Claude Duet review: Anthropic's proactive agent, behind a US waitlist

Claude Duet is Anthropic’s proactive agent app. Instead of waiting for you to open a chat, it reviews your email and calendar each morning, sends you a short brief, checks your connected accounts every few hours, and works on tasks from its own cloud computer while your phone is off. It reads, drafts and prepares freely, and stops to ask before it sends, pays, signs or deletes anything (Anthropic).

It’s a research preview, opened on October 21, 2026, with a waitlist for US subscribers on Claude Pro, Max and Ultra. If you’re in, it’s the cheapest way to get an always-on agent from a major AI lab. If you’re not, you’ll have to wait. For a short explainer, read What is Claude Duet?.

Who Claude Duet is for

Anthropic aims Duet at professionals and small business owners who have more to manage than they can keep track of (Anthropic). The connectors in its help center tell the same story: Gmail, Google Calendar, Drive, Slack, QuickBooks, HubSpot, Docusign and Calendly (Claude Help Center).

It fits best if you:

• already pay for Claude and have set up memory, skills and connectors there, since Duet inherits them from the first message;
• want a morning brief and drafted next steps more than an errand runner;
• are in the US and can wait for an invite.

It fits poorly for personal errands. Anthropic itself says Duet is weaker at things like booking tickets than at work tasks, and TechCrunch put it plainly: it “isn’t good at handling errands or booking tickets” (TechCrunch). It also isn’t for teams yet: Team and Enterprise plans aren’t supported.

What it actually does day to day

Duet is one ongoing thread you talk to by text or voice (Anthropic). A normal day looks like this:

• Morning: a short brief on the most important items across your connected apps.
• During the day: checks every few hours, with a notification only when something needs you. You can tell it which topics you want to hear about and how often. Anthropic explains why it doesn’t check constantly: that would burn through your plan’s limits.
• On request: small chores like scheduling, or bigger jobs such as dashboards, sales or financial forecasts, prototypes and web apps. It cites the sources behind numbers and summaries, and only you can mark a task complete (Claude Help Center).
• Buying things: it can check out through Link by Stripe or Shop Pay, and every purchase uses a one-time card number so your real card isn’t shared.

What it won’t do matters as much. Even with your approval, it won’t trade securities, move money between accounts or access other people’s accounts. It can’t call or text other people, can read but not edit your work in Claude, and can’t see your local files (Claude Help Center).

Limits

• Waitlist, US only. You join from the product page with your name, email, plan and phone type. Invites go out in waves with no guaranteed date, and the email should match your Claude account and your phone’s Apple or Google account (Claude Help Center).
• Shared usage. No extra fee during the preview, but Duet draws on the same limits as the rest of your Claude plan, and background work like the morning brief counts. The more you hand it, the faster you reach your plan’s limits.
• One agent, one model. You get a single Duet on Claude Opus 5.5, with no model choice.
• No browser view. You can’t watch or take over its browser yet.
• Gaps in connectors. Microsoft 365 isn’t available yet, and Slack works “with limits.” Slack is a connector, not a place where you talk to Duet.
• It will miss things. Anthropic warns it can misread a thread or miscalibrate urgency, and something urgent can arrive between checks.

Privacy

Anthropic has put more detail into Duet’s safety design than most rivals. Duet never directly sees your passwords or sign-in tokens; site passwords go through a separate on-device form. Its actions pass an auto mode classifier, it has prompt-injection defenses, and Anthropic says it was red-teamed, pen-tested by an outside firm and included in its bug bounty. A new classifier refuses to share the most sensitive data, such as card numbers and government IDs, and health or financial details only go out when you’ve explicitly told it to send them (Anthropic).

What Anthropic doesn’t say, in Duet’s launch article or help center, is whether Duet conversations are used for training. Check your Claude privacy settings rather than assuming either way. Also note that disconnecting an app stops future access, but Duet keeps its memory of what it already saw (Claude Help Center).

How Claude Duet compares

vs dots. OpenAI’s agent does more: Slack and Teams as real channels, calls, a cloud computer you can watch, and Codex for coding. It also needs ChatGPT Pro from $100 a month and excludes the EEA, the UK and Switzerland. Duet costs a fifth of that to get into and has published firmer limits on money and sensitive data.

vs Perplexity Teammates. Teammates lets you run several agents with their own roles and models, and you can take over their desktops. It’s $200 a month on Perplexity Max and web-only. Duet is the single, careful assistant; Teammates is a small team you manage.

vs Lindy. The nearest product in spirit: daily briefs, inbox and calendar work, approvals before outside actions. Lindy is available now with a free plan, works over iMessage and Slack, and covers teams. Duet’s advantage is your existing Claude memory and connectors.

vs Poke. Poke also messages you first, but lives in iMessage and Telegram with a free plan. If you want proactive check-ins in a chat app rather than a separate app, Poke is the easier start.

vs Muse and Instinct. Both are better at the personal errands Duet admits it’s weak at.

vs Rakazo and OpenClaw. Open source, your choice of model, several bots and real chat-app channels; Rakazo can even run on your existing Claude subscription. More setup, more control. (Disclosure: Rakazo is made by the company behind this site, and is ranked on the same terms.)

Verdict

If you’re a US Claude subscriber, join the waitlist now: Duet costs nothing extra, inherits the setup you’ve already done, and is the most cautious of the big-lab agents about money and sensitive data. It’s best as a work chief of staff, not an errand runner.

On our checklist it misses two points: it runs one agent and doesn’t live in your chat apps. If you need either, or you’re outside the US, look at dots, Lindy or the full best AI assistants ranking.