Boring AI vs Claude Duet
Claude Duet is Anthropic's always-on agent, launched as a research preview on 21 October 2026. You message it in one ongoing thread, by typing or by voice. It sends a morning brief, checks your connected apps every few hours, works on its own virtual machine with its own browser, and asks before it sends, pays, signs, or deletes. Anthropic built it for professionals and small business owners, and for one person's work it is a strong default. Boring runs on Claude by default too, so this is the same model family doing a different job. Boring runs the work a company answers for: agents owned by a team, started where work arrives, and reviewed by the person assigned to decide, who can fix the proposed call before it runs. Use both. This page shows where the line sits.
Updated
If the work is yours, try Duet. It watches your inbox and calendar, drafts the next step, and stays quiet until something needs you. It runs on Claude Opus 5.5 in its own virtual machine, so your laptop and phone can be off. It reaches more than 6,000 connectors and plugins from the Claude Marketplace, and during the preview it is included in Claude Pro and Max. Anthropic gave it real guardrails: it shows you the exact action before it sends, pays, signs, or deletes; some actions are off-limits even with your approval; and it never directly sees your passwords or sign-in tokens. Boring answers a different question: who operates a workflow when the company has to answer for it. Its agents belong to a workspace rather than one person's Claude account. They start from webhooks, forms, app events, the workspace API, and schedules as well as from people. By default, each connected action waits for an assigned reviewer who can approve it, modify the exact call, or reject it, with a due date. Give Duet your inbox. Give Boring your team's runbook.
Boring AI vs Claude Duet, capability by capability.
| Capability | Boring AI | Claude Duet |
|---|---|---|
| // shape of the product | ||
| What it is | A console where a team builds, operates, and reviews automation agents. | A proactive agent in its own app, built, in Anthropic's words, "for professionals and small business owners". A research preview, separate from the main Claude app for now. |
| Who owns and directs it | A workspace: roles, private or workspace visibility, live co-editing, and versioned definitions where every change is a diff you can roll back. | One person, signed in with their Claude account, carrying over that account's memory, skills, and connected apps. Not on Team or Enterprise plans yet; sharing a Duet with teammates is not specified in Anthropic's public docs. |
| Where you talk to it | The console and workspace assistant (⌘J), plus named Slack requests, supported review/input forms, and same-agent thread follow-ups. Linked identities and agent opt-in apply. Native mobile approvals are built; store release is pending. | Its strength: one ongoing thread, typed or by voice, in the iOS and Android apps or at claude.ai, with separate channels by topic that share the same memory. |
| A computer of its own | No computer use or signed-in browsing. Agents act through connected tools and can load public web pages. | A virtual machine with its own browser, so your devices can be off. You can't yet watch or take over that browser, per Anthropic's help center. |
| // judgment and control | ||
| What waits for a person | New agents start with every connected action waiting for approval. Owners loosen it to writes only or auto-approve eligible actions they trust, such as reads; an action Boring can't classify counts as a write. | It reads, drafts, and prepares on its own, then shows what it's about to do and waits before it sends a message, makes a payment, signs a document, or deletes anything. If you can't approve right away, it finishes the rest of the task and brings the approval back later. |
| Who decides, and what they can change | An assigned reviewer in a shared Tasks inbox can approve, modify the proposed call, or reject it, and the agent's owner or an admin can reassign it. Reviews can carry due dates, reminders, and a configured overdue outcome. | The person whose Duet it is. Editing the proposed action before approving it, or routing it to another reviewer, is not specified in Anthropic's public docs. |
| Hard limits on actions | Exact-action grants are enforced in code: an agent without the send action cannot send, whatever its instructions say. | Some actions are off-limits even with your approval, such as trading securities or moving money between accounts. You can disconnect any app it can reach, and it never directly sees your passwords or sign-in tokens. |
| Record of the work | One shared trace per run: reasoning, tool calls, approvals, and cost. Tool payloads are kept 90 days; the outcome, output, and step list stay. Run summaries export as CSV or JSON. | A Tasks tab shows ongoing work and flags what needs your input, and only you can mark a task completed. An activity log, or team access to one, is not specified in Anthropic's public docs. |
| // starting, recovering, model, and price | ||
| How work starts | Schedules, webhooks, inbound email, public forms, SaaS app events, the Chrome extension, the workspace API, MCP clients, and other agents. | Your messages, a morning brief, and checks of your connected accounts every few hours, so something urgent can arrive between checks. A webhook, form, or API that starts work is not specified in Anthropic's public docs. |
| When something fails | Resume eligible failed runs from the last completed step; in a numbered step plan, when an action's outcome is uncertain, Boring blocks automatic repetition instead of guessing. Recognized provider failures are labeled, with the provider's reference ID. | Work continues in its virtual machine after you close the app, and it tells you when a task finishes or needs review. Retry or resume semantics are not specified in Anthropic's public docs. |
| The model | Claude Sonnet 5.5 by default, managed GPT, Gemini, and Grok with no key, or your own Anthropic, OpenAI, Google, OpenRouter, or compatible-endpoint key. | Designed around Claude Opus 5.5. In the help center's words, "You can't choose a model." |
| Price and access | Early access, on design-partner terms, with white glove if you want the agents built and operated with you. | A research preview by waitlist for US Pro and Max subscribers, at no extra cost. It shares your plan's usage limits, and background checks count toward them. Not on Team or Enterprise plans yet. |
Details about Claude Duet come from Anthropic's launch announcement, the Claude Duet product page, and the help-center article "Get started with Claude Duet", checked 21 October 2026, launch day. Duet is a research preview and will change; where Anthropic's public docs don't specify something, we say that rather than call it missing. If anything here is out of date, tell us and we'll fix it.
