Claude Duet: Anthropic's Proactive Agent App on Opus 5.5

Anthropic on October 21, 2026 opened a research preview of Claude Duet, a standalone app in which Claude acts as what the company calls a "proactive agent." Built around Claude Opus 5.5, Duet runs on iOS, Android and claude.ai. It is rolling out by waitlist to US subscribers on Anthropic's Pro and Max plans, according to the company's announcement, and draws on the same usage limits as the rest of Claude.

The launch puts Anthropic directly against OpenAI and Meta, both racing to build agents that act on users' behalf, Reuters reported. It is also a break in habit: Anthropic usually folds new products into its main app, and Duet gets its own. The real test is whether small businesses will hand an AI their inbox, books and payment credentials.

Inside Duet's approval gate

Duet sends a short morning brief on the most important items across connected apps, then messages users during the day only when something needs attention. It checks accounts "every few hours" rather than continuously, which Anthropic says is because nonstop polling would drain plan limits much faster. Those background checks count against usage, and heavier jobs such as building a financial model or a deck consume more model turns. The agent works inside a virtual machine with its own browser, so a user's laptop or phone need not stay on.

The company pitches Duet at "professionals and small business owners with more on their plates than they can track." It asks for approval before it sends a message, makes a payment, signs a document or deletes anything. Some actions, including trading securities and moving money between accounts, are blocked even with approval. A dedicated classifier screens every message and is designed to refuse sharing credit card numbers or government ID. Anthropic says Duet also reuses its Managed Agents credential design, so it never directly sees passwords or tokens, and that an outside firm ran penetration tests. Fuller safety details are promised "in the coming days."

That is a long list of safeguards, but a self-reported one. No independent security researcher has yet publicly assessed an agent that, by design, reads email and can pay through Stripe's Link wallet and Shopify's Shop Pay. Anthropic itself concedes Duet "still might misread a thread or miscalibrate what's urgent," that personal errands like ticket booking are weaker, and that users can't yet watch its browser.

The 6,000-connector pitch against Dots and Muse

Duet can tap "over 6,000 connectors and plugins" in the Claude Marketplace, including Gmail, Slack, QuickBooks, Gusto, Docusign, HubSpot and Shopify. It is not yet offered on Team or Enterprise plans, the product page notes.

Rivals got there first. OpenAI launched always-on "dots" agents in September that pursue goals across apps such as Slack and Microsoft Teams, per The Next Web. Meta's Muse "drew millions of downloads and fueled investor interest," Reuters said, and TechCrunch notes Meta has added a small-business version, alongside startups such as Hark, Instinct and Fo. TechCrunch called the category "flooded," "with new ones launching every week." Anthropic's differentiation is largely restraint: work tasks over errands, and hard limits on money movement.

Early reception is mixed. Dan Shipper, chief executive of Every, who had early access, called Duet "the first personal agent I'd call a pro tool," adding that his OpenAI Dot "has become less proactive, and I've noticed it start to lose tasks." He also listed gaps: "it can see the Claude sessions on my computers, but it can't direct them; its email drafts are worse than Dot's; and its scheduled runs can't reach my custom connectors." Every says Anthropic had no input on the review.

Timing tracks the model. Duet arrives a month after Opus 5.5's September 22 release, which Anthropic says costs 40% less to run than Opus 5. Yet on Anthropic's own AutomationBench business-workflow test, OpenAI's GPT-6 Astra edged ahead, 41.4% to 40.0%, and the company admits benchmark margins "have become a less reliable guide to real-world differences."

The next markers are concrete: the safety disclosures Anthropic has promised within days, how quickly the waitlist clears, and whether Duet reaches Team and Enterprise customers. If background checks eat heavily into Pro allowances, cost rather than capability may decide whether small-business owners keep Duet running.