Claude Duet: What It Can See, What It Can't, and Alternatives for Work (2026)

TL;DR

• Claude Duet is Anthropic's proactive agent, opened as a research preview on 21 October 2026. It watches the apps you connect, sends a morning brief, and does larger jobs on its own cloud computer while your laptop is closed.
• It is good at personal follow-through: chasing unpaid invoices, drafting replies, scheduling and preparing documents. It asks before it sends a message, makes a payment, signs a document or deletes anything.
• For company use it is early. Access is by waitlist for US Pro and Max subscribers, Team and Enterprise plans are not included yet, and it sees only what one person connects.
• Sentra is the company memory Claude Duet and every other agent can check with, joining ERP, CRM and HRIS with what people said in email, chat and meetings, with a named person approving before anything goes out.

What is Claude Duet?

Claude Duet is a proactive AI agent from Anthropic that monitors the apps you connect, tells you what needs your attention, and carries out tasks in the background on its own cloud computer. Anthropic opened it as a research preview on 21 October 2026, by waitlist, starting with US Pro and Max subscribers (claude.com).

It lives in its own app on iOS and Android, separate from the main Claude app, and is also reachable on the web at claude.ai (claude.com). You message it a task, it works with its own browser and your connected apps, and it notifies you when the job is done or needs a decision. Anthropic says it runs on Claude Opus 5.5 and decides how much effort a job needs, and you cannot choose the model (support.claude.com).

The idea is a step past chat. A chat assistant waits for a question. Claude Duet checks your accounts at set intervals, sends a short morning brief, drafts next steps and stays quiet until something needs you (claude.com).

Claude Duet at a glance

Sources: claude.com, support.claude.com, claude.com safety page.

For context on what the underlying plans cost, Claude Pro is $20 a month or $17 a month billed annually, and Max starts at $100 a month (claude.com). Anthropic's launch post says Claude Duet can reach more than 6,000 connectors and plugins in the Claude Marketplace, and names Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva and Shopify (claude.com).

What Claude Duet does well

Anthropic's examples are concrete and lean toward the owner of a small business: finding unpaid invoices and drafting reminders, preparing payroll for review, and getting contracts ready to send (claude.com). The design choices around those jobs are careful.

• It keeps working when you leave. Tasks run on a virtual machine with its own browser, so your phone or laptop does not need to stay on (claude.com).
• Approvals are specific. Before a sensitive message, payment or deletion, it shows the action, the destination, the content and every recipient, and a separate automated check built on Claude Code's auto mode classifier runs outside the model (claude.com).
• Purchases are fenced. Each purchase uses a one-time card from a wallet such as Link from Stripe, limited to that merchant and amount (claude.com).
• Credentials stay out of the model. Anthropic says Claude Duet never sees your passwords, browser cookies or sign-in tokens for connected apps (claude.com).
• It carries your Claude context. It shares memory, skills and connectors from your Claude account (support.claude.com).

For one person who already lives in Claude and wants admin work handled, that is a strong first release.

What Claude Duet can see, and what it can't

The limits below come from Anthropic's own documentation. They matter most when a company, not one person, is the user.

It sees what one person connects. You start each connection yourself and can disconnect at any time (support.claude.com). The agent's picture of the business is the picture visible from your accounts. A decision a colleague made in a meeting you missed, or a commitment sitting in another team's channel, is outside its view.

Its memory is your memory. Claude Duet uses the memory from your Claude account and keeps its own history until you delete or reset it (claude.com). That is personal context. It does not start with the company's history of who decided what, or which of two conflicting numbers is the current one.

It checks on a schedule. It looks at your accounts at set intervals, so something urgent can arrive between checks, and Anthropic warns it may misread a thread or misjudge urgency (support.claude.com, claude.com).

Some tools are not there yet. Microsoft 365 is not available yet and Slack works with limits (support.claude.com). For companies on Outlook and Teams, that narrows what it can watch today.

It cannot see your computer. It cannot see all your projects or files on your computer, and you cannot watch or take over its browser (support.claude.com).

