Claude’s official account announced the research preview of a new app, Claude Duet. It is described as a “proactive agent” that surfaces things users should pay attention to, completes requested tasks, keeps working after they step away, and checks with them before acting on their behalf.
From AI that waits to be asked to AI that moves work forward
Until now, using conversational AI has generally meant a back-and-forth: people enter questions or instructions, and the AI responds. What stands out in the description of Claude Duet is that this sequence shifts somewhat. The AI takes the initiative, saying, “This is something you should take a look at,” and then moves ahead with requested tasks. The idea is to move from a passive tool to an assistant that becomes part of the workflow.
You might think of it not as a secretary waiting to be called, but as a colleague beside you, looking at the screen with you. The app’s name, “Duet,” meaning a performance by two, also evokes the image of a person and AI working side by side.
However, the post does not offer concrete examples of what it considers something that “requires attention.” For now, we’ll limit ourselves to outlining the framework of the features.
Working while you’re away: convenience and boundaries
Another important part of the announcement is the design that lets the AI keep working after users step away. This means AI is moving beyond being a sounding board and toward becoming something that takes on tasks continuously. Many people may imagine a way of working where tasks keep moving forward while they’re tied up in meetings or traveling.
At the same time, the AI is said to ask for confirmation before acting on the user’s behalf. This combination—advancing automation while leaving the final judgment to people—most succinctly captures Claude Duet’s design philosophy.
The more we entrust to AI, the more pressing the question of what it should not be allowed to do on its own becomes. Requiring confirmation before hard-to-reverse actions such as sending emails or making payments seems like a practical compromise for using agentic AI. We’ll have to wait for further information on when and how confirmation is requested, and what kinds of actions require it.
The research preview points to a direction
The announcement points toward developing Claude into a more proactive agent. Its status is strictly that of a research preview, and it should be distinguished from a formal launch or general availability.
This is not yet the stage to scrutinize the features’ level of polish in detail. What is more interesting for readers is that the question of how to draw the line between what AI is entrusted with and what requires human approval is built into the product’s design itself. When proactive AI becomes commonplace, the role of pressing the confirmation button at work is likely to carry more weight than ever.
Comments