Anthropic's Claude Duet announcement moves Claude from a chat you open to an agent that can brief you, watch connected work, carry out tasks in the background and ask for decisions. In his launch breakdown, Brock Mesarich walks through the article and asks whether this is enough to draw him away from Grok Bot. His video is a reading of Anthropic's materials and his own Grok Bot experience, not a hands-on Duet review: he says he has joined the waitlist and will test it when he gets access.
Claude Duet is an early, waitlisted research preview for US Claude Pro and Max subscribers. Anthropic says it brings your Claude memory, skills and connected apps into a separate mobile app or claude.ai, then works from one ongoing voice-or-text conversation. It runs on its own cloud computer, but background check-ins and longer tasks draw from the same usage limits as your other Claude work. There is no announced Portugal availability date in the launch article.
Watch Brock's Breakdown and Anthropic's Launch Film
Brock's commentary: His seven-minute video follows the launch article section by section. The time-linked chapters are below.
Official demo: This is the film embedded in Anthropic's October 21 launch article. It illustrates intended workflows; it is not a reliability benchmark or proof of general availability.
How Duet Is Supposed to Work
Anthropic describes a morning brief across connected apps, then periodic checks that surface items needing attention. It says Duet is built around Claude Opus 5.5 and carries over the memory, skills and connections in an existing Claude account. A user can type or speak in one ongoing thread. Duet has a virtual machine and browser in the cloud, so background work does not require the user's phone or laptop to stay on. The company says it checks accounts every few hours rather than constantly, partly to manage usage.
The announcement's examples range from finding missed Gmail and Slack messages to drafting invoice reminders from QuickBooks or Stripe, preparing a payroll run in Gusto, and assembling an agreement for Docusign review. A launch-film scenario has Duet flag a coffee-shop permit before a new location opens. These are vendor examples, not a promise that every connector supports every action or that Duet will correctly prioritize every business. The Claude Marketplace is the place to check an individual connector's access and capabilities.
Anthropic quotes early testers who used Duet to talk through a task list, catch a misdated property showing, or prepare information before a client call. Those anecdotes show what some invited users reported. They do not establish a measured productivity gain for everyone.
Approval and Privacy Boundaries
The central design claim is not that Duet has blanket permission to act. Anthropic says it can read, draft and prepare, but must show its intended step and get approval before sending a message, making a payment, signing a document or deleting anything. Trading securities and moving money between accounts are off-limits even with approval. Users can disconnect an app. For a business purchase, Anthropic describes a separate Link approval and virtual-card flow; it does not say that Duet sees a card number.
The company also says Duet never directly sees passwords or sign-in tokens. Its new classifier and safety prompt are designed to reject requests to share credit card numbers or government IDs, while health and personal-finance information can be sent only with explicit instruction. Anthropic reports red-team work, external penetration testing and bug-bounty coverage, and connects Duet to its trustworthy-agent framework. Those are meaningful published safeguards, not a guarantee that prompt injection, misclassification or an incorrect action is impossible. Give an agent only the accounts and permissions it needs, then inspect high-impact drafts and approvals yourself.
The Catch: Access and a Shared Meter
Access begins in the United States
What Anthropic announced
The research preview starts by waitlist for US Pro and Max subscribers. Duet is described as a standalone iOS and Android app and as available on claude.ai for admitted users. It lives separately from the main Claude app for now.
What that means here
Being a Claude subscriber in Portugal does not establish eligibility. The launch article does not give a European rollout date. Join the official waitlist if it accepts your details, but do not build a time-sensitive workflow around access you have not received.
Background work uses your Claude allowance
What is included
Anthropic says there is no separate research-preview fee on Pro or Max. Duet uses the same plan limits as the rest of Claude; its scheduled account checks count, and bigger work such as decks or financial models needs more turns.
What to watch
If Claude Code or long Claude sessions already take you near your cap, an always-on agent could make that limit more noticeable. Brock says Opus 5.5 has felt efficient in his own coding work, but he has not yet measured Duet's consumption. Start with one narrow responsibility and observe usage before adding more.