It connects apps, not the business. Connectors such as QuickBooks and HubSpot give it records from those systems. Anthropic does not describe Claude Duet joining a system of record with the conversations around it, for example matching a shipment in the ERP with the supplier email that delayed it. Each app is a separate source the agent reads when a task needs it.

None of this is a flaw in a research preview aimed at Pro and Max users. It is the difference between an agent that knows you and a memory that knows the company. Our guide to personal AI agents at work covers why that gap grows with headcount.

Claude Duet for business: where it stands today

Claude Duet for business is not yet a product you can buy for a team. Anthropic's help center says it is not available on Team or Enterprise plans yet and that it plans to bring it to those plans later (support.claude.com). In practice that means:

• No admin console, seat management or workspace policy for Claude Duet today. Each person joins the waitlist with a personal US account.
• The data handling page covers approvals, credentials, memory and deletion. It does not describe enterprise retention terms or admin controls (claude.com).
• Usage draws on individual plan limits, including background checks, with one full limit reset for admitted users available until 30 November 2026 (support.claude.com).

Founders and small teams can try it now on personal plans. A company that needs central control should pilot it with a few people and wait for the Team and Enterprise release before rolling it out. For the plan-level choice between assistants, see ChatGPT vs Copilot vs Claude vs Gemini.

Claude Duet vs ChatGPT dots

OpenAI launched dots at DevDay on 29 September 2026, three weeks before Claude Duet. Both run on their own cloud computer and keep working in the background. They differ most on who can get them and where they live.

Sources: support.claude.com, learn.chatgpt.com.

If your company needs an admin-enabled business option today, dots is further along. If your people already use Claude Pro or Max and want a personal agent with tight purchase controls, Claude Duet is the more careful design. Our ChatGPT dots page covers OpenAI's agent in detail.

Works with Sentra

Claude Duet knows the person it works for. Sentra knows the company. The two fit together.

Sentra reads what is happening across the company, from email, chat, meetings and documents to ERP, CRM, HRIS and BI, and works out what is true now, who owns what, and what is going wrong. It knows when a fact became true and when it stopped being true, so an agent asking about a price, a deadline or an owner gets the current answer rather than last quarter's. Answers respect each source's existing permissions, checked at query time, and a failed check counts as no access.

Any agent can use that memory over REST and MCP. Claude supports custom connectors through remote MCP on Pro and Max plans (support.claude.com), and Claude Duet uses the apps connected in your Claude settings. Anthropic's Claude Duet documentation does not yet say whether custom connectors carry over, so plan on connecting Sentra wherever Claude Duet supports MCP or custom connectors, and through Claude itself in the meantime.

Sentra also acts on its own. The Sentra Agent catches the promises you make in calls and threads, texts you before they are due on iMessage, WhatsApp or Slack, and does the work in a browser with your logins once you say yes. Anything Sentra raises waits for a named person to approve it before it reaches anyone else, and the company sets the guardrails.

Sentra is SOC 2 Type II and ISO 27001 certified, connects 200+ integrations, and deploys in the cloud, in an isolated VPC, or air-gapped on premises. It replaces none of your tools.

Book a demo at /demo to see Sentra working with the agents your team already uses.

Claude Duet alternatives

These are the agents buyers compare Claude Duet with for work, as of October 2026.

Sources: learn.chatgpt.com, town.com, sauna.ai, about.fb.com.

For a wider view of the category, see AI coworker platforms and Sentra vs Claude Tag, Anthropic's teammate inside Slack.

How to choose

1. Decide whose work it is. If the job is one person's inbox, calendar and errands, a personal agent such as Claude Duet fits. If the job is catching what slips between teams, you need something that sees across the company.

2. Check plan availability against your rollout. Claude Duet is Pro and Max only for now. If you need central admin and seat control this quarter, that rules it out for a company rollout.

3. List the systems that hold the truth. Write down where your numbers live: ERP, CRM, HRIS, BI. Ask each vendor how its agent joins those records with the conversations that explain them.

4. Test approvals on a real task. Give each agent the same job, such as chasing an overdue invoice, and see what it shows you before it sends.

5. Plan for more than one agent. Most companies will run several. Put one shared company memory underneath them so they give the same answer to the same question.

FAQ