Anthropic launches Claude Duet to automate business workflows

From Chatbot to Agent

Anthropic is moving beyond simple question-and-answer interactions with the release of Claude Duet. This new application is designed to act as a proactive assistant for professionals and small business owners who need to manage competing priorities.

Unlike standard AI tools that wait for every specific command, Claude Duet is built to handle a user's backlog independently. It aims to use judgment to determine which tasks it can complete on its own and which problems it needs to surface for human intervention. As Anthropic noted, the goal is to help people reclaim time for work that truly demands their focus.

The app runs on the Claude Opus 5.5 model and is accessible via the web, iOS, and Android. It is currently in a research preview phase, meaning it is being tested to see if the AI can reliably distinguish between routine work and complex decisions that require a human touch.

Independent Workflows

One of the most significant technical features of Claude Duet is its ability to work while a user is away from their device. To achieve this, the app utilizes its own virtual machine and browser environment. This allows the AI to continue processing a project after a user has stepped away, returning a completed result or a request for a specific approval later.

The scope of these automated tasks is broad. Beyond simple organization, Claude Duet can perform complex business operations such as:

• Scheduling meetings and preparing financial forecasts.
• Building dashboards and prototyping new products.
• Developing web applications and drafting job listings for platforms like LinkedIn.
• Preparing payroll details in Gusto or identifying unpaid invoices in Stripe.

To manage the consumption of subscription limits, the app monitors connected accounts every few hours rather than constantly. Users can customize which specific developments trigger a notification and how frequently they want to be contacted.

Ecosystem and Integrations

The utility of Claude Duet is bolstered by its access to the Claude Marketplace, which features more than 6,000 connectors and plugins. This allows the AI to interact with a wide array of communication, finance, recruitment, and marketing tools. For example, it can find unanswered emails in Gmail to draft replies or use DocuSign to prepare agreements.

Retailers can specifically use the tool for Shopify management, with plans for a more robust storefront editor coming in the future. The system also supports payments through Shopify's Shop Pay and Stripe’s Link wallet. Additionally, Anthropic plans to integrate with 1Password to allow the assistant to access saved logins securely.

The company is also looking at how different types of users interact with the tool. For instance, Boris Cherny, who leads the Claude Code team, uses the app to talk through his mental to-do list while walking. Meanwhile, Chelsea Larsson, who leads model and product behavior design, uses it to prepare evening reports while spending time with her children.

Safety and Oversight

Because Claude Duet has the ability to access sensitive business accounts, Anthropic has implemented several layers of oversight. The AI is required to show users a proposed action before it executes high-stakes tasks, such as sending messages, signing documents, making payments, or deleting data. Certain actions, such as securities trading or transfers between accounts, remain strictly prohibited even if a user provides approval.

Security is further bolstered by a dedicated classifier and a system prompt designed to prevent the disclosure of sensitive information. The system is programmed to block requests to share highly confidential details like government identification or credit card numbers. While it can handle health or personal finance disclosures, it only does so when the user provides explicit authorization.

Anthropic has also subjected the infrastructure to external penetration testing and adversarial exercises. The product is also covered under the company's public bug bounty program, where external researchers can be rewarded for finding vulnerabilities. Users maintain the ability to disconnect any application at any time to ensure they retain ultimate control over their data.