It is a preview, not a finished assistant
Current emphasis
Anthropic built Duet around professional and small-business work. It says the agent may misread a thread or judge urgency poorly and that personal errands such as ticket booking are less mature.
Still missing
The announcement says Duet cannot yet show users its browser. Anthropic has not announced multiple separate Duet agents or cross-agent collaboration. Those differences matter for anyone expecting Grok Bot's named, parallel team model.
Duet vs Grok Bot, Dots and Muse
Brock's own daily driver is Grok Bot, largely because he can split work among named agents. That is a workflow preference, not a controlled head-to-head Duet test. The official product pages support a narrower comparison:
Claude Duet
Center of gravity
Existing Claude memory, skills and connected apps, carried into one ongoing conversation. Briefings, background work and approval-gated actions are the announced design.
Decision point
Best positioned for someone already invested in Claude, provided US preview access and shared usage limits fit. Its eventual reliability and broader availability still need testing.
Grok Bot
Center of gravity
Several named Bots can work in parallel, coordinate and retain role-specific context. xAI's documentation says they share one cloud computer, including its files and logins, so role names alone are not data-isolation boundaries.
Decision point
Useful if distinct agent roles and handoffs are central to your workflow. Brock's statement that Grok Bot may route to multiple companies' models comes from a reported future plan, not a published per-task model audit; do not treat that as a verified quality advantage.
ChatGPT Dots
Center of gravity
Ongoing responsibilities built from ChatGPT memory, connected tools, cloud computer use and Codex handoffs. OpenAI describes reviewable work and user-controlled permissions.
Decision point
Check the current plan and regional rollout before comparing it with Duet. OpenAI's release notes exclude the EEA, including Portugal, from Pro access at launch, so neither launch should be presented as a tool local readers can use immediately.
Meta Muse
Center of gravity
A personal agent in its own secure VM, designed for messaging-style interaction in the Muse app or WhatsApp. Meta says users control connected-app access and approve sensitive actions.
Decision point
It is positioned more broadly around personal tasks than Duet's work-first preview. Meta's announced rollout and privacy claims are distinct from an independent comparison of actual outcomes.
The useful distinction is interface and operating model, not a speculative winner: Duet starts with Claude continuity, Bot exposes named cooperating roles, Dots integrates ChatGPT and Codex work, and Muse emphasizes a personal messaging agent. Pricing, regions, permissions and observed reliability can change; recheck the official pages before connecting sensitive accounts.
A Sensible First Duet Test
If you receive access, give Duet a reversible, low-risk responsibility: review one calendar and one inbox for tomorrow's conflicts, then prepare a morning brief with source links and draft-only follow-ups. For a week, record useful alerts, false alarms, missed items, approval requests and visible usage consumption. Expand to invoice, payroll or purchase workflows only after you know what it sees, what it can change and who reviews every consequential step. That would be evidence of fit for your work, which Brock's launch-day breakdown cannot yet provide.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | Anthropic announces Claude Duet |
| 00:27 | Who Duet is for |
| 02:18 | Safety claims and sensitive information |
| 02:49 | Early-user examples |
| 03:16 | How Claude Duet works |
| 05:13 | The shared usage-limit catch |
| 05:48 | Duet vs Grok Bot, Meta Muse and Dots |
| 06:19 | Why Brock uses Grok Bot |
| 06:41 | The model-routing question |
| 07:18 | How to request access |
Official Sources and Verification Limits
Primary video: Brock's breakdown and reader-supplied transcript. Anthropic: launch article and embedded film, Duet waitlist, Marketplace, and agent-safety framework. Comparison: xAI Bot docs, OpenAI Dots and rollout notes, and Meta Muse. Checked 22 October 2026. Duet access, usage, safeguards and example outcomes are Anthropic's descriptions; neither Brock nor this article independently tested Duet.