Claude Duet Wants to Be Your Chief of Staff: What Is Left for Indie Apps?
Claude Duet is Anthropic's research preview of a proactive agent that watches connected apps and acts with your approval. It makes thin assistant wrappers harder to sell, but it leaves real room for vertical depth, better approval flows and connectors.
Key takeaways
- Claude Duet is a standalone app that checks connected accounts every few hours and sends a morning brief; it is a research preview on a waitlist for US Pro and Max subscribers.
- It runs on Claude Opus 5.5, uses the same Pro and Max usage limits and does not let users pick a model.
- The thin layer of an AI assistant that summarises email or calendar is now a feature Anthropic ships itself.
- The openings are vertical workflows, approval UX and distribution through the connector marketplace.
- Proactive agents raise the stakes on permissions and audit trails, which is product work an indie team can own.
Claude Duet is Anthropic's research preview of a proactive agent, announced on 21 October 2026. It watches connected apps, sends a short brief each morning and acts on your behalf after you approve the action. For indie founders, the question is which parts of the assistant market it closes and which it opens.
TL;DR
- It is a standalone app on iOS, Android and claude.ai, included in Pro and Max plans.
- It runs on Claude Opus 5.5, users cannot choose a model, and the preview is a waitlist for US Pro and Max subscribers, not Team or Enterprise yet.
- Its approval step is the important design choice: it shows the action and waits before sending, paying, signing or deleting.
- A generic assistant wrapper is now a weak business, but vertical workflows, audit trails and connectors are still open.
Disclosure: ZeroRaised publishes this analysis and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 26 Oct 2026; ZeroRaised has no commercial relationship with the companies named.
Proactive agent is a system that monitors connected data and starts work without waiting for a prompt, then asks for approval before taking consequential actions.
What Claude Duet actually does#
Anthropic's launch post says: "Today, we're opening a research preview of Claude Duet, a new app where Claude works as a proactive agent" (Anthropic). It is a standalone app on iOS, Android and claude.ai, included in Pro and Max plans and using the same usage limits. It runs on Claude Opus 5.5, users cannot pick a model, and the preview is available by waitlist for US Pro and Max subscribers. Support documentation says it is not on Team or Enterprise plans yet (Anthropic Support).
Each morning it sends a short brief across connected apps. It has a virtual machine with its own browser, so it keeps working when your laptop or phone is off, and it checks accounts every few hours because, in Anthropic's words, checking nonstop would use up the plan's limits much faster. It can use over 6,000 connectors and plugins, with examples like unpaid invoices in QuickBooks or Stripe, payroll in Gusto, HubSpot follow-ups and a Shopify store. It can buy through Link, Stripe's digital wallet, with a virtual card after you approve the spend (Anthropic).
The approval rule is the most important design choice. Before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it will do and waits. Some actions are off limits even with approval, such as trading securities or moving money between accounts (Anthropic). That is the line between an assistant that drafts and an agent that acts.
Anthropic admits the gaps: it might misread a thread or miscalibrate urgency, personal errands like booking tickets are not as strong, it cannot yet show you its browser and it lives in a separate app from main Claude (Anthropic). Those limits map what is still unclaimed.
Why thin assistant wrappers are a hard sell#
If your product reads email, summarises a calendar and drafts replies, Duet now does that inside a Pro or Max subscription many users already pay for. The wrapper is not competing with a better model. It is competing with a bundled feature. A standalone assistant must cover model costs; Duet's marginal cost to a subscriber is the plan they already bought. If you price an AI feature, how to reduce LLM costs is the relevant homework.
Where indie founders can still win#
Vertical depth. Duet can check QuickBooks for unpaid invoices, but it is not an accounts receivable system. It can draft a HubSpot follow-up, but it is not a pipeline with your team's stages, discount rules and handoff logic. The more a workflow depends on domain rules and clean records, the less a general agent can own it.
Audit trails. Duet's approval is basic: show the action, wait, then act. Regulated buyers need more: who approved what, when, under which policy, with what evidence, and how to reverse it. That is product work, and it is hard to bundle.
Connectors. Duet can use over 6,000 connectors and plugins in the Claude Marketplace (Anthropic), and Anthropic's help centre says it supports plugins such as a small business plugin (Anthropic Support). If your product is where work already happens, a connector puts you inside the agent's loop. The guide to monetising an MCP server covers the commercial side.
Approval UX is the product, not a checkbox#
Approval is where trust is won or lost. Anthropic shows the action and waits before sending, paying, signing or deleting. That is not a popup; it is a compact contract with the user. Indie products can go further: show the source of every claim, let the user edit before it runs, make the audit log exportable, set policies per vendor and amount, and let a manager approve a junior's queue. Buyers pay for this when money is involved.
An agent with a browser, a wallet and thousands of connectors is a larger attack surface than a chat box. The vibe-coded app security checklist covers the basics, and the PHP2Shell disclosure is a reminder that an agent inherits the security of everything it can touch.
Distribution: build for Duet, do not fight it#
Treat Duet as a channel. If the agent can use your product through a connector, your product becomes part of a workflow a subscriber already trusts. That is the same logic as building for any platform: go where the attention and the payment relationship are.
The catch is dependency. Platform features change and the platform can ship your feature. Keep your own login, data, billing and a reason to return directly. The API marketing guide covers discoverability without owning the whole surface.
AIMap's weekly roundup says all three frontier labs now have proactive personal agents: Meta's Muse, OpenAI's Dots and Anthropic's Duet (AIMap). This is a category forming, not a single launch, which means the horizontal assistant layer will keep getting cheaper and more bundled.
What this means for bootstrapped founders#
- Do not start with a generic assistant. Start with a job that has rules, records and consequences.
- Pick a vertical where the data model is the moat, not the prompt.
- Treat approval, audit and permissions as features, not overhead.
- Build a connector if buyers live in Claude, but keep your own billing and login.
- Watch the gaps Anthropic admits to. That is where a small team can be sharper.
How to launch an AI app and how to price a SaaS product cover the build and pricing side.
Where ZeroRaised fits#
ZeroRaised is a launch and discovery site for bootstrapped products with no institutional VC or accelerator equity. If you are building a vertical agent, an approval layer or a connector, submit your product. The pricing page explains the options, and joining the ZeroRaised newsletter is the simplest way to follow what other makers ship.
No launch guarantees traffic. A listing gives buyers a clear page about the product.
Sources#
Frequently asked questions
Is Claude Duet available to everyone?
No. It is a research preview available by waitlist, starting with US Pro and Max subscribers. Anthropic's support documentation says it is not on Team or Enterprise plans yet.
Which model does Claude Duet use?
It is built around Claude Opus 5.5, and users cannot choose a different model. The app uses the same usage limits as the rest of a Pro or Max plan.
What can Claude Duet do without asking?
It checks connected accounts and prepares work, but before it sends a message, makes a payment, signs a document or deletes anything it shows what it will do and waits for approval. Some actions are off limits even with approval, such as trading securities or moving money between accounts.
Should I build an AI assistant startup now?
Only if the value is in something Duet does not do well yet: a specific vertical, a workflow with real operational rules, or a permission and audit experience your buyers trust. A generic assistant that summarises email is a weak starting point.
Can indie apps get distribution through Duet?
Possibly. Duet can use over 6,000 connectors and plugins in the Claude Marketplace, and Anthropic says it supports plugins such as a small business plugin. Building a connector is a distribution route, not a guarantee of usage.