Claude Duet arrives late to the personal-agent race, but bets on work, not errands

By CNBC · October 21, 2026.

Anthropic on Wednesday announced a research preview of Claude Duet, an AI agent designed for small businesses and professionals. According to the company, it can send daily summaries, monitor connected apps and carry out tasks such as building dashboards, scheduling meetings or creating financial forecasts. Several outlets (CNBC, Engadget, VentureBeat, The New York Times) have covered the same launch.

Access is through a waitlist. Eligible users will be able to use it from a standalone app or from the Claude website. During the preview it is available only to Pro and Max subscribers.

Anthropic sums it up this way on its blog, as quoted by CNBC: the agent “comes to you with what needs your attention, completes the tasks you ask of it, keeps working even when you step away, and checks with you before acting on your behalf.” A proactive agent is one that doesn't wait for you to write to it: it observes your context and proposes or carries out work on its own initiative.

The company itself acknowledges that it is at an early stage and that it “could misread a thread or misjudge what is urgent.” It also admits that, because it is geared toward the workplace, personal errands such as booking tickets are not “as robust.”

The competitive landscape explains the rush. According to CNBC, Muse, Meta's “personal agent,” launched last month and reached the top spot on Apple's App Store. Mark Zuckerberg told analysts on Wednesday that more than 10 million people in the U.S. and Canada have set up a Muse agent in a month and a half. OpenAI unveiled its Dots, “always-on” agents, weeks later at its annual developer conference.

The territory is being carved up. Meta is aiming mainly at consumers, OpenAI covers both enterprise and consumer, and Anthropic is staking out work. That is consistent with its history: it has competed for enterprise market share more than for mass attention.

**Our take.** We are moving from AI that answers to AI that watches and acts. The era of the chatbot waiting for instructions is ending, and that shift matters more than any model improvement. What is at stake now is who manages to be present in the user's daily workflow, and that looks more like a distribution war than a benchmark race.

We see three things. First, trust is the product. An agent that monitors your connected apps and acts without being asked only works if its mistakes are cheap and reversible. The promise to “check before acting” is the most important design decision in the announcement, and the admission that it may misjudge urgency is an unusual degree of honesty. It deserves credit, though the real test will come with use.

Second, Meta's figure should be read carefully. Ten million agents set up measures initial adoption, not retention or proven usefulness. The same goes for the top spot on the App Store: it signals curiosity, not sustained value. There is still no independent data on how much useful work these agents actually do.

Third, the impact on jobs is real, and it is arriving through this route. Broadly speaking, agents that prepare reports, schedule meetings and draw up forecasts touch precisely the administrative and coordination work that is most exposed, according to our sector analysis. For a small business with no back office, an agent like this can be an employee it could not afford. For those who make a living from those tasks, it is a hard transition in the short term, and there is no point sugarcoating it.

The long-term horizon is brighter. If these agents become reliable, they free up time and management capacity for those who today have no team: freelancers, small businesses, solo professionals. It is a modest but concrete form of the abundance we advocate: that the capabilities of a large organization stop being the privilege of the large.

One practical caveat: which users it will work for, and under what privacy conditions, remains to be seen. An agent with permanent access to your apps concentrates a great deal of sensitive information, and questions of security and data access should be at the center of the debate ahead of the marketing. If Anthropic lives up to the caution it preaches, its edge will be less about speed than about trust.