Anthropic is rolling out Claude Duet, a new way to use its Claude chatbot that behaves less like a search box and more like an assistant you text. The company announced the research preview on Tuesday, and invitations are going out in waves to Pro and Max subscribers in the US through a waitlist.
The pitch is simple: tell Claude Duet what you need, typed or spoken, and it keeps working on it inside the apps you've already connected, even after you've closed the window. Anthropic says the agent runs on its own cloud-based computer rather than depending on an open browser tab or app session, which is what lets it keep going in the background.
That convenience comes with a catch. Claude Duet is free to use for now, but every minute it spends working counts against your existing plan's usage limits, including time spent on tasks you're not actively watching. Anthropic itself warns this could push subscribers toward their caps "sooner than with chat alone."
What it can actually do
Claude Duet pulls from the same memories, skills and connectors as the regular Claude app, and it can dig through past conversations so you don't have to repeat context you've already given it. A dedicated Tasks tab tracks what the agent is working on and flags anything that needs a decision from you. Anthropic is explicit that only the user, not the AI, can mark a task as finished.
The agent plugs into Stripe to flag unpaid invoices and draft reminders, and it can use Link and Shop Pay to complete purchases, though only with your approval first. Anthropic says Duet is built to check in before it sends a message, signs a document, makes a payment or deletes anything, which is the company drawing a line around how much autonomy it's handing over at launch.
Under the hood, Claude Duet runs on Claude Opus 5.5, and it's available as standalone apps for iOS and Android as well as through the web at claude.ai, for anyone who gets an invite off the waitlist.
Where this fits
The move puts Anthropic in a race that OpenAI and Meta are already running. OpenAI has been building out proactive features under the ChatGPT umbrella, and Meta has been testing its own agent-style tool called Muse. All three companies are betting that the next phase of chatbots isn't answering questions but quietly handling the small administrative tasks people put off, like chasing a late invoice or clearing a calendar conflict.
For now, Claude Duet is locked out of Team and Enterprise plans, though Anthropic says that's coming. The bigger test will be whether a research preview this hands-off, checking in before every payment or signature, can actually save subscribers time without burning through their usage limits before lunch. Anthropic hasn't said when the waitlist will clear or when the feature might leave preview status.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.