Most AI tools still work the same way: you open a chat window, type a question, and wait for an answer. Anthropic's new research preview, Claude Duet, is built around a different idea. It is meant to function as a standing work surface that can open with a daily brief, keep an eye on connected apps, and carry recurring work forward between prompts.
Anthropic describes Duet as a new app and a new way to work with Claude, now in research preview. CNBC reports that access is waitlisted for Pro and Max subscribers through a standalone app and the Claude website. Anthropic's release notes also identify Duet as a research preview for those plans.
What Claude Duet is built to do
Anthropic's product page and CNBC's reporting point to a familiar set of workplace jobs: daily briefs, connected-app monitoring, dashboards, meeting scheduling, and financial forecasts. The product page also describes handing off tasks, scheduled work, connectors, and the ability to edit output before it goes out.
The important change is persistence. Instead of treating every request as a fresh chat turn, Duet is designed to keep a thread of work alive across a day or a week.
Why the “agentic harness” matters
A harness is the product system around an AI model: the context it can draw on, the task surface it uses, and the way work can keep moving when you are not typing. Standard chat is mostly prompt and response. A proactive harness gives the model a place to notice, organize, and return to recurring work.
CNBC places Duet alongside OpenAI's Dots and Meta's Muse. It is too early to declare a winner, and CNBC does not detail every feature of those products. The useful comparison is simpler: all three point toward AI products that try to become an ongoing layer of work, rather than another tab you remember to open.
Three good first trials
The best early test is something you already repeat and can judge quickly.
- A weekly customer-feedback digest. Ask Duet to pull themes from the places feedback already lands and give you a short Monday readout.
- Meeting and action tracking. Let it pull decisions and open items into one running view, so follow-ups do not disappear into a week of calls.
- A campaign or metrics readout. Turn numbers you already inspect into a standing snapshot that is ready when you are.
These tests are interesting because they reveal whether a proactive assistant saves real attention, rather than merely producing a competent one-off answer. Duet is still an early preview, so start with work you can evaluate easily while learning how its connected-app controls work.
The real test is whether it earns a place in your week
Claude Duet is not interesting because it can schedule a meeting or summarize a dashboard. Plenty of tools can do pieces of that. The question is whether it can create a useful rhythm around recurring work, surfacing what matters before you go looking for it. That is the promise Anthropic is now putting into the field.
If Claude Duet has you thinking beyond one-off prompts and toward AI that can support real work, the Generative AI for Everyone course is a strong next step. It gives you a practical foundation for judging where generative AI can help, where it needs human direction, and how to choose useful first use cases.*