{"id":100987,"date":"2026-10-22T06:08:08","date_gmt":"2026-10-22T10:08:08","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=100987"},"modified":"2026-10-22T06:08:08","modified_gmt":"2026-10-22T10:08:08","slug":"claude-duet-ai-agent-100987","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/claude-duet-ai-agent-100987\/","title":{"rendered":"Claude Duet Enters AI Agent Race Against Grok Bot, Muse, Dots"},"content":{"rendered":"<p>Anthropic has entered the <a href=\"https:\/\/overcentral.com\/en\/ai-agent-skills-gap-96625\/\" title=\"AI Agent Skills: The Gap Between Theory and Practice\" data-iacss-internal=\"1\">AI agent<\/a> race with Claude Duet, a standalone research-preview app built on Claude Opus 5.5 that runs as a proactive agent. The app is available on iOS, Android, and the web at <a href=\"https:\/\/claude.ai\" target=\"_blank\" rel=\"noopener noreferrer\" data-iacss-external=\"1\">claude.ai<\/a> and is aimed at professionals and small business owners who have &#8220;more on their plates than they can track.&#8221; With this launch, Anthropic is answering the same question that has pushed Grok Bot from SpaceXAI, Muse from Meta, and Dots from OpenAI: what happens when an <a href=\"https:\/\/overcentral.com\/en\/ai-assistant-skills-96649\/\" title=\"Stop Trying to Build Your AI Assistant from Scratch\" data-iacss-internal=\"1\">AI assistant<\/a> does not wait to be asked?<\/p>\n<p>Duet is best understood as the opposite of a conventional chatbot. In the standard Claude app, the user opens a conversation and drives it. With Duet, the agent comes to the user. It watches connected accounts, works in a cloud virtual machine, and only breaks the silence when something needs attention. That shift from pull to push is the central design bet of the new wave of personal agents.<\/p>\n<h2>What is Claude Duet and how does it differ from the Claude app?<\/h2>\n<h3>What is Claude Duet?<\/h3>\n<p>Claude Duet is a research-preview standalone app from Anthropic that runs on Claude Opus 5.5 as a proactive agent. The agent monitors connected apps, completes assigned tasks on a remote cloud computer, and asks for approval before taking sensitive actions. It is available on iOS, Android, and the web to Pro and Max subscribers in the United States through a waitlist; Team and Enterprise plans are not yet included.<\/p>\n<p>The main difference from the regular Claude app is the direction of initiative. The Claude app is a pull product: users open it, type a request, and receive a response. Duet is a push product. Every morning it sends a short summary of the most important items from connected apps, and after that it reaches out only when a task or event needs the user&#8217;s attention.<\/p>\n<p>Duet begins with what the user already has. It pulls memory, skills, and app connections from an existing Claude account, so the first run does not start from a blank profile. That also means the agent inherits the user&#8217;s preferred tools and context instead of forcing them to rebuild that environment inside a new product.<\/p>\n<h2>Duet starts with a briefing and takes work in one continuous thread<\/h2>\n<p>Anthropic designed the agent&#8217;s rhythm around attention. Instead of firing notifications all day, Duet delivers one morning digest, then waits. That makes it a quieter assistant than the word &#8220;agent&#8221; often implies. The intent is to give users a sense of what changed overnight without overwhelming them.<\/p>\n<p>Users can assign tasks by text or voice message in a single ongoing thread. Anthropic says the thread is broad enough to cover everything from scheduling to dashboards and financial forecasts to prototypes and <a href=\"https:\/\/overcentral.com\/en\/claude-code-web-apps-playbook-96678\/\" title=\"Building Web Apps with Claude Code: A Senior Developer&amp;apos;s Playbook\" data-iacss-internal=\"1\">web apps<\/a>. In effect, the conversation doubles as a running brief for the agent.<\/p>\n<ul>\n<li>Scheduling and administrative follow-ups<\/li>\n<li>Dashboards and financial forecasts<\/li>\n<li>Prototypes and web applications<\/li>\n<li>Invoice follow-ups and payroll preparation<\/li>\n<\/ul>\n<p>The breadth is possible because Duet connects to services such as Gmail, Slack, Stripe, Docusign, and Shopify through more than 6,000 connectors and plugins in the Claude Marketplace. That gives the agent enough reach to track down unpaid invoices, draft payment reminders, or get payroll ready without moving between several tools.<\/p>\n<h3>How does Claude Duet handle purchases?<\/h3>\n<p>For purchases, Duet uses Stripe Link with a virtual card. The practical effect is that the user&#8217;s real card number never gets shared with a merchant. Combined with an approval step before payment, it is a deliberate answer to one of the trickiest questions in agent commerce: how to let software spend money without exposing credentials or losing control.<\/p>\n<h2>Why Duet needs its own cloud computer<\/h2>\n<p>One detail separates Duet from a phone-based assistant: it is not primarily running on the user&#8217;s device. Anthropic says the agent operates on a virtual machine, its own cloud computer with a browser, so the user&#8217;s laptop or phone does not need to stay on. This is what makes &#8220;keep working while you are away&#8221; more than a marketing phrase.<\/p>\n<p>A local app could pause when the device locks, lose connectivity when the user moves between networks, or drain battery during long tasks. By running in the cloud, Duet can keep polling apps, assembling documents, and preparing work for the next time the user checks in.<\/p>\n<p>The trade-off is that this kind of background activity has a real cost, and Anthropic is upfront about how that cost is handled. Duet checks connected accounts every few hours, and those checks count against the same subscription limits that already apply to other Claude products.<\/p>\n<h2>Claude Duet enters an agent field that already includes Grok Bot, Muse, and Dots<\/h2>\n<p>The launch follows a burst of similar products from Anthropic&#8217;s rivals. SpaceXAI opened the Grok Bot beta in August, and Meta followed with Muse in September. OpenAI also introduced Dots in September. Each company is betting on personal agents that can handle tasks independently and act on their own initiative.