Claude Duet: What the New Daily-Briefing Agent Does
Disclosure: this article was drafted with Claude, which is made by Anthropic, the company behind Claude Duet. I have tried to stick to what Anthropic and independent outlets actually report, and to flag every place where the details are thin or the sources disagree.
Most AI assistants wait. You open the app, type a question, get an answer, and close the app. Claude Duet, which Anthropic opened as a research preview this week, is built around the opposite idea: it checks your accounts on its own, sends you a short briefing each morning, works on tasks while your devices are off, and pings you when something needs a decision.
That makes it more useful than a chat window for busy people, and also riskier, because it can touch your email, your calendar, and in some cases your money. This guide explains what Duet does, who can use it, what it will not do, and what to think about before you join the waitlist.
What Claude Duet Is
Anthropic describes Claude Duet on its product page as a personal AI agent that you message. It is labeled a research preview. According to the page, it handles inbox, calendar, and follow-up work: it sends a morning briefing, drafts next steps, and takes on larger jobs such as scheduling meetings or building dashboards. It reports back when work is done or when it needs your decision, and it says it is “powered by Opus 5.5.”
Think of it as an assistant that works between your messages rather than only inside them.
Who Can Use It Right Now
Availability is narrow, and this is the first thing to check:
- Country: the United States only, according to Anthropic’s page.
- Plans: Claude Pro and Max subscribers. Team and Enterprise plans are not included yet.
- Access: through a waitlist. One report says invitations go out in waves with no guaranteed date.
- Where it runs: as a standalone app on iOS and Android, and on the web at claude.ai, rather than as a mode inside the main Claude app.
Headlines this week said Anthropic “opened” Duet to Pro and Max users. The body of at least one report is more precise: it opened a waitlist. Do not expect to switch it on today just because you pay for a plan. One outlet also reports that users in India cannot access the current preview even on Max, so non-US readers should assume they are outside the preview.
What Duet Does
A morning briefing
Each day starts with a short summary of the important items across the apps you have connected. Independent coverage adds that Duet checks those accounts every few hours and messages you during the day when it judges that something needs attention. Digital Applied reports that Anthropic chose “every few hours” because checking nonstop would use up plan limits much faster. That is the author’s relay of Anthropic’s explanation, so treat it as reported rather than confirmed.
Background work on its own computer
You assign a task by typing or speaking, in one ongoing thread. Reports say Duet runs the work on a hosted virtual machine with its own browser, so your phone or laptop does not have to stay on. It notifies you when the work is finished or needs a decision.
Examples given by the outlets covering the launch include finding missed Gmail and Slack messages, drafting invoice reminders from QuickBooks, preparing payroll, and building forecasts, presentations, dashboards, and web apps.
Memory and skills carry over
Duet uses the memory, skills, and connected apps from your existing Claude account. If you have already taught Claude how you like things done, that setup should follow you.
Connectors
Anthropic says Duet can use more than 6,000 connectors and plugins from the Claude Marketplace. The official page names Gmail, Google Calendar, Slack, Stripe, and QuickBooks as examples. Other outlets add apps such as Shopify and Gusto, and one notes that the launch post does not say which connectors are available at the start of the preview. In practice, expect the first version to cover popular business apps and some gaps. One report says Microsoft 365 is not available yet, and that Duet cannot reach files stored on your own computer.
What It Asks Before It Acts
This is the most important design choice. Anthropic’s page says Duet asks for approval before sensitive actions such as sending messages or making payments. Independent reports spell out the rule more fully: Duet can read and prepare work on its own, but it must show you exactly what it is about to do and wait for approval before it:
- sends a message,
- makes a payment,
- signs a document, or
- deletes anything.
Two things are off-limits even with approval: trading securities, and moving money between accounts. These are reported as examples of its restrictions, not as a complete list.
For purchases, reports say Duet can pay through Stripe’s Link wallet using a virtual card, or through Shopify’s Shop Pay, with approval steps in between.
Privacy and Safety, as Anthropic Describes It
The details below come from coverage that relays Anthropic’s launch post. I could not independently test any of these claims.
- Credentials: Duet is said to follow the design used for Claude Managed Agents, so it never directly sees your passwords or sign-in tokens. Sign-in through the 1Password password manager is described as coming later.
- Sensitive data: a dedicated classifier checks outgoing messages and is designed to block sharing the strictest data, such as card numbers and government ID numbers. Health and personal finance information is shared only when you explicitly instruct it.
- Testing: reports say the system was red-teamed, tested by an external security firm, and covered by a public bug bounty.
These are design claims. They are reassuring, but a research preview is exactly the stage where surprises show up, so connect only what you are comfortable with.
