Anthropic is launching Claude Duet, an AI agent that doesn't just wait for requests but takes on tasks on its own, even when the user isn't online. With this, the company is catching up with similar products from xAI, Meta, and OpenAI. This is interesting for working professionals and small businesses, because it could change the way people handle emails, chats, and payments in everyday life.

What happened

According to The Decoder, Anthropic is launching a proactive AI agent with Claude Duet. "Proactive" means that the program doesn't only become active once it receives a command, but keeps working in the background and completes tasks independently. This also applies when the user is offline.

The key points from the reports at a glance:

  • Technical basis: The app runs on Claude Opus 5.5, a language model from Anthropic.
  • Connection to services: Via more than 6,000 connectors, Claude Duet accesses applications such as Gmail, Slack, or Stripe.
  • Approval for sensitive actions: Before delicate steps, such as payments, the agent first obtains the user's consent.
  • Competition: According to The Decoder, Anthropic is responding to comparable products from xAI (Grok Bots), Meta (Muse), and OpenAI (Dots).

What The Verge adds

The Verge describes Claude Duet as aimed primarily at working professionals and owners of small businesses. The product launches as a "Research Preview." With this label, Anthropic marks it as an early version that is still being developed and in which features may change.

Claude Duet is available as a standalone app for iOS and Android as well as through Claude on the web. According to Anthropic's description, the agent sends a short overview of the most important items from the connected apps every morning. Over the course of the day, it keeps an eye on these apps and only gets in touch when something needs attention. In addition, users can hand tasks over to it, which it then works through.

t3n also reports on the launch and presents it as an attack on Meta's Muse and OpenAI's Dots. Claude Duet is described there as an agentic assistant.

Background

To put the news in context, it helps to look at how AI assistants have developed. The first popular chatbots worked according to a simple pattern: the user asks a question, the model answers. After that, nothing happens until the next question comes. This is practical for drafting, summarizing, or looking things up, but it requires the human to trigger every step.

From chatbot to agent

With AI agents, this pattern shifts. An agent is given a goal, plans several steps on its own, and uses tools to do so: it reads messages, sorts appointments, creates drafts, or triggers actions in other programs. For this to work, it needs access to the services in which the work actually takes place. This is exactly where connectors come into play, that is, standardized connections between the AI model and external applications.

The next step: proactivity

Most agents nevertheless react to assignments. A proactive agent goes further: it continuously observes what is happening in the connected services and decides on its own when it is worth addressing the user. This is more like an assistant in the office who goes through the mail and only asks questions when something is important, than like a tool that one picks up deliberately.

The fact that several major providers are moving in this direction at once is shown by the competition that The Decoder describes. xAI has Grok Bots, Meta has Muse, OpenAI has Dots, and now Anthropic follows with Claude Duet. For Anthropic, this is an obvious step: the company is known above all for its Claude models, and a dedicated app for continuously running tasks brings these models closer to users' everyday lives.

More on this topic:

What this means

Whether Claude Duet actually saves time in everyday use cannot yet be judged from the reports so far. Since it is a research preview, caution is warranted. Still, a few practical implications for different groups can be drawn.

For small businesses and the self-employed

People who handle many tasks on their own often lose time switching between their inbox, chat and payment services. An agent that delivers an overview in the morning and only reports during the day when needed could reduce these interruptions. The fact that services such as Gmail, Slack and Stripe are connected fits the typical workflows of small businesses: customer inquiries, internal coordination and payments.

Once such access is open, however, it is worth checking carefully what the agent is allowed to do. Anyone who gives it access to payment services should look at the approval settings and, when in doubt, initially hand over only low-risk tasks.

For developers

The figure of more than 6,000 connectors shows that providers are betting heavily on integration. For developers, this means: anyone who runs their own services or software may increasingly face agents accessing them through standardized interfaces. It may be worthwhile to document interfaces cleanly, to grade permissions finely, and to think about how an agent's actions can be distinguished from those of a human.

For private users

Even those who do not run a company can benefit from a proactive assistant, for example through a short morning summary. At the same time, with every connected app the amount of personal data that a service can see grows. It is advisable to link only the apps that are truly necessary and to regularly check which accesses are active.

What to look out for

An agent that works even when you are away raises a number of questions. The following checklist can help when trying out such a product:

  • Check permissions: Which accounts are connected, and what is the agent allowed to read or change there?
  • Use approvals: For payments and other sensitive actions, the confirmation prompt is an important layer of protection. You should not turn it off just to save time.
  • Spot-check the results: Even good models can make mistakes, for example when categorizing messages or phrasing replies.
  • Start small: An early version is initially suited to tasks where a mistake does not cause great harm.

Competition as an opportunity

The fact that several providers are bringing out similar products in parallel can benefit users: features, usability and security mechanisms get compared, and what works well is more likely to prevail. At the same time, things become less clear. Anyone who chooses an agent often links many accounts to it, and a later switch can mean effort. It is therefore worth comparing the offerings calmly before connecting sensitive services.

Fre­quently asked questions

What is Claude Duet?

Claude Duet is a proactive AI agent from Anthropic. It completes tasks on its own, even when the user is offline.

Which model is Claude Duet based on?

The app is based on Claude Opus 5.5.

Which services can Claude Duet access?

Through more than 6,000 connectors, the agent accesses services such as Gmail, Slack or Stripe.

Does the agent act on its own when it comes to payments?

No. Before sensitive actions such as payments, Claude Duet first asks the user for approval.

Which products does Claude Duet compete with?

Anthropic is responding to similar products from xAI, Meta and OpenAI, including Grok Bots, Meta Muse and OpenAI's Dots.