Anthropic Also Enters AI Agent Race with 'Claude Duet' Reveal

What happened

On May 22 (US time), Anthropic showed a research preview of Claude Duet. It is built on Claude Opus 5.5, one of its top AI models. Unlike a normal chatbot, Duet does not wait for step-by-step orders. It watches linked files and apps and finds work to do. For now, it is open to paying subscribers in the United States who pay more than $20 a month.

Key facts

• Price limit: paid plans of $20 or more per month, US only for now
• Works on the Claude website and on iPhone and Android apps
• Connects to more than 6,000 apps, including Gmail, Slack, LinkedIn, Stripe, and Shopify
• Keeps working on its own virtual computer, even when your computer is off

What it can do

Users can ask Duet for a short daily briefing with the main news from their apps. They can also tell it to check email or messages and send replies when needed. Through Stripe and Shopify digital wallets, it can even pay for items directly. Anthropic says it made the tool for busy experts and small business owners who find it hard to manage their schedules alone.

Why it matters

AI agents that act for people, not just answer questions, are a new race among big tech firms. Giving an AI the power to pay money or read private data raises safety questions. Anthropic added safeguards: it asks for approval before sending emails, paying, signing documents, or deleting files. It bans high-risk actions like stock trades and money transfers, even with approval. It also blocks requests to share the most sensitive data, such as credit card or ID numbers. Anthropic admits the early version can misread context or judge urgency wrong.

What to watch

• Whether the tool expands beyond the US and to free users
• How well Duet avoids mistakes when it acts on its own
• Whether Meta's Muse and OpenAI's Dots gain more users