Anthropic Launches Claude Duet, an AI Agent That Messages You First

Most AI assistants wait for you to type something. Anthropic's newest one is built to speak first. On Wednesday, October 21, the San Francisco company opened a research preview of Claude Duet, an app that watches your connected work apps, tells you what needs attention and keeps working on tasks after you put your phone down.

Claude Duet is aimed at professionals and small business owners with more on their plates than they can track. It arrives just weeks after Meta and OpenAI launched their own proactive agents, which makes the race for an AI that acts on your behalf one of the busiest contests in Silicon Valley this fall.

What Claude Duet Does

Each morning, Duet sends a short brief on the most important items across the apps you have connected, such as email, chat, calendar and accounting tools. Throughout the day it checks those accounts every few hours and sends a message only when something needs you. You talk to it by typing or by voice in one ongoing thread, so a quick reply and a weeklong project start in the same place.

You can also hand it work. Anthropic says the tasks range from scheduling meetings to building dashboards, creating financial or sales forecasts, prototyping and building web apps. Duet runs on a virtual machine with its own browser, so your laptop or phone does not need to stay on. When a task is finished or needs a decision, it sends a notification.

Duet is built around Claude Opus 5.5, Anthropic's latest model, and it carries over the memory, skills and connected apps from a person's existing Claude account, so there is little to set up.

The Tools It Can Reach

A lot of Duet's usefulness comes from the thousands of connectors and plugins in the Claude Marketplace. In its announcement, Anthropic described Duet finding missed Gmail and Slack messages and helping reply, checking QuickBooks or Stripe for unpaid invoices and drafting reminders, preparing payroll in Gusto, posting an approved job listing to LinkedIn, sending agreements for signature through Docusign, drafting follow-up emails from HubSpot deal history, turning a product launch into a Canva campaign and setting up a Shopify store.

It can also buy things for a business. Duet uses Link, Stripe's digital wallet, which asks the user to approve the spend and then pays with a virtual card so the real card number is not shared. It can check out with Shop Pay at online stores, and sign in through 1Password is listed as coming soon.

Early testers named in the announcement describe an assistant that catches the things people miss. Jess Provencher, a real estate broker, said Duet flagged a property showing that was on her calendar for the wrong day: "[This is] exactly the kind of thing I want caught without having to ask." Amin Ebrahimi, who owns the Shahs of Kabob restaurants in South Florida, uses it for accounting, customer support and marketing. "It brings it all together," he said. "I don't need to open up 100 tabs."

How Anthropic Is Handling Safety

An agent that reads your email and acts for you raises obvious questions, and Anthropic spent much of its announcement on them. Duet follows the company's framework for trustworthy agents, the same guidance behind Claude Code and Claude in Chrome. It never directly sees passwords or sign-in tokens, a classifier approves its actions, and its defenses against prompt injection, hidden instructions planted in web pages or messages, build on Anthropic's browser research. Red teams and an outside firm tested it, and it is covered by a public bug bounty.

Duet reads, drafts and prepares work on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows the user exactly what it plans to do and waits for approval. Some actions, including trading securities and moving money between accounts, are off-limits even with approval. A new classifier blocks requests to share the most sensitive information, such as credit card numbers and government IDs, and health or personal finance details go out only when the user explicitly says so. Anthropic published a further safety, privacy and security overview on October 23.

Who Can Use It and What It Costs

The preview is open by waitlist to U.S. subscribers on the Pro and Max plans, on iOS, Android and the web. There is no separate charge. Duet draws on the same usage limits as the rest of Claude, and its background checks and larger jobs, like building a financial model or a slide deck, use those limits faster than ordinary chats. Pro costs $20 a month, and Max starts at $100 a month, according to Anthropic's pricing page.

Anthropic is also upfront about the limits of this early version. Duet may misread a thread or misjudge what is urgent, personal errands such as booking tickets are weaker than work tasks, it cannot yet show its browser to the user, and it lives in its own app, separate from the main Claude app.

What It Means for Silicon Valley

The three biggest proactive agents now come from companies a short drive apart. Meta introduced Muse on September 8, a personal agent with a free tier that works on its own cloud computer and keeps going after the app closes, and OpenAI followed at the end of September with Dots, agents that keep moving projects forward between conversations. Anthropic is pitching Duet less at shopping and travel and more at running a small business.

The partner list reads like a Bay Area roll call, too. Stripe, Slack, Gusto, Docusign and Intuit, the maker of QuickBooks, all started or are based in the region. If agents like Duet become the way small companies get through invoices, payroll and hiring, the software those companies already pay for becomes something an AI operates, and the fight for the customer moves from the app screen to the agent.

Will Proactive Agents Earn Our Trust?

The possibilities are real. A small business owner who spends evenings chasing invoices, wading through email and fixing calendar errors could win back hours each week. Agents that spot an issue before it becomes a problem, like a client call booked right before a flight, are closer to what people want from a capable human assistant than a chatbot that only fields questions. If the preview proves itself, startups and freelancers could get the kind of back-office support that used to mean putting someone on the payroll, and Silicon Valley's app makers could find new demand through agent connections.

The concerns are just as tangible. Duet needs deep access to email, documents, money and messages, so every connected account becomes a target. The industry has yet to solve prompt injection, and one well-crafted fake message could try to push an agent off course. Then there is judgment. Anthropic itself says Duet can misread urgency, and an agent that interrupts too much, or not enough, is of little use. Cost is a factor too, since background checks draw down the same plan limits subscribers need for everything else. For workers, there is a quieter question: what becomes of bookkeeping and admin jobs when one owner and an agent can do the work of a part-time assistant?

A broader debate is underway over whether this is how people want to work in the first place. Some are happy to let an assistant make the first move. Others are worn down by a steady stream of AI nudges and are not keen to hand more of their judgment to software. Liability is another open issue: if an agent puts an error into a contract or an approved payroll run, who is responsible? Supporters argue that approval gates and a list of banned actions keep the user in command, and that an overloaded owner left to catch everything alone pays a price too.

The signals to watch are specific. How quickly will Anthropic work through the waitlist and expand beyond the United States? Will Duet come to Team and Enterprise plans with the audit controls businesses expect? Will testers see fewer missed tasks without being flooded with alerts? And will Meta and OpenAI answer with deeper business tools of their own?

Agents Need Open Doors

At Sparknify we keep track of how AI agents get along with the rest of the internet. Last week we looked at AI Agents Promised to Do Everything Online. The Web Is Locking Them Out, on banks, airlines and retailers blocking automated assistants. Duet takes a different route by working mostly through official connections that businesses have agreed to, which may matter as much as its model.

The assistant that messages you first is here; now it has to prove it knows when to stay quiet.

Subscribe to Sparknify for Silicon Valley tech news and the people behind it.