← Back to the blog AI Consulting · Toledo, Ohio

Claude Duet Can Make Purchases for You. Decide What Yours Gets Access To First.

Jayson Hines · October 24, 2026 · 4 min read

Anthropic previewed Claude Duet on Friday, a personal AI agent that does work on its own instead of waiting for you to ask. According to CBS News it's aimed at professionals and small business owners. It sends you a daily brief, schedules meetings, builds sales forecasts, drafts job postings, and can buy things on your behalf. Right now it's a research preview, U.S. only, by waitlist for Claude Pro and Max subscribers, with invites going out in waves.

I'm not telling you to sign up and I'm not telling you to stay away. My point is that when your invite shows up, you'll be looking at a screen asking what to connect, and that's a bad moment to make the decision.

What it needs to be useful

To do anything worthwhile, Duet has to see your email and calendar. It also carries over the memory, skills, and connected apps from your Claude account, so whatever you've already hooked up comes along. Anthropic says it never directly sees your passwords or sign-in tokens, and that it checks with you before it acts. Fine. But "checks with you" only protects you if you read what it's asking. Anyone who has clicked Allow on forty cookie banners knows how that goes by week two.

Security people quoted in the coverage said the same thing I'd say: limit what you share.

The buying part

Of everything on that list, I'd go slowest on purchases. A bad meeting invite is annoying. A bad order is money gone. I wrote last week about Musk saying SpaceX would cover Grok's purchase mistakes while the terms said otherwise (that one's here). The CBS piece doesn't say who eats the cost if Duet orders the wrong thing, and I haven't seen it spelled out anywhere yet. Until I do, nobody's putting the company card on it.

The ten minute plan

Start with who gets to use it. Your own inbox is one thing. Your office manager's inbox is another, because it's full of customer info, vendor invoices, and payroll questions. And watch for the sneaky one, which is staff signing up with a personal Claude account and connecting their work email to it. If you don't have a rule on that yet, make one this week.

Then pick the low stakes jobs first. The daily brief and meeting scheduling are good places to begin. Keep it away from money and away from anything that goes out to customers until you've watched it work for a month.

Last, write down what you connected and why. When something goes sideways, or when you hire someone new, you'll want a real answer to "what can this thing see?" and you won't remember it off the top of your head.

These agents will get better and they'll land in more of your tools, so the habit of deciding access up front is worth building now. It's the same habit I push in the AI policy conversation with every owner I work with.

I run AI workshops and one-on-one AI consultations for companies around Toledo, Northwest Ohio, and Southeast Michigan, and a big part of that is helping owners set the rules before an AI agent gets near their email or their credit card. If you'd like a hand with that, send me a note.

Email Jayson