Anthropic Shipped Claude Duet, an Agent That Checks Your Apps Every Few Hours and Acts for You. Here's What It Replaces in a Solo Operator's Week.
This week Anthropic opened a research preview of Claude Duet, a standalone app where Claude works as a proactive agent: it watches your connected apps, sends a morning brief, takes on tasks, and keeps working in the cloud after you close your laptop. It is available by waitlist to US Pro and Max subscribers, built around Claude Opus 5.5, with no separate price listed on Anthropic's page. I have not had access, so everything below comes from Anthropic's own description, and the useful question for a one-person business is which parts of your week it could realistically take.
What Anthropic says it does
Anthropic's page describes Duet as built for professionals and small business owners. It sends a short morning brief on the most important items across your connected apps, monitors those apps, and messages you only when something needs attention. It takes on tasks from scheduling meetings to building dashboards, financial or sales forecasts, prototypes, and web apps.
It runs in the background on its own virtual machine with its own browser, so your laptop or phone does not need to stay on. It carries over your memory, skills, and connected apps from your Claude account, and it connects to over 6,000 connectors and plugins in the Claude Marketplace, with examples that include Gmail, Slack, QuickBooks, Stripe, Gusto, HubSpot, Canva, and Shopify. It can make purchases through Link, Stripe's digital wallet, using a virtual card, and can check out with Shop Pay. It runs as an iOS and Android app and on claude.ai, separate from the main Claude app for now.
The gates are the product
Anthropic says Duet reads, drafts, and prepares work on its own, but before it sends a message, makes a payment, signs a document, or deletes anything, it shows you the action and waits for approval. Trading securities and moving money between accounts are off-limits even with approval. The page also says approvals are handled by Anthropic's auto mode classifier, and that Duet never directly sees your passwords or sign-in tokens.
That design tells you what Anthropic thinks the failure modes are. It also admits them: Duet may misread a thread or misjudge urgency, personal errands like booking tickets are weaker than work tasks, and it cannot yet show you its browser. An early reviewer cited by one newsletter said it flagged too much, which is the failure I would rather have than the opposite.
One more limit to plan around. Duet checks connected accounts every few hours rather than constantly, because Anthropic says constant checking would burn through plan limits, and background checks count against your plan like everything else. If you build a workflow that depends on a response within minutes, this is not the tool.
What I'd hand it first
For a solo operator the strongest case is the morning brief across the apps where things quietly go wrong: failed Stripe payments, a customer thread that went unanswered for two days, an invoice due in QuickBooks. Reading and summarizing across tools is the lowest-risk, highest-boredom work in my week. If the brief is wrong, I lose a minute.
Second, I would try recurring drafts: the weekly metrics summary, the follow-up emails I put off, the sales forecast spreadsheet I rebuild by hand every month. Drafts sit behind your approval, so a bad one costs you review time, not a mistake in front of a customer.
What I would not hand it yet
I would keep purchases off for the first few weeks. A virtual card plus an agent that is still learning what counts as urgent is a combination worth watching before you trust it. I would also not let it reply to customers on a first-name basis until I had read a month of its drafts and knew how often I edited them. The approval step is only a safeguard if you actually read what you are approving, and a long queue of requests trains you to click through.
What I'd actually do
Join the waitlist if you are on a US Pro or Max plan, connect two or three read-heavy apps, and run it for a week in brief-only mode. Count how many items it surfaced that you would have missed and how many were noise. If fewer than half of the alerts mattered, tighten what it watches before you give it any write access.
Where this could go wrong
It is a research preview from a company that says plainly it will misjudge things. The unknowns that matter most to a solo operator are the ones the page does not answer: how fast it burns your plan limits when you give it larger tasks, since Anthropic says deck and model building use more, and whether the "every few hours" cadence misses the one email that mattered. I also do not know how it behaves when two connected apps disagree about the same fact. Treat the first month as a test, not a delegation.
Author
Lukas
@lukcombinator