The differences that decide it.
A proactive agent for one person, by design.
Anthropic built Duet "for professionals and small business owners", and everything about it centres on one person. It carries over the memory, skills, and connected apps from your Claude account. It sends you a morning brief, follows up on quiet threads, and messages you only when something needs attention. Its Tasks tab tracks what it's working on for you, and only you can mark a task completed. For one person's inbox, calendar, invoices, and decks, that is a sensible and well-made design.
A company workflow has a different shape. The refund queue, vendor onboarding, and the Monday customer report each have several people who edit them, review what they did, and answer for the result. Boring puts agents in a workspace, with roles, private or workspace visibility, live co-editing, and one shared record per run, and every change to a definition is a versioned diff you can roll back. Duet isn't on Team or Enterprise plans yet, and Anthropic's public docs don't specify sharing a Duet with teammates. When that changes, we'll update this page.
Anthropic built real guardrails. The question is who uses them.
Duet reads, drafts, and prepares work on its own, then stops before anything consequential. Before it sends a message, makes a payment, signs a document, or deletes anything, it shows you what it's about to do and waits. Some actions, like trading securities or moving money between accounts, are off-limits even with your approval. Credentials follow the same design as Claude Managed Agents, so Duet never directly sees your passwords or sign-in tokens, and payments go through a virtual card so your real card number isn't shared. Anthropic adds classifiers that screen actions and sensitive information, defenses against prompt injection, external penetration testing, and a public bug bounty.
The difference is who gets asked and what they can do. Duet asks the person whose Duet it is. Boring routes the decision. New agents start with every connected action waiting for approval, and each request leads with the exact tool call and its scrubbed arguments. It goes to an assigned reviewer with a due date, reminders, and a configured overdue outcome of reject or skip, and that reviewer can approve, reject, reassign, or modify the exact arguments before the call runs. Editing a proposed action before approving it, or routing it to someone other than the owner, is not specified in Anthropic's public docs for Duet.
Where work arrives, and what happens when it breaks.
Duet starts from your messages, a morning review of your email and calendar, and checks of your connected accounts every few hours. Anthropic is candid that something urgent can arrive between checks and that Duet can misjudge what's urgent. That covers a great deal of personal work. A company's work also arrives as a webhook from another system, a form a customer submits, an event in a SaaS app, or a call to an API, and it often needs to start the moment it arrives. Boring starts runs from all of those, plus schedules, inbound email, the Chrome extension, MCP clients, and other agents.
When a Boring run breaks, its saved plan and progress stay, and eligible failed runs resume from the last completed step. In a numbered step plan, when a write's outcome is uncertain, Boring blocks automatic repetition instead of guessing. For eligible runs, provider throttling and outages are waited out within a bounded window; if the provider stays down, the run ends with its saved progress and a clear reason. Recognized provider failures are labeled billing, rate limit, or context size, with the provider's reference ID. Duet keeps working in its own virtual machine after you close the app and tells you when a task finishes or needs review; retry and resume semantics are not specified in Anthropic's public docs.
Same model family, different choices.
