Introducing Claude Duet
On October 21, 2026, Anthropic announced a research preview of Claude Duet, a standalone agent application powered by the Claude Opus 5.5 model. Unlike previous Anthropic offerings integrated into the main platform, Duet functions as a dedicated "proactive agent" available on iOS, Android, and the web. The app is currently rolling out via a waitlist for Pro and Max subscribers in the United States.
Functionality and Safety Measures
Duet is designed to assist professionals and small business owners by managing background tasks across more than 6,000 integrated services, including Gmail, Slack, HubSpot, and Shopify. The agent operates within a virtual machine, allowing it to function even when the user's devices are offline. A core feature of Duet is its "approval gate"; the system requires explicit user authorization before performing critical actions such as sending messages, making payments, or signing documents. Anthropic claims the system uses a credential design that prevents the agent from directly accessing passwords and employs a classifier to block the sharing of sensitive data like government IDs.
Market Positioning and Challenges
The launch marks a significant shift in Anthropic's strategy, positioning it directly against competitors like OpenAI and Meta, who have already deployed similar agent technologies. Anthropic differentiates Duet by focusing on professional workflows and implementing strict guardrails on financial movements. However, the product faces scrutiny regarding its impact on usage limits, as background polling consumes the same quotas as standard model interactions. While early feedback from power users has been mixed—noting both the tool's utility and its current limitations in email drafting and custom connector reach—the company has promised further safety disclosures in the coming days. The ultimate success of Duet will likely depend on whether it can prove its reliability and security for small business operations without becoming prohibitively expensive due to model usage costs.
Source: AI Daily News. This story was rewritten and translated for our readers.