Back to Blog
    AI & AutomationOct 27, 2026

    Claude Duet: The Proactive AI Agent, Explained for Business

    Amir

    Amir Farid

    7 min read

    Claude Duet: The Proactive AI Agent, Explained for Business

    Claude Duet is Anthropic's new proactive AI agent: instead of waiting for a prompt, it comes to you. Announced as a research preview on 21 October 2026, it sends a morning brief from your connected apps, checks them every few hours, and works in the background on its own cloud computer — even when your device is off. It drafts replies, documents, dashboards and forecasts for approval, and asks before sending, paying, signing or deleting.

    For a business, the shift matters more than the product. Until now, AI has been reactive: you ask, it answers. Duet flips that — an agent that watches, notices and prepares. The catch: a US-only research preview for Pro and Max subscribers, on a waitlist, with Team and Enterprise not yet included. So the real question is what to build now, so proactive working is already normal in your business when tools like this reach you.

    Claude Duet: The Proactive AI Agent, Explained for Business

    What Claude Duet actually is

    Claude Duet is a standalone app on iOS, Android and claude.ai, built on Claude Opus 5.5 — not a chatbot with a new coat of paint, but an agent designed around initiative. Connect your apps and Duet checks them every few hours, carrying over memory, skills and connectors from your Claude account.

    A morning brief, then background work

    The centrepiece is the morning brief: a summary across your connected apps of what needs attention and what Duet has prepared. Behind it sits Duet's own cloud computer — a virtual machine with a browser — that works while your phone is switched off. Drafts arrive ready for approval, not as blank pages waiting for instructions.

    6,000+ connectors, but only for some subscribers

    Duet connects to more than 6,000 apps, including Gmail, Slack, HubSpot, Stripe, QuickBooks and Shopify. That breadth is what makes a proactive agent plausible: it can only notice what it can see. But access is narrow — a waitlist of US Pro and Max subscribers only, Team and Enterprise not yet included, no separate price beyond the plan, though background work counts toward usage limits.

    Reactive vs proactive: why the difference matters

    Proactive agent illustration: a morning briefing dashboard surfacing tasks for a business owner

    Every AI tool most businesses use today is reactive. You open it, you type, it responds. The work of noticing — a quiet client, an overdue invoice, a lead needing follow-up — still falls on a person. A proactive agent takes on the noticing, a genuine change in where human time goes.

    The honest comparison

    • Reactive agents answer well but never start anything; they save typing time, not thinking time.
    • Proactive agents start work, surface problems and prepare answers — but they can also start the wrong work, which is why Duet's approval step matters.
    • The value is not more output; it is fewer things slipping through the cracks.

    What Duet will — and will not — do for you

    Duet's guardrails are unusually clear, and the most useful part of the announcement for anyone planning their own agent. It asks for approval before sending a message, making a payment, signing a document or deleting anything. Even with approval, it cannot trade securities or transfer funds between accounts — hard limits, not settings.

    Why hard limits are a feature, not a flaw

    A proactive agent with no limits would be a liability, not a labour saver. The businesses that get value fastest draw these lines early: what it may draft, what it may send, what it may never touch. Duet's design is a template worth copying — approval gates on irreversible actions, some actions off the table. Any AI automation project should start with the same list. Anthropic concedes Duet may misread a thread or misjudge urgency — so approval must stay a real review, not a rubber stamp.

    What to delegate, and what to keep human

    The practical question is not "what can the agent do?" but "what should it do without me?" Rule of thumb: delegate preparation, keep decisions.

    Safe to delegate

    • Morning summaries across email, chat and CRM — the "what needs me today" brief.
    • Draft replies to routine enquiries, drawn from your past answers.
    • Dashboards, forecasts and reports assembled from systems you already trust.
    • Follow-up reminders: quiet clients, overdue invoices, unanswered quotes.

    Keep human

    • Anything that sends money, signs a commitment or deletes records — approval at minimum, and some actions never automated at all.
    • Delicate conversations: complaints, negotiations, apologies.
    • Judgement calls where the context is not in any app — the phone call, the relationship, the history no system recorded.

    How to build your own proactive agent now

    Waiting for a waitlist is the slowest option. The pieces Duet combines — scheduled checks, connected apps, draft preparation, approval gates — are all buildable today.

    Pick one workflow, not ten

    Choose one recurring workflow where things slip: lead follow-up, invoice chasing, or daily reporting. One workflow, done well, teaches you more than five half-built ones — the daily numbers brief from your restaurant ERP, say, or the quiet-lead alert from your CRM system.

    Connect the systems it needs to see

    A proactive agent is only as good as its view. Connect the inbox, the chat tool and the system of record — the trio Duet leans on with Gmail, Slack and HubSpot. Businesses already using WhatsApp bots or a chatbot have a head start: those channels produce structured conversations an agent can watch. The same applies to voice: an AI calling agent prepares, a human approves. See also our guides to AI agents for small business automation and chatbots versus AI agents.

    Build the brief, then the drafts

    Start with the morning brief alone — nothing sent automatically. Once the brief is accurate for a few weeks, add draft replies for routine items, each needing one-tap approval. Only after that earns trust should anything send without review, and only for the lowest-risk categories — observability first, autonomy later, as in serious SaaS development. If the agent must answer from your own documents, use retrieval — see our RAG chatbots guide.

    What to do this week

    Proactive agents are no longer a curiosity; they are a research preview with a waitlist, so general availability follows. The businesses that benefit first already know which workflow to automate and where the approval line sits.

    • Join the Duet waitlist if you are a US Pro or Max subscriber — and treat it as research, not a rollout plan.
    • Pick your one workflow and write its approval rules this week.
    • Build the morning brief first, using the tools and connectors you already have.

    Budget for running costs, not just the build — as a rough estimate, a small scheduled agent costs in the low hundreds of dollars a month; see the Gemini 4 Argon pricing deadline for how fast introductory pricing can double. If you want help building that first proactive workflow — the brief, the connectors, the approval gates — talk to a team that builds AI agents for production, not demos. Good AI automation starts with the workflow and the guardrails, then the model.

    Frequently asked questions

    What is Claude Duet?

    Claude Duet is Anthropic's proactive AI agent, announced as a research preview on 21 October 2026. Built on Claude Opus 5.5, it sends a morning brief across your connected apps, checks them every few hours, and works in the background on its own cloud computer, drafting replies, documents, dashboards and forecasts for approval. Coverage: 9to5Mac and Unite.AI.

    Who can use Claude Duet today?

    Only a waitlist of US Pro and Max subscribers for now; Team and Enterprise accounts are not yet included. A standalone app on iOS, Android and claude.ai, with no separate price beyond the plan, though background work counts toward usage limits.

    Can Claude Duet send messages or make payments on its own?

    It asks for approval before sending a message, making a payment, signing a document or deleting anything. Even with approval, it cannot trade securities or transfer funds between accounts. Treat approval as a real review: Anthropic concedes Duet may misread a thread or misjudge urgency.

    Should my business wait for Claude Duet or build its own proactive agent?

    Build now. The pieces — scheduled checks, app connectors, draft preparation and approval gates — are all available today. Start with a morning brief, add drafts with approval, automate sending only for the lowest-risk items.

    #Claude Duet#Proactive AI Agents#AI Automation