Anthropic Debuts Claude Duet AI Agent Powered by Claude Opus 5.5
Anthropic has launched Claude Duet, an AI agent for professionals built on Claude Opus 5.5. The research preview features virtual machine browsing and workflow automation.
Anthropic has unveiled Claude Duet, a new AI agent aimed at professionals and small business owners. The tool positions Anthropic directly against rival agentic solutions, including Meta’s Muse and OpenAI’s Dots. Currently launching as a research preview, Claude Duet is accessible across mobile platforms via dedicated apps on iOS and Android, as well as on the web through the main Claude interface.
Powered by Claude Opus 5.5 and Virtual Machines
At the core of Claude Duet is Anthropic’s Claude Opus 5.5 model. The agent is built to integrate deeply into a user's daily work by leveraging memory, custom skills, and connected applications linked through Claude. To perform complex operational tasks, Claude Duet makes use of an internal virtual machine equipped with its own web browser.
In daily operation, the agent provides a brief summary every morning highlighting crucial updates across a user's connected software tools. Throughout the remainder of the day, it continuously monitors these connected applications in the background, sending notifications only when an item specifically requires user attention.
Users can delegate a wide spectrum of responsibilities to Claude Duet. On the simpler end, it handles routine administrative tasks such as scheduling meetings. On a higher technical level, it can construct complex performance dashboards using dozens of integrated tools, generate financial or sales forecasts, build prototypes, and assemble complete web applications.
To ensure user control over sensitive actions, Claude Duet incorporates a strict approval mechanism. Before carrying out major steps—such as sending out a payment or permanently deleting an item—the agent actively prompts the user for explicit confirmation.
Security Controls and Early Limitations
Because Claude Duet interacts with sensitive business context and data, Anthropic has implemented dedicated safety mechanisms. The system uses a specialized classifier along with a dedicated safety system prompt. These safeguards are engineered to automatically reject any user requests to share the strictest tier of sensitive data, such as credit card numbers.
At the same time, Anthropic notes that Claude Duet is still an early version and possesses clear operational limits. The company cautions that the agent may occasionally misread an email or chat thread, or miscalibrate which events actually require urgent user attention. Additionally, the agent is acknowledged to be less capable when managing personal errands, such as booking tickets for events or travel.
For those interested in testing the preview, Anthropic has opened a waitlist. Access is currently restricted to Claude Pro and Max tier subscribers located within the United States.
What it means for developers
The introduction of Claude Duet underscores the growing industry trajectory toward full-featured AI agents capable of operating web browsers and managing complex application stacks. By pairing Claude Opus 5.5 with a dedicated virtual machine, Anthropic demonstrates how foundational models can extend beyond conversational chat into active task execution, such as building web applications and drafting financial forecasts.
For developers exploring autonomous agent design, Anthropic's integration of safety classifiers and mandatory human confirmation steps offers a concrete blueprint for managing high-risk actions. As teams experiment with building or evaluating agentic systems powered by top LLMs, developers can try top AI models cheaply through one API at https://apixoai.online.
As Anthropic gathers feedback from its research preview, developer focus will remain on how effectively AI agents can maintain reliability when handling multi-tool background tasks while minimizing false notifications and operational mistakes.
Source: Anthropic takes on Muse with Claude Duet, its new AI agent — The Verge AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
