Meta set the pace first. Its consumer-oriented Muse launched Sept. 8 and quickly amassed millions of downloads; by Oct. 21, Mark Zuckerberg said more than 10 million people in the US and Canada had created a Muse agent, with roughly a quarter using it on a typical day.¹

OpenAI followed on Sept. 29 with Dots, another assistant designed to keep working while users are away. That sequence raised the stakes for Anthropic: without a comparable personal agent, the company risked surrendering the daily interactions that make an assistant indispensable, even among users who prefer Claude itself.¹

On Wednesday, Anthropic answered with a research preview of Claude Duet, available by waitlist to US Pro and Max subscribers. The company’s pitch is deliberately narrower than its rivals’: Duet is for professionals and small-business owners handling invoices, payroll, follow-ups and other administrative drag—not primarily for booking dinner or travel.²

Duet runs on Opus 5.5, draws on users’ existing Claude memories and connected apps, and can send a morning brief, monitor accounts and tackle work ranging from meeting scheduling to forecasts and dashboards. Addy Osmani described it as a proactive agent that works across “6,000 connectors/plugins” and “keeps going when you step away.”Addy Osmani (addyosmani)³

Anthropic’s differentiator is meant to be restraint as much as initiative. The agent asks for approval before major actions such as sending a payment or deleting information, while a dedicated classifier is designed to block requests involving the strictest sensitive data, including credit-card numbers.⁴

Yet the company is not overselling the preview. It says Duet may “misread a thread or miscalibrate what’s urgent,” and its background checks count against ordinary Claude usage limits.⁴ That leaves Anthropic trying to turn a cautious, work-first assistant into a daily habit—while rivals already have a head start.