<\/p>\n<table>\n<thead>\n<tr>\n<th>Product<\/th>\n<th>Developer<\/th>\n<th>Availability<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Claude Duet<\/td>\n<td>Anthropic<\/td>\n<td>Research preview; waitlist for Pro and Max subscribers in the US<\/td>\n<\/tr>\n<tr>\n<td>Grok Bot<\/td>\n<td>SpaceXAI<\/td>\n<td>Beta opened in August<\/td>\n<\/tr>\n<tr>\n<td>Muse<\/td>\n<td>Meta<\/td>\n<td>Launched in September<\/td>\n<\/tr>\n<tr>\n<td>Dots<\/td>\n<td>OpenAI<\/td>\n<td>Launched in September<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>What stands out about Duet is the depth of its workplace integration. The existing Claude ecosystem, the Claude Marketplace connectors, and the virtual machine give Anthropic a different attack angle from consumer-oriented assistants. Duet is not just a chatbot with tools; it is designed as a back-office worker that happens to communicate through an app.<\/p>\n<h2>Claude Duet can propose, but it cannot send or pay without approval<\/h2>\n<p>Anthropic has set hard boundaries around what the agent may do autonomously. Duet can read, draft, and prepare on its own. Before it sends a message, makes a payment, signs a document, or deletes anything, it shows exactly what it plans to do and waits for the user to approve. Trading securities and moving money between accounts are off the table even with approval.<\/p>\n<p>This makes Duet a supervised agent rather than a fully autonomous one. It can do the legwork, but the final decision on external actions stays with the user. For businesses testing the product, that is likely a deliberate choice: an agent that can spend money or sign contracts is useful, but only if it cannot do both silently.<\/p>\n<h3>What technical safeguards protect Claude Duet?<\/h3>\n<p>Anthropic says the agent never sees passwords or login tokens directly. Credential handling follows the design of Claude Managed Agents. To defend against prompt injection, Duet uses protections drawn from Anthropic&#8217;s research on browser use. Red teams and an outside firm also ran attack tests against the preview.<\/p>\n<p>Anthropic also built a new classifier specifically for Duet. The screening system works with special safety instructions in the system prompt to check every message for signs of sensitive-data leakage. The combination is designed to refuse requests to share credit card numbers or ID details. Health or personal finance information only gets through when the user explicitly says so.<\/p>\n<h2>Anthropic acknowledges what Duet cannot do well yet<\/h2>\n<p>The company is careful not to overpromise. The early version can misread a message thread, misjudge urgency, and miss items that arrive between two checks. It is also weaker at personal errands such as booking tickets, because its core features are built for work rather than consumer life.<\/p>\n<p>The polling cadence reinforces those limits. Duet checks connected accounts only every few hours, so it is not a continuous monitor. Anthropic says that frequency comes down to usage: constant checking would burn through subscription limits much faster. Since Duet shares quotas with other Claude products, users who rely heavily on Claude for coding, writing, or analysis will need to watch their remaining allowance.<\/p>\n<p>There is also a product-placement question. Duet runs separately from the Claude app for now. Anthropic has not said how the two should coexist, and the answer may depend on what early testers say.<\/p>\n<h2>What the research preview will tell Anthropic about agent judgment<\/h2>\n<p>Duet&#8217;s first real test is not whether it can connect to apps or run code. It is whether it can learn when to interrupt. Anthropic plans to use tester feedback to tune when the agent should reach out, which is another way of saying that the early versions are expected to get the balance wrong. The waitlist model gives Anthropic room to correct course before a wider release.<\/p>\n<p>Team and Enterprise plans are not part of the preview, which suggests the agent is not ready for organization-wide deployment. But the direction is clear. Anthropic is treating the agent relationship as a product in its own right, not as a feature inside the chat app.<\/p>\n<p>The next phase of Claude Duet&#8217;s development will depend on whether users trust the agent enough to let it work unsupervised in low-risk areas, and whether the approvals it requests feel like safety checks or friction. If Duet can master that balance, it could become a template for how AI assistants earn a permanent role in daily work. If not, it will be an instructive early example of how hard it is to build proactive agents that people actually want to keep around.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Anthropic has entered the AI agent race with Claude Duet, a standalone research-preview app built on Claude Opus 5.5 that runs as a proactive agent. The app is available on iOS, Android, and the web at claude.ai and is aimed at professionals and small business owners who have &#8220;more on their plates than they can [&hellip;]<\/p>\n","protected":false},"author":7,"featured_media":100991,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/100987.png","fifu_image_alt":"Claude Duet Enters AI Agent Race Against Grok Bot, Muse, Dots","footnotes":""},"categories":[31],"tags":[],"class_list":["post-100987","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/100987.png","fifu_image_alt":"Claude Duet Enters AI Agent Race Against Grok Bot, Muse, Dots","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/100987","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=100987"}],"version-history":[{"count":2,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/100987\/revisions"}],"predecessor-version":[{"id":100990,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/100987\/revisions\/100990"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/100991"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=100987"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=100987"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=100987"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}