When AI Stops Waiting and Starts Watching
Anthropic's new Claude Duet agent works in the background and messages you when a decision is needed — a shift that moves AI from answering questions to noticing what requires your attention.
- Oct 21, 2026
- research preview launched
- 6,000+
- connectors in the Marketplace
- Claude Opus 5.5
- model powering Duet
- Pro & Max only
- access tiers (US waitlist)
Most AI tools work only when you open them and type a question. Claude Duet turns that around: the agent messages you first. Anthropic launched a research preview on October 21, and it is built to watch your connected apps — email, calendar, whatever you link — and send you a short briefing each morning on what needs your attention. Through the day it keeps watching, and it contacts you only when a decision is needed.
It can do real work in the background — drafting documents, building dashboards, even buying things through Stripe or Shopify Pay. But it always asks before the sensitive steps: sending a message, making a payment, anything that affects another person or moves money. Hard limits stay in place: it will not trade securities or shift funds between accounts, full stop.
That design matters more than it sounds. Right now, a lot of the value people get from AI depends on how good they are at writing prompts. Duet moves the work earlier: the agent notices what needs doing before you have to think of it. The bottleneck moves from doing to deciding — your job becomes approving or redirecting, not starting from scratch.
This is still early. Microsoft 365 is not yet connected, Slack works with limits, and Anthropic says the preview may misjudge what is urgent. But the architecture is clear — and worth thinking about now. Ask which parts of your workflow are really just 'notice something, route it to the right place.' Those are exactly the tasks Duet is built for, and the teams that map this out first will gain the most.