Your calendar, your inbox, and that client who emails at 7am all want a piece of you before lunch. Anthropic’s new bet is an AI agent that watches all of it overnight and only speaks up when something actually needs you. It’s called Claude Duet, and it quietly launched yesterday.
What Claude Duet actually is
Claude Duet is a proactive AI agent. Most chatbots sit there waiting for you to type. Duet flips that: it checks your connected apps on its own, sends you a short brief each morning, and pings you during the day only when something needs a decision.
Under the hood, it runs on Claude Opus 5.5 on its own virtual machine in the cloud. That detail matters more than it sounds. Your laptop can be off, your phone buried in a drawer, and the agent keeps working anyway. When it finds something worth your attention, it shows up with a summary and a suggestion, not a wall of text.
Anthropic built it for a specific kind of person: someone with more on their plate than they can track. Freelancers juggling five clients. Small business owners drowning in admin work. If your day is three hours of real work interrupted by forty small decisions, congratulations, you’re the target market.
What it does all day
The morning brief
Every morning, Duet sends a short brief covering the most important items across your connected apps. Think of it as an assistant who read everything overnight and left you a two-paragraph memo. You can adjust which subjects it flags and how often it checks, so it doesn’t have to nag you about everything.
One expectation to set: the agent checks apps every few hours, not every second. A genuinely urgent email can land between checks. So it’s background assistance, not a security system.
Tasks you can hand it
The brief is the passive half. The active half is you handing Duet work from your phone: scheduling meetings, building dashboards that pull from a dozen tools, drafting financial or sales forecasts, even prototyping full web apps. Anthropic’s examples run from “move my meeting” to “build me a dashboard”, which is a wild range for a phone app.
Here’s the part I like. Before Duet sends a message, makes a payment, signs anything, or deletes a file, it stops, shows you exactly what it’s about to do, and waits for your approval. You say yes, then it acts. No surprise purchases from your AI while you sleep.
How to get access right now
Access is throttled on purpose while Anthropic works out the kinks. Here’s the path:
- Be on a Claude Pro plan ($20/month) or Max plan (from $100/month)
- Join the waitlist through Anthropic’s official announcement
- Watch your email for the invite
- Open the standalone Duet app on iOS or Android, or use claude.ai on the web
A few catches. It’s US-only for now, so if you’re elsewhere, the waitlist won’t help yet. Team and Enterprise plans can’t get in at all during the preview. And there’s no separate Duet charge right now: the agent is included with Pro and Max while it’s in preview.
Also know that Duet shares your plan’s usage limits with the rest of Claude. Those background checks nibble at the same allowance you use for chatting. Bigger jobs, like building a financial model, burn through more turns than a quick question.
What it won’t do (yet)
Anthropic was unusually honest about the limits, which I respect:
- Personal errands are weak. Booking flights or concert tickets is the wrong job for this tool right now.
- You can’t watch its browser work in real time. Some rival agents show their browsing like a screen recording; Duet doesn’t yet.
- It lives in its own app, separate from the main Claude app, which is mildly annoying.
- It will sometimes misread a thread or get urgency wrong. Anthropic says so itself.
For errands that happen over the phone, there’s already a workaround we covered in Yelp Check for Me: let AI call the restaurant for you. Different tool, different job, but it fills the gap Duet leaves open.
Should you trust it with your accounts?
Fair question, since Duet reads your email and calendar to be useful. Anthropic’s security pitch works on two levels. First, some credentials stay out of the agent’s reach entirely, so a leak can’t expose everything at once. Second, its auto mode runs classifiers that review what the agent does, and its browser defenses treat a sneaky instruction hidden inside an email as an attack, not as a task.
That second part matters. Prompt injection is the attack nobody talks about at dinner, and it’s exactly the one a proactive agent invites. An AI that reads your inbox is an AI a scammer wants to write to. Duet at least ships with defenses designed for that, instead of pretending the problem doesn’t exist.
How Claude Duet stacks up against the other agents
Claude Duet isn’t stepping onto empty ground. OpenAI has Dots. Meta has Muse. xAI has Grok Bot. Then there’s a bench of smaller players like Hark, Instinct, and Fo. TechCrunch counted at least six rivals in the space before Duet even launched.
The difference is the pitch. Most of these agents compete on doing more things. Anthropic’s angle is judgment: an agent that knows when to bother you and when to shut up. If that holds up in daily use, it’s the version of this future nobody will mind living in. If it doesn’t, Duet becomes one more notification source to mute.
What it costs, in plain numbers
No separate Duet fee exists right now, so the real decision is which Claude plan you have:
| Claude Pro | Claude Max | |
|---|---|---|
| Price | $20/month (or $200/year) | starts at $100/month |
| Duet access | included during preview | included during preview |
| Extra agent fee | none | none |
| Usage | shares your plan’s limits; background checks count | shares your plan’s limits; background checks count |
| Availability | US, waitlist | US, waitlist |
My take: if you’re already on Pro and the waitlist clears, try it. The agent costs you nothing extra. Just watch your usage meter, because a proactive agent that checks apps every few hours is a completely different beast from a chat window you open twice a day.
If you’re on the free tier, don’t buy a plan just for this. The preview is early and US-only, and honestly, neither plan is a purchase I’d make on the promise of an agent that might misread half my emails. We broke down the value math across Claude’s subscription plans if you want the full picture before spending anything.
Meanwhile, if you want an agentic AI coworker today instead of after a waitlist, Google’s version is already live for most people. We explained it in Gemini agent: Google’s AI coworker explained.
What to do today
Already on Pro or Max in the US? Join the waitlist now and let the invites catch up with you. Everyone else: bookmark the announcement, keep your plan as-is, and check back once the preview widens.
Claude Duet is the clearest signal yet that AI is moving from “ask and wait” to “watch and tell you”. Try it early, and you’ll know whether that future saves you hours or just adds noise.
