The takeaway
Duet reframes the agent product around approvals and background execution rather than conversation, which makes identity, credentials and permission design the first engineering problem for anyone building proactive agents.
Why it matters for builders
Treat the approval gate as architecture: model identities, credential custody and scopes before the task loop. The morning brief plus a short decision queue is now a proven interface pattern that automation teams can ship on their own stack, and the 6,000-connector marketplace shows integrations are the distribution channel for agent apps.
Anthropic launches Claude Duet, a proactive agent app
Anthropic opened a research preview of Claude Duet on Wednesday, a standalone agent app for iOS, Android and claude.ai aimed at professionals and small business owners rather than consumers. It runs on Claude Opus 5.5, is included in Pro and Max plans, starts in the US and requires joining a waitlist.
The pitch is a break from chat: Duet comes to the user with what needs attention, keeps working while they are away, and asks before it acts.
What Duet actually does
Each morning Duet sends a short brief on the most important items across connected apps. Through the day it keeps tabs on those apps and messages the user only when something needs attention. Workloads range from scheduling meetings to building dashboards across dozens of tools, financial and sales forecasts, prototyping and shipping small web apps, according to Anthropic's announcement.
Tasks run in a background cloud computer, continuing while the user steps away. The app carries over the memory, skills and connected apps from the existing Claude app and works with more than 6,000 connectors and plugins, per the product page.
The race moves from assistants to agents
Duet lands in a suddenly crowded field. TechCrunch puts it against Meta's Muse, OpenAI's Dots, Grok Bot and Fo, framing Anthropic's angle as professional work rather than errands. Anthropic draws the line itself: Duet is not built for booking tickets, and it does not show browser use live.
That positioning matters commercially. Assistants won the last two years on conversation quality. Agents are judged on whether they can be left alone with a task list and a company's data without causing an incident.

Approvals are the product
Most of Anthropic's announcement is safety engineering, which is the interesting part for anyone shipping agents. Credential handling follows the Claude Managed Agents design, so Duet never directly sees passwords or sign-in tokens. Actions are gated by the same auto mode classifier that guards Claude Code, and Anthropic added a dedicated classifier plus safety system prompt that rejects sharing credit card numbers and government identification; health and financial data move only when the user explicitly asked.
TechCrunch reports purchases route through Stripe Link or Shop Pay with user approval, and that Duet will not trade securities or move money between accounts. The preview was red-teamed internally and by an external penetration testing firm and is covered by Anthropic's bug bounty.
What builders should take away
The approval gate is now architecture, not a feature toggle: identity, credentials and permissions get designed first, then the task loop. Duet also mirrors agent commerce, where Walmart opened an agent gateway while Amazon closed its doors and credential brokers like 1Password race to keep passwords out of agent context.
The practical lesson is narrower than the launch hype: a morning brief with a short queue of decisions is becoming the default interface for proactive agents, and any automation team can build that format on its own stack today.
The Automation Brief
Read 5 AI stories instead of 50.
The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.
No noise. Unsubscribe anytime.
Editorial notes
Stefan Trbojevic
n8n Lab Editorial
22 October 2026
22 October 2026
Sources
AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.




