudge urgency or misunderstand a thread.

Its current design is focused more strongly on professional work than on personal errands such as booking tickets.

Which applications and services can Claude Duet use?

Claude Duet connects to business tools through the Claude Marketplace, which Anthropic says provides access to more than 6,000 connectors and plugins.

These integrations allow the agent to work across applications rather than confining a task to a single chat window.

Examples described by Anthropic and reported by Reuters include:

• Gmail and Slack: Find messages that need attention and help prepare replies.
• QuickBooks and Stripe: Review unpaid invoices and draft payment reminders.
• Gusto: Help prepare payroll information, including hours, bonuses and reimbursements.
• HubSpot: Draft follow-up messages using customer and sales information.
• DocuSign: Prepare agreements for review and signature.
• Canva: Help develop social media campaigns.
• Shopify: Set up and manage an online store.

The availability of a connector does not mean every possible action within that service is automatically permitted.

Access depends on the integration, the user's authorization and Claude Duet's safeguards.

Some capabilities are also still under development.

Claude Duet can also support certain purchases through Stripe's Link digital wallet and Shopify's Shop Pay.

Anthropic says these payment flows include additional confirmation steps rather than giving the agent unrestricted control over a user's payment credentials.

How Claude Duet handles permissions and sensitive actions

Autonomous software introduces a difficult question: how can an AI agent complete useful work without gaining more authority than the user intended?

Anthropic says Claude Duet requires approval before performing consequential actions such as sending a message, making a payment, signing a document or deleting something.

The user can review the proposed action before it proceeds.

Some actions remain prohibited even when a user approves them.

According to Anthropic's announcement, Claude Duet cannot trade securities or move money between accounts.

These restrictions distinguish the ability to prepare work from permission to execute every possible financial or administrative operation.

Anthropic also says Claude Duet does not directly see users' passwords or sign-in tokens.

Its credential-handling approach is based on the company's Claude Managed Agents design, while safeguards are intended to protect against malicious instructions embedded in webpages or other content the agent encounters.

Such attacks, commonly known as prompt injection, attempt to manipulate an AI system by placing instructions inside material it is asked to process.

Anthropic adds further security protections

On October 23, Anthropic published additional information about Claude Duet's safety, privacy and security architecture.

The company said the product uses model-level safety training alongside protections designed specifically for agents working across connected accounts.

Its announced measures include adversarial testing, penetration testing by an external firm, coverage under a public bug bounty program and a dedicated classifier intended to reduce the risk of sensitive information being disclosed improperly.

Anthropic says the system is designed to reject requests to share highly sensitive information, including credit card numbers and government identification.

Information such as health or personal financial details is subject to additional restrictions and is allowed through only when explicitly instructed by the user.

These are safeguards described by the company, not proof that the product is immune to security failures.

Claude Duet is still in a research preview, and its protections will need to be evaluated as people use it in real workflows.

When can users access Claude Duet?

As of October 26, 2026, Anthropic's announcement describes Claude Duet as a research preview available through a waitlist, initially for subscribers in the United States on Claude Pro and Max plans.

It can be accessed through a standalone app on iOS and Android, as well as through the Claude website.

The preview is included in eligible Pro and Max subscriptions, but it does not come with a separate, unlimited usage allowance.

Claude Duet shares the user's existing Claude plan limits, and its background checks count toward those limits.

Larger projects can consume substantially more usage than a typical short conversation because they require additional model interactions and background work.

Users who want access must join the waitlist and wait for an invitation.

The announcement does not establish a general availability date or confirm when access will expand beyond the initial U.S. group.

Why Claude Duet matters in the AI agent race

Claude Duet arrives as AI companies compete to build agents that can complete tasks across applications rather than simply generate answers.

Reuters has positioned Anthropic's launch alongside OpenAI's Dots agents and Meta's Muse AI agent.

These products reflect a broader industry shift toward software that can manage ongoing goals, use connected services and perform multi-step tasks with less continuous direction from a person.

For businesses, the potential value is straightforward: employees spend considerable time collecting information, coordinating meetings, following up with customers and moving data between applications.

An agent that can reliably handle parts of that work could reduce manual effort and help teams identify problems earlier.

However, usefulness depends on more than the number of connected applications.

The agent must understand which information matters, respect access permissions, avoid taking unintended actions and make it clear when human judgment is needed.

Errors in an ordinary summary may be inconvenient; an error involving a customer communication, payroll preparation or financial record can have more serious consequences.

Anthropic's approval requirements and limits on certain financial actions show that the company is not presenting Claude Duet as an unrestricted autonomous operator.

Instead, its design combines background task execution with checkpoints for consequential decisions.

The central question will be whether users can trust the agent's judgment enough to delegate meaningful work while retaining appropriate control over the results.

What remains uncertain

Claude Duet is still at an early stage, and several important questions remain unanswered publicly.

Anthropic has not announced a general release date or confirmed when the research preview will expand to other countries or subscription tiers.

Its initial announcement also does not provide independent performance benchmarks showing how accurately the agent prioritizes tasks, how frequently it makes mistakes or how much time it saves across different types of businesses.

Usage limits may also influence how people use the product.

Because background monitoring and complex projects consume the same allowance as other Claude interactions, users will need to determine whether the benefits justify the additional usage.

The October 23 security update provides more detail about Anthropic's protections, but real-world reliability will depend on how the system performs across different applications, permissions and unexpected situations.

For now, Claude Duet is best understood as an early attempt to turn Claude from a primarily conversational assistant into a proactive work partner.

Its success will depend on whether it can consistently identify the right tasks, complete them accurately and recognize when it should stop and ask for human approval.