Anthropic announced on Wednesday (21) Claude Duet, an artificial intelligence (AI) agent capable of organizing tasks, tracking emails and calendars, and continuing to work even when the user closes the app. The release intensifies the competition with OpenAI's ChatGPT Dots and Meta Muse in a market that is betting on assistants capable of performing tasks, rather than merely answering questions.

The feature is being rolled out gradually as a research preview to Pro and Max plan subscribers in the United States. Access is through a waitlist, with invitations released in stages.

Claude Duet comes with no additional cost for eligible subscribers, but it uses the usage limits already included in their plans. Since tasks performed in the background also count toward these limits, Anthropic warns that users may reach their limits faster than they would in regular conversations.

How does Claude Duet work?

  • Claude Duet is designed to function as an assistant that follows the user's needs and takes initiative when it identifies something that requires attention;
  • The agent can consult connected apps, organize information, and continue tasks without requiring the user to keep the app open. To do this, it operates on its own cloud-based computer, allowing work to continue even after the user steps away;
  • The system also uses the same memories, skills, and app connections available in regular Claude. In addition, it can search previous conversations to retrieve information and avoid requiring the user to repeat instructions;
  • Among the possibilities described by Anthropic are preparing a summary of the day’s priorities based on emails and calendar commitments, drafting responses for review, following up on conversations that went unanswered, and performing more complex tasks;
  • Users can also request larger projects, which may result in editable files, such as research reports, financial models, presentations, and data dashboards. Completed work is available in the Make tab;
  • Interaction can take place through typed or voice messages. It is also possible to create separate channels for different topics while maintaining shared memory between them.

The agent can perform tasks, but requires authorization

One of the central points of Claude Duet is the combination of autonomy and human oversight.

According to Anthropic, the system was designed to request authorization before sending a message, making a payment, signing a document, or deleting any content.

The tool also offers integrations aimed at business activities. With Stripe, for example, it can check unpaid invoices and prepare payment reminders. Integrations with Link and Shop Pay, meanwhile, allow purchases to be made with user approval.

The agent also supports plugins, including tools intended for small businesses, and can suggest extensions suited to the requested work.

The aim is to allow AI to advance on multi-step tasks without taking control of important actions away from the user.

TAG58@@

Logo da AnthropicOne of the central points of Anthropic’s Claude Duet is the combination of autonomy and human oversight – Image: daily_creativity/Shutterstock

Read more:

  • OpenAI races to raise R$163.5 billion before Anthropic’s IPO
  • Anthropic reveals a loss of R$435.8 billion and billion-dollar AI costs
  • Anthropic project foresees an AI shutdown in the face of catastrophic risks

Will Claude Duet be free for subscribers?

Claude Duet is included in eligible Pro and Max plans, with no additional charge for the feature.However, use of the agent consumes the plans’ allowance, including when tasks are run in the background.

Initial availability is limited to the United States and is being rolled out through invitations sent gradually to users on the waitlist.

Anthropic announced that Claude Duet is not yet available on the Team and Enterprise plans, but it intends to expand access to these options in the future.

Anthropic enters the race for personal AI agents

The launch places Anthropic in growing competition with companies seeking to turn AI assistants into agents capable of performing work proactively.

OpenAI introduced ChatGPT Dots as a tool that can run tasks in the background, work with connected apps, and contact the user when it needs attention or additional information.

Meta Muse, meanwhile, is betting on a personal assistant capable of taking on tasks and projects in different areas of everyday life.

Claude Duet follows a similar path but leverages the infrastructure of memories, skills, and integrations already available in the Claude ecosystem. The tool aims to transform the assistant into a collaborator that monitors recurring demands and continues working even when the user is not present.

With the launch, Anthropic is expanding its focus on AI agents capable of performing complex tasks and managing workflows, rather than limiting itself to the traditional question-and-answer format.

Rodrigo Mozelli
Rodrigo Mozelli

Rodrigo Mozelli is a journalist graduated from São Paulo Methodist University (UMESP) and is currently a writer at Olhar Digital.

View all articles →
Ana Luiza Figueiredo
Ana Luiza Figueiredo

Ana Figueiredo is Olhar Digital's technology editor. She graduated in Journalism from the Federal University of Uberlândia (UFU).

View all articles →
Tags: anthropic claude Artificial Intelligence