Claude Duet showed up on October 21 as a waitlisted research preview — a proactive AI agent that checks your apps, sends a morning brief, and asks before it acts. Right now it's US-only and gated to Pro and Max subscribers. For most businesses outside that narrow slice, nothing changes this week. For everyone building with AI agents, the approval-gate pattern it ships with is worth studying regardless of whether you can log in.
What is Claude Duet, exactly?#
Anthropic introduced Claude Duet on Wednesday, October 21, as a standalone app running on Claude Opus 5.5 — separate from the core Claude product. It connects to your apps, sends a short brief each morning on what needs attention, and checks in every few hours rather than waiting for you to open a chat.
It runs tasks inside its own virtual machine with a browser, so it can do real work without touching your actual computer. Per Thurrott's coverage, it connects to a large catalog of apps (Thurrott puts the figure north of 6,000); MacRumors frames it as Anthropic's answer to Meta's Muse and OpenAI's Dots, which we covered when Muse shipped.
What it's actually doing in that virtual machine, per both reports:
- Sending a short morning brief on what needs your attention across connected apps
- Checking in every few hours instead of waiting to be opened
- Drafting things — forecasts, simple dashboards, meeting scheduling
- Asking for a yes before it sends a message, pays, signs, or deletes anything
Reports disagree on mobile availability — MacRumors says iOS only, Thurrott says Android and iPhone both are live. I haven't been able to pin down which is right, so take the platform list as "web, plus phone, details pending."
Does this actually change anything for your business?#
If you're not a US-based Pro or Max subscriber, no — there's nothing to sign up for yet. If you are, this is still a day-two research preview, not something to wire into customer-facing work. The part that matters either way is the design pattern: it won't send a message, make a payment, sign a document, or delete a file without your yes first. That's the bar every agent in this category should now be measured against, including the ones we build.
We've written before about how fast this category moves — Microsoft shipped its own agentic model the week before, and Anthropic's Q3 revenue shows how much is riding on exactly this kind of product landing well.
The honest trade-offs#
The upside is real: a tool that checks your inbox, calendar, and task list every few hours and only interrupts you when something's actually worth interrupting for is a genuinely different shape of software than a chatbot you have to remember to open.
The downside is just as real. It's a waitlist product with no published pricing beyond "counts against your existing plan," no Team or Enterprise date, and — per both reports — no public detail yet on exactly how the safety classifier decides what counts as sensitive. Anthropic has said more detail is coming. It isn't out yet. That's a fair thing to wait on before you'd trust it with anything that touches a client.
How we're handling it#
At WebEpex, Claude's coding agent already runs inside a client's production trading-platform stack, watching an Ostium-based execution engine for errors around the clock. That's a different job than Duet does, but it's the same lab, and we trust the approval-gate instinct because we've already built our own version of it.
Every WhatsApp and voice flow we ship for clients across the GCC, Europe, USA, Canada, and India hard-codes the same rule Duet launched with: a human confirms before the bot speaks for the business, not after.
We reviewed Duet the day it came out and made a simple call: we're adding it to the watchlist we keep for client advisory conversations, but every automation we ship still runs on infrastructure we control end to end — self-hosted VPS, PM2, n8n, our own approval logic — not a subscription waitlist we don't operate. My read is that changes once Duet (or whatever Meta and Microsoft ship in response) gets enterprise access and a real API. It hasn't yet.
What I'd tell a client asking about this#
Join the waitlist if you've got a US Pro or Max seat and want to see where the category's heading — but point it at your own busywork first, not anything customer-facing. If you're outside the US or on a Team plan, there's genuinely nothing to do this week. Worth remembering for your next vendor conversation, though: if someone's pitching you an "autonomous" bot, ask whether it asks permission before it acts. If it doesn't, it's behind what Anthropic just shipped on day one.
If you're trying to work out whether any of this hits your own setup, send me what you're running and I'll tell you straight — takes two minutes, no pitch attached. cal.com/webepex/growth-review
