For three years, using AI meant opening a window and typing. Claude Duet flips the order. According to the official page:
- every morning it tells you what needs your attention, with next steps already drafted;
- between tasks it stays quiet, and only reaches out when a decision is needed;
- longer jobs run on its own cloud computer, with the apps you connect: more than 6,000 connectors, including Gmail, Slack, Stripe and Google Drive;
- before sending a message or making a payment, it asks.
It is a research preview behind a waitlist, for Pro and Max subscribers in the US. Team and Enterprise plans are not included yet.
I was already doing this with an email
The morning brief is not a new idea. At work I taught people to set up a scheduled prompt on Copilot: every morning an email arrives, you click and land on the day’s briefing. It worked for one reason only: the AI sat where the day was already starting.
Duet does the same without the email trick. And it adds the part that changes everything: it does not just tell you, it acts.
When AI answers, you read the mistake. When it acts, your customer does.
The right question is not how good it is
Anthropic itself writes that Duet can misread a thread, get urgency wrong and miss something that arrives between check-ins. That is honest, and inside a company it shifts the question: not how intelligent the agent is, but what it can touch.
An agent that summarises your email and one that can change an opportunity in the CRM are not the same risk. Put them through the same approval process and the first gets slowed down for nothing while the second gets too much. Autonomy belongs to the individual action, not to the agent.
There is also a cost you do not see: background check-ins use up the plan’s limits. An agent that is always on keeps spending even when you are not using it.
What I would do
Before connecting it to anything work-related, I would write three columns on a sheet of paper: what it can do alone, what it can do with my approval, what it must never do. If that sheet does not exist, it is not the agent that is not ready.
Why operational memory has to stay with the company.
If an agent could write to your customers tomorrow, could you say which messages it may send without asking?
Sources
- Anthropic, Claude Duet, official product page