AI Firm Walls Off New Work Bot In Phone App For 2 Paid Plans
Anthropic opened Claude Duet in research preview for Pro and Max subscribers, offering a mobile-only proactive agent app that automates ongoing paid work and advances its move toward paid workflows.
Key Takeaways
- Anthropic opened Claude Duet in research preview for Pro and Max subscribers
- Product commentator Casey Newton said the Claude Duet app is mobile-only and separate from the main Claude experience
- Claude Duet is a proactive agent app intended to automate ongoing work for paying users
- Anthropic added monthly API credits to Max and Team plans on Oct. 7 and folded Cowork features into ordinary conversations on Sept. 16
Anthropic posted the change in its Claude app release notes, saying the feature is available to Pro and Max plans and linking users to a separate product post.
Product commentator Casey Newton said the app is mobile-only and separate from the main Claude experience.
Claude Duet Starts Outside The Main App
That packaging matters.
Phone App Creates Friction
Claude Duet only becomes useful when it can sit where work already happens, across desktop, browser and phone. A separate phone app creates friction at the exact point Anthropic is trying to prove a new behavior.
Users must decide not just whether Claude can act for them, but where that action lives and how often they need to switch into it.
Anthropic has been widening Claude beyond chat for months. September brought Claude Cowork into every conversation, and plugin submissions followed soon after.
The new phone app pushes further by treating the assistant less like a prompt box and more like software that can keep working after the first request.
Shift Toward Paid Workflows
The release is the latest step in Anthropic’s move from model access toward packaged labor. Oct. 7 updates added monthly API credits to Max and Team plans, while Sept. 16 changes folded Cowork features into ordinary conversations.
That sequence shows a product line moving toward paid workflows, not just better answers.
Pro and Max users already buy more capacity and better model access, and this adds a second pitch, that subscribers are paying for ongoing action rather than a faster chatbot.
Also Read: Zoom Unlocks AI That Turns Meetings Into Tasks After 2 Years
The commercial logic is clear. Consumer AI plans have become crowded, and labs need reasons to justify premium tiers.
A dedicated agent product can support higher pricing if it saves users real time on repeat tasks that do not need a new prompt each time.
Anthropic sells access to Claude through subscriptions and APIs, so interface design becomes a revenue question as well as a product detail. If the agent stays trapped in a separate phone app, the company may learn that users like the idea of delegated work but resist moving it outside the main workspace.
The distinction between model updates and agent products is important.
Model releases compete on speed, reasoning and price. Agents compete on trust, persistence and handoff.
If users cannot see what the agent is doing, or cannot reach it from their real work tools, adoption can stall even if the underlying model is strong.
How Agents Differ From Chatbots
A proactive agent differs from a chatbot because it keeps a task alive after the first request. In a normal Claude chat, a user asks for research, code or writing help and gets responses inside a thread.
In this phone app, Anthropic is testing a product where the assistant can take responsibility for a continuing job, useful for consultants, founders and other paid users who want fewer status checks, fewer repeated prompts and more visible task progress.
Claude Duet aims to deliver that kind of ongoing task ownership for subscribers.
Phones are good for alerts, approvals and quick handoffs. Laptops remain where most long documents, spreadsheets and internal tools live.
The new app can prove strong for triage and follow-up, but Anthropic still has to show whether it can become a full work surface rather than a sidecar.
Read Next: Anthropic Unleashes AI Attack Bots To Test If One Breach Spreads