Duet is designed around Claude Opus 5.5, and Anthropic's help center says plainly that you can't choose a model. Its background work shares your Pro or Max plan's usage limits with the rest of Claude, so a busy month of check-ins and long tasks draws down the same allowance as your chats. For a personal agent in research preview, that is a clear and reasonable trade.
A team choosing where its operations run usually wants to set those terms itself. Boring runs new agents on Claude Sonnet 5.5 by default, offers managed GPT, Gemini, and Grok with no key, and accepts your own Anthropic, OpenAI, Google, OpenRouter, or compatible-endpoint key. Pin a different model for one agent or the whole workspace when a job calls for it, and see each run's cost in its trace. And we never train foundation models on your content.
Sometimes the answer is Claude Duet.
- The work is yours: your inbox, calendar, follow-ups, invoices, and the briefs, decks, and dashboards you'd otherwise build yourself.
- You want an agent that comes to you, with a morning brief and a message only when something needs your attention.
- You already pay for Claude Pro or Max in the US, and want it included at no extra cost during the preview.
- You want to reach your own apps through more than 6,000 connectors and plugins in the Claude Marketplace, and buy things through Link or Shop Pay with your approval.
- You want to message it in one ongoing thread, typed or by voice, on your phone or at claude.ai.
- The workflow belongs to the company: several people edit it, review its runs, and answer for what it did.
- Someone other than its author has to approve a step, with the power to change the proposed call before it runs, and a configured outcome if nobody decides by the due date.
- Work arrives as webhooks, form submissions, app events, or API calls, and has to start when it arrives, not at the next check-in.
- The team needs to choose the model and the keys: Claude Sonnet 5.5 by default, managed GPT, Gemini, or Grok, or your own provider key.
- You'd rather have the agents built and operated with you: white glove is half the offer.
Can I use Claude Duet and Boring together?
Yes, and that's the split we'd suggest. Give Duet your inbox: your morning brief, your follow-ups, the drafts and decisions only you make. Run what the company answers for through Boring: the queues, hand-offs, and recurring jobs that several people own, that start from other systems, and that need an assigned reviewer and a shared record. Both run on Claude, and the two don't need to be connected to be useful side by side.
Duet already asks before it sends, pays, signs, or deletes. Why isn't that enough?
For your own work, it often is. Duet shows you the action before it acts, keeps some actions off-limits even with approval, and never directly sees your passwords. Boring doesn't claim stronger enforcement than that. The difference is who gets asked and what they can do. In Boring, the request goes to an assigned reviewer with a due date and reminders, leads with the exact tool call, and can be approved, rejected, reassigned, or modified before it runs, and the decision stays in a run record the team can review.
How is Claude Duet different from Claude Cowork?
Both are Anthropic products for one person. Anthropic describes Duet as a proactive agent in its own app: it checks your connected apps on its own, sends a morning brief, and messages you when something needs attention. Cowork is how you hand Claude a task and come back to the finished work. Boring sits beside both: it runs the workflows a team owns. We compare Boring and Cowork on its own page.
Which models do they run on?
Anthropic designed Duet around Claude Opus 5.5, and its help center says you can't choose a model. Boring runs new agents on Claude Sonnet 5.5 by default, offers managed GPT, Gemini, and Grok with no key, and accepts your own Anthropic, OpenAI, Google, OpenRouter, or compatible-endpoint key. Pin a different model for one agent or the whole workspace.
Can I talk to Boring by voice, the way I message Duet?
Boring does not offer voice conversations. Linked teammates can work with enabled agents in Slack: address a published agent by name in a DM or @mention, answer supported questions, review actions, and follow up in the same thread after a run finishes. Those requests use the same permissions and approval rules as other runs. The console and its workspace assistant (⌘J) remain available for managing the work and reviewing proposed changes. A native mobile companion for approvals is built, with its store release pending.
Is this comparison fair?
We've tried to make it so. Every fact about Duet comes from Anthropic's own announcement, product page, and help center, checked on 21 October 2026, the day it launched. We left out details we could only find in press coverage, and where Anthropic's docs don't specify something we say exactly that rather than call it missing. The page also says plainly where Duet wins: it comes to you, it has a computer of its own, it reaches thousands of connectors, and it's included in plans many people already have. Duet is a research preview and will change quickly. If we got something wrong, tell us and we'll fix it.
Let Duet watch your inbox. Bring us your team's runbook.
Request access and describe it in a sentence — or ask about white glove and our team will build and run it with you.