Claude Duet: Anthropic bets that judgment, not reach, will decide the personal-agent race

Anthropic announced Claude Duet on Wednesday. Business Insider reports that the company calls it a "proactive agent": an app that runs in the background, surfaces tasks to the user and handles a range of work on their behalf. Duet is built on Opus 5.5, Anthropic's flagship model, released in late September. Business Insider describes Opus 5.5 as already one of the industry's most popular models. For now Duet is a research preview with a waitlist, so it won't compete right away for the mass market that Meta's Muse and OpenAI's Dots are chasing.

The design follows the template that is emerging across the industry. According to Anthropic, Duet sends users a brief each morning about what is happening in their connected apps. It messages them when something needs attention. It runs outside the user's computer and asks for approval before certain tasks. Mike Krieger, co-lead of Anthropic's Labs team, said in the launch video that the agent "can have the discernment and judgment around when to be proactive, when to act on your behalf." Karan Ahuja leads development. His LinkedIn says he ran product management for Meta Superintelligence Labs until July and joined Anthropic in August. On X he wrote that the team built Duet "for professionals and small business owners juggling a lot at work and at home."

The timing was pointed. Anthropic launched Duet on the day of Meta's earnings report and made a point of naming the hire it had poached from Meta. Meta noticed: Alexandr Wang, head of Meta Superintelligence Labs, posted a meme about how much Duet resembles Muse. The scale gap is real. Data firm Apptopia estimates Muse had more than 14 million installs as of Wednesday. Apptopia also found something more nuanced: Claude users who installed Muse went on to use Claude more than before, though they still spent fewer minutes a day on Claude than on Meta's app. All this happens as Anthropic heads toward an IPO that, according to the report, could value it at up to $2 trillion.

**Our thesis:** Duet confirms that the agent war has moved from "who has the smartest model" to "who earns permission to act." Meta has distribution, with millions of installs. Anthropic has the model and a reputation for caution. Krieger's own framing gives this away: he sells "discernment," not raw capability. For an always-on agent, the product is not intelligence. It is knowing when to interrupt you and when to stay quiet.

**Our reading.** We have argued before that whoever controls the "plumbing" wins: distribution, integration, connections to apps. Duet is Anthropic admitting that a great model is not enough without a place in the user's daily routine. The Apptopia figure undercuts the zero-sum story, since using one agent seems to raise use of the other rather than replace it. The short-term risks deserve a plain statement. An agent that reads your connected apps and acts on your behalf builds the most complete record of your working life. That raises the same privacy and confidentiality questions we flagged with chatbots as confidants. Approval gates on certain tasks are the right instinct, but a research preview proves nothing yet about how they hold up at scale. Today Duet is a promise, not a demonstrated capability.

The long-term direction is still worth taking seriously. The target user here is not a tech enthusiast. It is the small-business owner who spends hours on inboxes, follow-ups and admin. That is the back-office work we have identified as most exposed to AI. When an agent handles it well, the benefit goes beyond efficiency and gives people back time for the parts of their work that require judgment, relationships and passion. The transition will be uneven, and trust will have to be earned one approval at a time. But if these agents work, and if they are governed with care, the result looks less like replacement and more like having a capable assistant, something until now only large companies could afford.