An agent that acts without continuous instructions
Anthropic enters the market for AI agents for individual use with Claude Duet, presented in research preview on October 22, according to ETNews. The newspaper places the move within a competition already energized by Meta with Muse and by OpenAI with Dots: three companies aiming to transform the AI assistant from a reactive tool into an autonomous executor of tasks. The technical basis is Claude Opus 5.5, described as the company's high-end model.
The difference from a traditional chatbot lies in how it works: the user does not have to issue step-by-step orders, because the agent observes the connected data and applications and identifies on its own the activities to be performed. Anthropic justifies the development with the need of professionals and small business owners who, having too many tasks, struggle to manage their own schedules alone. The agent operates on its own virtual computer and continues working even if the user's device is turned off or the user moves away.
Integrations, payments, and safety guardrails
The service is accessible from the Claude website and the iPhone and Android apps. The initial phase is reserved for the United States and subscribers to paid plans costing at least $20 per month, who can request access. Among the indicated uses is receiving a daily briefing with key points, as well as the ability to have emails and messaging reviewed and to send communications when necessary.
The scope is broad: more than 6,000 apps can be connected, including Gmail, Slack, LinkedIn, the Stripe payments platform, and Shopify e-commerce. The relevant point is that the agent can pay for goods directly through Stripe and Shopify digital wallets, a step that moves AI from advice to the execution of economic transactions.
On this front, Anthropic has built an authorization system: sending emails, payments, signing documents, and deleting files require explicit user consent. Operations considered high-risk, such as securities trading and fund transfers, are entirely prohibited even with approval. Requests for external sharing of highly sensitive data, such as credit card numbers or identity documents, are automatically refused, while health and financial information is transmitted only on explicit instruction.
A preview with acknowledged limitations
Anthropic admits that this is an initial version and that the agent may misinterpret context or misjudge the urgency of a situation. This clarification matters, because it concerns precisely the tasks in which the agent's autonomy encounters the user's data and decisions. The material does not indicate prices beyond the $20 threshold, availability dates outside the United States, or timelines for moving from the research phase to the commercial one. It therefore remains to be seen how the scope of authorizations will evolve and what controls will be added before a broader release.
