OpenAI's ChatGPT Dot gains ability to initiate text conversations via iMessage and RCS, U.S. Pro preview
Claude Duet proactive agent
Latest: Anthropic Opens Research Preview of Claude Duet Proactive Agent
Duet represents Anthropic entering the proactive agent category alongside Meta's Muse and OpenAI's Dots, shifting from reactive chat toward agents that initiate contact and act autonomously on users' behalf. The safety and cost tradeoffs, particularly around background check frequency and approval requirements, set a design point that differs from competitors.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
The full picture
Claude Duet is in research preview, available to U.S. Pro and Max subscribers via a waitlist at claude.com/product/claude. The app runs on Opus 5.5 on a cloud virtual machine and operates while the user's phone and laptop are off. It monitors Slack, email, and calendars, supports over 6,000 connectors including invoicing and payroll tools, and integrates with QuickBooks, Gusto, DocuSign, HubSpot, and Shopify. Rather than checking continuously, Duet checks connected accounts every few hours and sends a morning brief when action is needed. Background checks draw on the same usage allowance as regular Claude chats, which means continuous checking would consume Pro and Max limits faster; at Opus 5.5 API pricing, six background checks per day could cost roughly $34 per month at API rates, exceeding the Pro subscription price. All irreversible or sensitive actions require explicit user approval before execution, and securities trading and inter-account money transfers are blocked entirely. Purchases require two approvals: Duet confirms the item and Stripe's Link wallet approves the spend, after which Duet pays with a virtual card that keeps the real card number private. Credential handling follows the Claude Managed Agents design, so Duet never directly sees passwords or sign-in tokens. A safety classifier blocks sharing of credit card numbers or government IDs, and sensitive health or financial data is only transmitted when the user explicitly instructs Duet to send it.
How it developed
Transformer News reported that Anthropic is preparing a new model described as especially good at discerning a person's underlying intentions and goals even without much information from the user, a development separate from Duet, the proactive agent app Anthropic launched October 21 on Opus 5.5.
Anthropic opened a research preview of Claude Duet on October 21, 2026, a standalone Opus 5.5 app that monitors Slack, email, and calendars across more than 6,000 third-party integrations, including invoicing and payroll tools.
Duet sends a morning brief on important items and handles work from admin tasks to building web apps; early testers report reduced context switching. All irreversible actions require explicit user approval, securities trading and inter-account transfers are blocked, and a waitlist is open to US Pro and Max subscribers.
Anthropic made Opus 5.5 fast mode broadly available on the Claude Platform at $8/$40, delivering up to a 2.5x speedup.
Coverage of Duet's October 21 launch also confirmed the app runs on a cloud virtual machine to operate while the user's phone and laptop are off, and named specific integrations including QuickBooks, Gusto, DocuSign, HubSpot, and Shopify. Anthropic acknowledged urgency as a weak spot, noting the meaningful signal is how often a notification proves worth the user's attention.
Sources
3 more sources
- Meta's Muse personal AI agent
- Claude's Gmail and Google Drive integration
- OpenAI's Dots always-on agents
- Anthropic background computer-use in Claude apps
BackgroundAI agents: models that take actions
Get the daily email
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.