Claude Duet: What the Proactive Agent Does and Shows You (2026)
Vorec Team · 2026-10-25 · About 8 min read
Claude Duet is an Anthropic app, in research preview since 21 October 2026, in which Claude works as a proactive agent. It sends a morning brief from your connected apps, messages you when something needs you, and takes on tasks in the background on its own cloud computer and browser. Anthropic says it is designed to show you the action and wait for your approval before it sends a message, makes a payment, signs a document or deletes anything. You can't watch it work in its browser; you follow its work through a task list instead. Access is by waitlist for US Pro and Max subscribers.
This page sums up what Anthropic has published, checked on 25 October 2026: the announcement (21 October 2026), the Get started with Claude Duet help article and the safety, privacy and security article (23 October 2026). Anthropic says features can change during the research preview, so treat the details as a snapshot. Where something is our reading rather than Anthropic's statement, we say so.
Claude Duet at a glance
| Question | What Anthropic says |
|---|---|
| What it is | An app where Claude works as a proactive agent |
| Who can get it | US Pro and Max subscribers, through a waitlist, in waves |
| Team and Enterprise | Not available yet; Anthropic plans to bring it later |
| Where | A standalone app on iOS and Android, and on claude.ai (per the announcement) |
| Model | Claude Opus 5.5; you can't choose a model |
| Price | Included with Pro and Max during the preview; uses your plan's usage limits |
| Where it works | Its own computer in the cloud, with its own browser |
| Can you watch it? | No. You can't watch it work in its browser or take over that browser |
What Claude Duet does
The help article describes four main jobs:
- A morning brief. Each morning it looks across your email and calendar, picks what most needs your attention, and can start first steps, such as drafting replies for you to review.
- Check-ins during the day. It sends updates on its work and follows up on threads that have gone quiet. It checks your connected accounts at set intervals rather than nonstop, which Anthropic says is to avoid using up your plan's limits.
- Larger jobs. You describe what you need and it works out which apps to use. Bigger results come back as files you can review and edit, such as a research brief, financial model, slide deck or dashboard, in a Make tab.
- Websites without a connector. For sites with no connector, it uses its own browser in the cloud.
It works with the apps you've already connected in Claude. The help article lists Gmail, Google Calendar, Google Drive, Slack, QuickBooks, HubSpot, Docusign and Calendly, plus Stripe, and Link and Shop Pay for purchases you approve. It also carries over your Claude memory, skills and connectors, and supports plugins.
What it asks before acting
Anthropic's safety article covers this in detail. Before Claude Duet sends a message, makes a payment, signs a document or deletes anything, the safety article says it shows you the action, the destination and the content, including every recipient, and waits for your approval.
A few more rules from the same pages:
| Rule | Detail |
|---|---|
| Separate intent check | An automated check, built on the auto mode classifier from Claude Code, runs outside the working model and assesses whether each action matches your intent. |
| Never allowed | Trading securities, moving money between accounts and accessing someone else's account stay off-limits even with your approval. |
| No "don't ask again" | Some steps, such as paying or sending money, deleting an order or creating an account, keep asking however often you've approved them. |
| Purchases | Each purchase uses a one-time card that only works for that purchase. |
| Credentials | Claude Duet never sees your passwords, browser cookies or sign-in tokens; they pass through separate machines the model can't access. |
| Sensitive information | A separate classifier checks outgoing messages. It's designed to reject the strictest tier, such as card numbers and government ID, and to let health or personal-finance details through only when you've told it to send them. Anthropic says these safeguards will occasionally miss something or flag a harmless message, which is why they work alongside the approvals. |
| If you're away | It finishes the rest of the task and brings the approval back to you later. |
Anthropic also says it spent time calibrating how often Duet asks, because too many requests teach people to tap through them.
What you can and can't see
The help article is blunt about one limit: you can't watch Claude Duet work in its browser or take over that browser. The browser runs in its own virtual machine in the cloud, and the agent drives it through a fixed set of commands such as open a page, click and type.
What you get instead, according to Anthropic's pages:
- A task list. You can see what Claude Duet has done and is doing, each task has its own page, and you can steer it mid-task. The Tasks tab highlights tasks that need your input, and only you can mark a task as completed.
- Approval cards showing the action, destination, content and recipients before it sends a message, makes a payment, signs a document or deletes anything.
- Sources for numbers. When it gives you a number or a summary, it tells you which email, file or spreadsheet it came from.
- History you control. It keeps your conversation history and memory until you delete or reset them.
Anthropic also lists what Duet can't do yet: it can't see the files on your computer, it can read but not change your work in Claude, it doesn't make phone calls or send texts, and Microsoft 365 isn't available yet. It "can also make mistakes", such as misreading a thread or misjudging what's urgent, which is why Anthropic tells you to review drafts and approval requests before you confirm them.
Why this matters if your work involves screen recordings
Our reading: agents are splitting into two styles of oversight.
- Show the work as it happened. Coding agents such as Cursor's attach screenshots and videos of their runs to a pull request, and some let you watch or take over the desktop. We covered this in our Cursor computer use guide.
- Show the decision before it's final. Claude Duet doesn't show you its browser. It shows you a task page, an approval card and the source of a number.
Both styles are built for review: they help you decide whether to accept what an agent did or is about to do. A tutorial has a different job. It's made on purpose to teach a person the workflow, with pacing and an explanation of each step. If a process an agent handles today has to be learned by a person tomorrow, someone still needs to record it with explanation, which is what a narrated tutorial is for. We looked at the trust side of this in our write-up of OpenAI's report on agents weakening their own monitoring, and at screen-driving agents in general in Computer Use Agents: What the Vendor Docs Say.
Where Vorec fits
Vorec makes the teaching kind of recording. It has its own macOS recorder, and an AI agent can drive it for you through the Claude Code plugin, capturing locally so you can review the take before anything is uploaded. It then drafts narration matched to the workflow it captured and generates the voiceover, so nobody speaks into a microphone. In the editor you can add smooth cursor motion, cursor-follow zoom and click-based auto-zoom, and freeze-sync can hold a frame to give an explanation time to finish. The same capture can also produce a written step-by-step guide, with annotatable screenshots. You can upload an existing recording instead. See how AI agents record screen tutorials. Plans start at $9/month (Starter); a free trial is available (current allowances on the pricing page).
FAQ
What is Claude Duet?
An Anthropic app, in research preview since 21 October 2026, where Claude works as a proactive agent: it sends a morning brief, messages you when something needs you, and works on tasks in the background on its own cloud computer.
Who can use Claude Duet?
US subscribers on the Pro and Max plans, through a waitlist. Invitations go out in waves. It isn't available on Team or Enterprise plans yet.
How much does Claude Duet cost?
During the research preview it's included with Pro and Max at no extra cost, and it uses your plan's usage limits, including background work such as the morning brief.
Can I watch Claude Duet work?
No. Anthropic's help article says you can't watch it work in its browser or take over the browser. You can follow each task on its own page in the task list and steer it mid-task.
Does Claude Duet ask before it acts?
Anthropic says it's designed to, for sending a message, making a payment, signing a document or deleting anything. It shows the action, destination, content and recipients, then waits for approval. Some actions, such as trading securities or moving money between accounts, are off-limits even with approval.
What model does Claude Duet use?
Claude Opus 5.5. You can't choose a different model during the preview.
Want the recording to narrate itself? Record with Vorec, or upload one you already have, and get a narrated tutorial plus a written guide. Start free. Trial includes up to 3 projects; exports carry a watermark.