Blog
Claude Duet Research Preview: The Closest Lab Product to Day-to-Day Business Ops
Anthropic opened Claude Duet as a waitlisted research preview for US Pro and Max users. Human approval before send, pay, or sign is the feature that matters.

Claude Duet Research Preview: The Closest Lab Product to Day-to-Day Business Ops
On October 21, 2026, Anthropic opened a research preview of Claude Duet, a proactive agent aimed at professionals and small business owners "with more on their plates than they can track." Access starts by waitlist for US Pro and Max subscribers. A safety, privacy, and security follow-up landed on October 23, 2026.
This is not another chat window with a longer memory. Anthropic's framing is an agent that comes to you with what needs attention, keeps working when you step away, and checks with you before it acts. For operators, the approval gate is the product.
What Anthropic shipped
Claude Duet is available as a standalone app on iOS and Android, and on claude.ai. For the research preview, Anthropic says it is included in Pro and Max plans and uses the same usage limits as the rest of Claude.
Anthropic designed Duet around Claude Opus 5.5. Each morning it can send a short brief on important items across connected apps. Throughout the day it checks those apps and messages when something needs attention. You can also hand it tasks ranging from scheduling to dashboards, forecasts, prototypes, and web apps.
Work can continue when you leave. Anthropic says Duet has a virtual machine with its own browser, so it does not require your laptop or phone to stay on. You get a notification when a task finishes or needs review.
Human approval before send, pay, or sign
The line that matters for real businesses is explicit in Anthropic's launch article:
Before Duet "sends a message, makes a payment, signs a document, or deletes anything, it shows you exactly what it's about to do and waits for your approval."
A few actions stay off-limits even with approval, including trading securities or moving money between accounts. You can disconnect any connected app.
That is the right default for operators. Draft and prepare in the background. Keep irreversible steps behind a human click. Anthropic also says Duet never directly sees your passwords or sign-in tokens, following the credential design used for Claude Managed Agents.
Connectors that map to real ops
Anthropic says Claude Duet can use the over 6,000 connectors and plugins in the Claude Marketplace. Named examples include:
- Gmail and Slack for missed messages and replies
- Intuit QuickBooks or Stripe for unpaid invoices and draft reminders
- Gusto for payroll prep with hours, bonuses, and reimbursements filled in
- LinkedIn for drafting and, after edit and approval, posting a job
- Docusign for preparing agreements
- HubSpot for follow-up drafts from contact and deal history
- Canva for turning a product launch into a social campaign
- Shopify for store setup and management, with a richer visual editor coming later
Anthropic also describes purchases through Link, Stripe's digital wallet, with spend approval and a virtual card so the real card number is not shared, plus Shop Pay for online checkout. Coming soon, with permission, saved 1Password logins.
Those are product claims about available connectors and flows. They are not proof that every connector is reliable for your books tomorrow morning.
Early business use, as Anthropic presents it
Anthropic published early-user stories. These are vendor testimonials, not independent production audits:
- Amin Ebrahimi, owner of Shahs of Kabob with about 110 employees, used Duet for accounting, customer support, and marketing, working through invoices store by store: "It brings it all together. I don't need to open up 100 tabs."
- A broker flagged a property showing booked on the wrong day without being asked
- A brand strategist got a meeting brief and a warning that the call sat right before a flight
Useful as a picture of the intended buyer. Not a reliability study.
Stated limits of the research preview
Anthropic is direct about what still breaks:
- Duet "might still misread a thread or miscalibrate what's urgent"
- Personal errands "aren't quite as strong"
- It "can't yet show you its browser"
- Background checks count toward plan usage limits
- Larger tasks need more model turns and background work
Independent measurement of reliability in production SMB use remains unknown while this is a research preview.
What this is not
Duet is not a license to turn off judgment. Research preview means Anthropic is still learning when the agent should nudge you and when it should stay quiet. It is also not a separate unlimited plan. Background checks spend the same Pro or Max budget you already use for Claude.
Do not treat waitlist access as a full-time COO. Treat it as a supervised operator on non-critical paths first.
How to evaluate it if you get access
- Connect one or two systems you already live in, such as Gmail plus QuickBooks or HubSpot.
- Start with drafts only: invoice reminders, meeting briefs, follow-ups that require your send click.
- Keep money movement, signatures, deletions, and customer-facing sends behind the approval prompt every time.
- Watch usage. Background checks are not free, even when you are not chatting.
What it means for operators (Jaden's take)
I have been waiting for a lab product that looks like actual business ops instead of a demo that renames my files. Duet is the closest I have seen from a major lab: morning briefs, invoice reminders, payroll prep, HubSpot and Shopify work, and an explicit stop before send, pay, sign, or delete.
My rule is simple. Join the waitlist if you are on US Pro or Max. Test on workflows where a wrong draft is cheap and a wrong send is blocked by the approval gate. Do not hand it your sharpest customer thread or your only payroll run on day one.
Want the agent that prepares the work. Keep the click that ships it. That is how you get leverage without buying a new class of mistakes.