Technology

Anthropic Launches Claude Duet, a Triage-Style AI Agent, in Research Preview

Martin HollowayPublished 8m ago3 min readBased on 1 source
Reading level
Anthropic Launches Claude Duet, a Triage-Style AI Agent, in Research Preview
Photo by Matheus Bertelli on Pexels

Anthropic has launched Claude Duet, its AI agent for consumers, as a "research preview." The Verge reported the launch on 13 November 2026.

Access is limited to Pro, Max and Ultra subscribers in the United States. Users without access can join a waitlist. Claude Duet has its own standalone app, so users do not need to open the main Claude interface to reach it.

Anthropic is pitching the product at professionals and small business owners who need help deciding what to tackle first. The agent surfaces to-dos as cards. Users swipe left to hide a card or right to accept it. The final decision on each item stays with the user. The agent proposes and sorts; the user decides.

The broader context here is the shift from chat as the primary interface toward agents that act on a queue of tasks. A chat window waits for a prompt. A triage surface pushes work to the user, which changes what the model has to get right. Ranking matters as much as generation. A suggestion that is plausible but misprioritized costs the user attention, and a suggestion accepted in error costs more. The swipe gesture gives the user a fast, reversible way to correct the agent. It also means the quality of the queue depends on the model's judgment about urgency, a judgment that can look reasonable on any single card and still be wrong in aggregate.

For technical readers, the open questions are the ones a preview is designed to answer: how the agent decides what needs attention, which data sources it reads, how it handles conflicting signals, and what happens when an accepted to-do turns out to be wrong. The research preview label and the subscriber-only, US-only waitlist suggest Anthropic expects to learn these things from a limited group before widening access. That is a common pattern for consumer agents, and it leaves the most consequential behavior to be observed in use rather than inferred from the launch.

In my view, the most useful test for Claude Duet will be how it behaves on ordinary, low-stakes work. Small business owners already juggle invoices, reminders and follow-ups across several tools, and a reliable triage layer could ease that routine. The features that make the product convenient, including its ability to propose actions the user then accepts, are the ones that will need the closest scrutiny in the coming months. The swipe design keeps a human in the loop for every commitment, a sensible default for a product still in preview.

If the ranking holds up, the practical gain is modest but real: fewer items slipping through and less time spent deciding what to look at first. That is a narrow benefit, and it tends to matter most to people who run small operations alone. Whether it holds up with a wider user base is the question the preview is built to answer.