On 21 October, Anthropic opened a research preview of Claude Duet, a new app where Claude works as a proactive agent. It is built for professionals and small business owners who have more to keep track of than they can manage. Here is what it does, who can use it today, and what I make of it.
What Claude Duet is
Claude Duet is a separate app for iOS and Android, also available on claude.ai, that you message the way you would message a person. It works on its own computer in the cloud, using the apps you have connected, so it keeps going after you close the app. It messages you when a task is done or when it needs a decision.
It uses your existing Claude account, so your memory, skills and connected apps are there from the first message. It runs on Claude Opus 5.5, and you cannot pick a different model.
What it can do
- A morning brief: a short summary of what needs your attention across your email, calendar and other connected apps.
- Everyday admin: finding messages you missed and drafting replies, setting up meetings, and checking QuickBooks or Stripe for unpaid invoices and drafting the reminders.
- Bigger jobs: building dashboards, financial or sales forecasts, prototypes and web apps. Finished work comes back as files you can review and edit.
- Business tools: it can use the connectors and plugins in the Claude Marketplace, such as Gusto, HubSpot, Docusign, Canva and Shopify.
Anthropic says it checks your connected accounts every few hours, not constantly, because checking nonstop would use up your plan's limits much faster. Something urgent can arrive between checks.
Who can use it today
Access is by waitlist, and only for Pro and Max subscribers in the US. Team and Enterprise plans are not included yet. During the preview it is included in those plans at no extra cost, but its background work counts toward the same usage limits as the rest of Claude, so you may reach your limits sooner. If you are outside the US, you will have to wait.
How it handles control and safety
This is the part I would read most carefully before connecting anything, because an agent that reads your email and sends messages is trusted with a lot.
- Before it sends a message, makes a payment, signs a document or deletes anything, it shows you the action and waits for your approval.
- Some actions are off limits even with your approval, such as trading securities or moving money between accounts.
- It never sees your passwords, cookies or sign-in tokens. They pass through separate machines the model cannot access.
- A separate classifier checks each message before it goes out. It is designed to refuse to share the most sensitive information, such as card numbers and government ID numbers.
Anthropic is also open about the limits. It can misread a thread or misjudge what is urgent, it is weaker at personal errands such as booking tickets, and you cannot watch its browser as it works.
What I make of it
For the daily admin of running a business, the morning brief, chasing invoices and drafting replies, an agent that does the first pass and asks before it acts is a sensible idea. Those are the jobs that eat an owner's week, and having a draft ready for approval is a real saving.
Anthropic also lists building web apps and prototypes among its tasks. That is where I would be careful. A first version that works in a demo is quick to get. What decides whether a website or app holds up with real customers is different: security, performance, how it behaves when something fails, and who looks after it six months later. If you have built something with an AI tool and want it checked or finished properly, that is the work I do.
It is early, it is US only, and Anthropic calls it a research preview. I would treat it as worth trying, not as something to rebuild how you work around yet.
What I would do this week
- If you are in the US on Pro or Max, join the waitlist.
- When you get in, start with low-stakes tasks: the morning brief and invoice reminders.
- Connect only the apps a task needs, and keep the approval step on for anything that sends, pays or signs.
- Check what each connected app can see before you add the next one.
Building something with AI and want it solid?
If you have a website or app that was built with AI tools, or you are planning one, get in touch and I will tell you what I would fix first. I also cover this in AI MVP cleanup.