What It Costs and How Limits Work
Anthropic’s page lists no separate price. Duet uses your plan’s existing usage limits, shared with the rest of Claude. Morning briefs, background checks, and larger projects all count toward those limits. No outlet I read gives numbers for how much a check or task uses.
If you already run into limits on Pro, adding an agent that works all day may make that more noticeable. This is an inference from how shared limits work, not something any source states.
Known Limitations
Anthropic itself lists several, and they are worth taking seriously:
- Duet may misread a thread or misjudge how urgent something is.
- It can miss something that arrives between check-ins.
- Personal errands, such as booking tickets, are less developed.
- Reports say you cannot yet watch or take over its browser. Anthropic’s page says you can see what it is doing and steer it, so the likely difference is between seeing activity and watching a live screen. Check which one applies when you get access.
A Day With Duet (An Illustration)
To make this concrete, here is an imagined example, not a recorded test. A freelance designer connects Gmail, Calendar, and Slack.
- 7:30 a.m.: Duet sends a brief: two client emails need replies, a meeting moved, and an invoice is overdue.
- 10:00 a.m.: It messages to say a client has asked for a new date, and shows a drafted reply for approval.
- Midday: The designer asks it by voice to prepare a simple dashboard of this month’s invoices. It works in the background and notifies her when it is done.
- Afternoon: It drafts an invoice reminder and waits. Nothing is sent until she approves.
The value here is not that any single step is hard. It is that the small steps happen without her opening five apps.
How Duet Differs From a Normal Chat
| Regular Claude chat | Claude Duet | |
|---|---|---|
| Starts when | You send a message | You message it, and it also checks in on its own |
| Where it runs | The chat window | A hosted virtual machine with a browser (reported) |
| Needs your device on | Yes, to chat | Not for background work |
| Acts on your apps | When you ask | Prepares work, then asks before sending, paying, signing, or deleting |
| Availability | Broad | US Pro and Max waitlist only |
For a comparison with another vendor’s approach to agents, see our article on ChatGPT Pro’s parallel agents.
Should You Join the Waitlist?
Ask yourself four questions:
- Do you live in the US and pay for Pro or Max? If not, you cannot use the current preview.
- Is your work inbox-and-calendar heavy? Duet’s reported strengths are email, scheduling, forecasts, and routine business tasks.
- Are you comfortable connecting your accounts? Start with read-focused connections and add more once you trust how it behaves.
- Can you review approvals promptly? An agent that waits for your approval is only fast if you answer.
If you join, a cautious approach is to begin with the morning brief and low-stakes drafting, then expand slowly. If your chats contain private material, our guide to ChatGPT’s Locked Chats shows how one competitor approaches privacy, which is a useful reference point when you compare.
Questions to Ask Before You Connect an Account
Whatever agent you use, a few questions help you decide what to hand over:
- What is the worst thing that could happen if it misreads this account? An email mix-up is annoying. A wrong payment is worse. Start where the cost of a mistake is low.
- Can I undo it? Drafting is easy to reverse. Sending, paying, signing, and deleting are not, which is why those are the actions Duet is reported to pause on.
- Would I be comfortable if a colleague saw these messages? If not, think twice about connecting the account.
- Do I know how to disconnect it? Before you add an app, find where you can remove access later.
Common Mistakes to Avoid
Treating the waitlist as instant access. Check whether you have an invitation.
Connecting everything on day one. Add apps gradually.
Approving without reading. The approval step only protects you if you read what is about to be sent or paid.
Assuming it never misses anything. Anthropic says it can misjudge urgency and miss items between checks.
Ignoring usage limits. Background work counts against your plan.
Expecting full coverage of your tools. Not every connector is available, and Microsoft 365 is reported as missing for now.
Frequently Asked Questions
What is Claude Duet? A research-preview personal agent from Anthropic that sends a morning briefing, works on tasks in the background, and asks for approval before sensitive actions.
Who can use it? US Claude Pro and Max subscribers, through a waitlist. Team and Enterprise plans are not included yet.
How much does it cost? No separate price is listed. It shares your plan’s usage limits.
Can it spend my money? It can prepare purchases, and reports say it uses Stripe Link or Shop Pay with approval steps. It must wait for your approval before making payments, and it refuses to trade securities or move money between accounts.
Does it see my passwords? Anthropic says it does not directly see passwords or sign-in tokens, according to coverage of the launch.
Which model does it use? Anthropic’s page says it is powered by Opus 5.5.
The Bottom Line
Claude Duet is a move from “ask and answer” to “check in and act,” with an approval step in the middle. The design looks careful: it prepares work, shows you what it will do, and waits before sending, paying, signing, or deleting. But it is a research preview, limited to US Pro and Max users on a waitlist, with known weak spots around urgency and personal errands. If you qualify, try it with low-stakes tasks first. If you do not, watch how it develops before deciding whether an always-on agent belongs in your workflow.
