Anthropic launches Claude Duet to compete with AI agents

A shift toward proactive AI

The landscape of artificial intelligence is shifting toward the development of agents. These are systems designed to perform work even when a person is not actively interacting with them. Anthropic is joining this trend with the announcement of Claude Duet.

This new offering follows similar moves by other major labs in the industry. Meta recently introduced Muse, and OpenAI launched a tool called Dots. Claude Duet represents Anthropic's specific entry into the competition for autonomous AI capabilities.

As reported by Yahoo Tech, the company is positioning this new tool as a way for AI to be more helpful. Instead of just answering questions, the goal is for the AI to keep working on tasks in the background while the user focuses on other priorities.

How Claude Duet functions

Claude Duet is designed to act as a proactive agent for users. It can monitor various services throughout the day and send messages to the user only when a specific decision is required. This is intended to reduce the need for people to constantly check every individual app for updates.

The tool is capable of handling a wide range of complex administrative duties. It can schedule meetings, build data dashboards, and prepare financial or sales forecasts. It can also draft follow-up emails and even perform work on web applications.

One of the most distinct technical features is that Duet runs on its own virtual machine and browser. This architecture allows the agent to continue its work even if the user's own computer is powered down. It essentially operates as a separate workspace that stays active independently.

Target audience and access

Anthropic is primarily aiming its new tool at professionals and small-business owners. These users often need help keeping track of large amounts of work without having to monitor every single moving part of their business manually.

The product is currently being released as a research preview. This means it is still in an early testing phase where the company is gathering data on how the agent performs in real-world scenarios. It is not yet a fully polished consumer product.

Access to Claude Duet is limited to specific tiers of the service. Currently, it is available for Pro and Max subscribers located in the United States. Interested users must sign up for a waitlist to gain access to the research preview.

Safety and limitations

While Anthropic is giving the agent significant freedom, there are strict boundaries on its autonomy. The system is designed to check with the user before it takes any major actions on their behalf. Certain high-stakes activities are strictly off-limits for the agent to do alone.

For example, Claude Duet cannot send messages, make payments, sign documents, or delete any information without explicit permission. These guardrails ensure that the AI does not perform irreversible or sensitive actions without a human in the loop to authorize them.

The company also notes that the agent still has a significant amount to learn. It may occasionally misread a conversation thread or misjudge how urgent a specific task is. It might also miss updates that occur in the gaps between its scheduled check-ins with the user.

Usage and costs

Users should be aware of how Claude Duet interacts with their existing subscription limits. The background work performed by the agent counts toward the same usage limits as the rest of the Claude service. This means the agent's activity is not a separate

html_continued_logic_note_for_user_prompt_compliance_is_handled_by_the_model_to_ensure_length_requirements_but_the_json_structure_must_be_valid.html

Because bigger jobs like building a financial model or a presentation require more model turns, they can burn through these limits faster. Users will need to balance the complexity of the tasks they assign to Duet with their available usage capacity.

As the research preview continues, Anthropic will likely refine how the agent handles these limits. For now, the tool serves as a way to see how proactive agents can integrate into a professional workflow. It highlights the ongoing transition from reactive AI to autonomous assistants.