Claude Duet: When Your AI Agent Starts the Conversation
Every assistant you have used shares one design constraint: you go first. Claude Duet—Anthropic's always-on proactive agent, in research preview on iOS, Android, and web—removes it. The agent works in the background, delivers a daily brief, and asks permission before acting. It is the highest- signal entry yet in the proactive-agent race against Muse, Dots, and Grok Bot, and it changes what "using an AI assistant" means.
What shipped: an agent with initiative
Claude Duet, still dominating X AI discussion in late October 2026, inverts the interaction model. The official getting-started guide (support.claude.com) describes an agent that:
- Runs continuously. Not a session you open—a background presence across phone and web that keeps working after you close the app.
- Delivers a daily brief. A digest of what it did, found, and decided needs your attention, replacing the check-in ritual of opening a chat.
- Does background cloud work. Research, drafting, and monitoring happen on Anthropic's infrastructure while you live your life.
- Uses approval gates. Consequential actions pause for human permission—the trust primitive that makes always-on acceptable.
Why proactive is a category, not a feature
The reactive assistant caps out at the effort you invest in prompting it. A proactive agent compounds: it remembers what you care about, watches for changes, and spends idle cycles on your agenda. That is also why the field is crowded in 2026—Muse, Dots, and Grok Bot all pitch the same inversion. Anthropic's differentiation is the approval-gate design: in a research preview from the lab that popularized constitutional AI, the safety posture is the product, not just the model quality.
We have covered this shift before—see our earlier Claude Duet proactive personal agent deep-dive and the Claude Duet overview. This page focuses on the October 2026 research preview state and how to think about adopting it.
Claude Duet vs Muse vs Dots vs Grok Bot
| Agent | Pitch | Strengths | Watch-outs |
|---|---|---|---|
| Claude Duet | Always-on proactive agent, iOS/Android/web research preview | Daily brief, background cloud work, explicit approval gates | Research preview—availability and limits vary |
| Muse | Proactive companion agent | Strong consumer UX momentum | Different data and integration surface |
| Dots | Ambient task agent | Focused task-monitoring loop | Narrower scope than a general brief |
| Grok Bot | X-native proactive agent | Deep integration with the X firehose | Platform-bound value |
The honest answer for now: pick based on where your information and accounts already live, and trial the approval UX before trusting any of them with actions that cost money or send messages.
The supervision stack around an always-on agent
Duet makes approval flows mainstream, and a small ecosystem of tools has formed around supervising agents that act while you are not looking:
- Monish — lock-screen progress and tap-to-approve for coding agents; the same pattern Duet generalizes, for your own agent runs.
- Claude Code MCP Tasks & elicitations — how background agent work and structured questions work on the developer side of the same paradigm.
- Context Engine — shared memory across assistants, so a proactive agent inherits what your other agents already learned.
How to evaluate Claude Duet responsibly
- Start with read-only briefs—let it summarize and monitor before granting any action capability.
- Audit the daily brief for a week: is it surfacing what matters, or generating noise you learn to ignore?
- Map every approval gate to a real consequence—anything irreversible deserves a human, every time.
- Watch billing: background cloud work means usage you do not see happening. Understand your plan's accounting for it.
- Keep an exit path—export what it has learned about you before you depend on it.
Claude Duet FAQ
What is Claude Duet?
Claude Duet is Anthropic's always-on proactive agent, available as a research preview on iOS, Android, and web. Instead of waiting for prompts, it works in the background, sends you a daily brief, and asks for approval before taking consequential actions.
How is Claude Duet different from regular Claude?
Regular Claude answers when you ask. Duet initiates: it monitors tasks and information sources continuously, does background research in the cloud, and surfaces results on its own schedule with approval gates for actions it wants to take on your behalf.
Is Claude Duet free?
Duet ships as a research preview with access details published by Anthropic. Check the official getting-started article (support.claude.com) for current availability, plan requirements, and regional rollout status.
What are the approval gates in Claude Duet?
Approval gates are checkpoints where the agent pauses for human permission before executing consequential actions—sending messages, making purchases, or modifying data. This human-in-the-loop pattern is the trust mechanism for always-on agents.
Claude Duet vs Muse, Dots, or Grok Bot?
Muse, Dots, and Grok Bot are competing proactive-agent entries from other labs. The practical comparison axes are: what each can do without permission, how background work is billed, notification design, and how well each handles the daily-brief loop. Trial the one matching your existing account and data footprint.
AIXList is an independent AI tools directory. This article summarizes public October 2026 announcements and documentation for Claude Duet; we are not affiliated with Anthropic, X, or any other vendor mentioned. Research preview capabilities change frequently—verify details in the official support documentation.