What Anthropic's Claude Duet agent means for marketing teams

Anthropic opened a US waitlist for Claude Duet on 21 October 2026, a proactive agent that briefs you each morning, works through more than 6,000 connectors and plugins on its own cloud computer and asks before it sends, pays, signs or deletes anything. For marketing leaders, the first decision is which company apps staff may connect to it.

Key takeaways

• Anthropic announced Claude Duet as a research preview on 21 October 2026, available by waitlist to US Claude Pro and Max subscribers only.
• Duet runs on Claude Opus 5.5 on its own virtual machine with its own browser, so it keeps working when your laptop and phone are off.
• Anthropic names HubSpot, Canva, LinkedIn, Shopify, Gmail and Slack among more than 6,000 connectors and plugins Duet can use.
• Duet shows what it will do and waits for approval before it sends a message, makes a payment, signs a document or deletes anything.
• Team and Enterprise plans are not included yet, and Duet’s background checks count toward the same usage limits as the rest of Claude.

What Anthropic launched on 21 October

Anthropic’s announcement describes Duet as a separate app, on iOS, Android and claude.ai, where Claude works without waiting for a prompt. Each morning it sends a short brief on the most important items across your connected apps. During the day it checks those apps every few hours and messages you only when something needs you. You talk to it in one ongoing thread, by typing or by voice, and it carries over the memory, skills and connectors from your existing Claude account.

Anthropic lists jobs well beyond a chat reply, such as scheduling meetings, building dashboards from dozens of tools, sales and financial forecasts, and prototyping web apps, and it delivers finished documents, sheets and decks for review. 9to5Mac’s write-up adds that a Tasks tab tracks work in progress and only the user can mark a task finished.

Duet reads, drafts and prepares work on its own, then stops for approval before anything is sent, paid, signed or deleted. Trading securities and moving money between accounts are off limits even with approval, and Anthropic says Duet never directly sees passwords or sign-in tokens.

Which marketing jobs could it take?

Several of Anthropic’s own examples are marketing work. It describes Duet drafting HubSpot follow-ups, making social campaigns in Canva, posting a LinkedIn job once approved and setting up Shopify stores. One of its testers, the brand strategist Mya Dalton, had Duet prepare a brief for a client call and offer to move the call when it noticed the slot fell right before her flight.

How does Duet compare with dots?

OpenAI got there first. Its dots, which we covered in our post on dots and ChatGPT plugins, launched at DevDay on 29 September 2026.

A chat assistant waits for someone to open it, while Duet decides when to message you, so a marketing team should test its judgement on what to flag and what to leave alone before anything else.

Why the plan type matters most

Duet is on individual Pro and Max plans only. Team and Enterprise, the plans a company buys and runs centrally, have to wait. So the first people in a marketing team to use Duet will be on individual accounts that no company admin controls, connecting it to the company’s Gmail, Slack and HubSpot with their own logins.

Our view at Manual Focus is that marketing leaders should treat Duet as a personal tool until Anthropic ships it for Team and Enterprise, and should decide now which company systems a personal agent may connect to, so the answer comes from a policy and not from a surprise in an audit log. Anthropic’s approval list also leaves gaps for marketers. It covers messages, payments, signatures and deletions, and says nothing about publishing a post, editing a live web page or changing a HubSpot list. Write rules for those yourself, using the approach in our brand guardrails as code playbook.

Duet’s background checks also spend the same allowance as everything else a person does in Claude, and 9to5Mac quotes Anthropic warning that users may hit limits “sooner than with chat alone”. Our post on the hidden cost of AI adoption covers the running overheads teams underestimate once AI tools are in place.

What to do before access reaches your team

1. Ask your admins about personal connections. Find out whether staff can connect third-party apps to Google Workspace, Slack and HubSpot with their own accounts, and who gets told when they do.
2. Write the missing approval rules. List the marketing actions Anthropic’s approvals don’t name, such as publishing, editing live pages and changing audience lists, and decide which need a second person.
3. Choose one recurring job to pilot. The Monday performance brief is a good start, with someone reading every output for two weeks. Our daily briefing pipeline playbook sets out what a good brief contains, and our evaluation frameworks playbook shows how to score one.
4. Agree who pays for seats. Decide whether the company buys Max plans for named people or bans expensed personal plans, before a US colleague or agency partner gets an invitation.
5. Run the same job through dots. Give both agents one identical brief and compare what each flags, using the checks in how to evaluate AI tools without falling for the demo.

What Anthropic hasn’t said

Anthropic gives no date for the UK or any other market, and no timetable for Team and Enterprise plans. It doesn’t say how much of a plan’s allowance one background check uses, or whether a company admin will be able to see what a staff member’s Duet has done. Anthropic says it will publish more safety design detail in the coming days. Its own list of limits says Duet may misread a thread or misjudge urgency, can miss something that arrives between check-ins and can’t yet show you its browser.

Further reading

• What OpenAI’s dots and ChatGPT plugins mean for marketing teams, on the rival agent and the rules it needs
• Brand guardrails as code, for turning brand rules into checks an agent can follow
• Daily briefing pipeline, on what a useful morning brief should contain
• The hidden cost of AI adoption, on the running overheads teams underestimate
• How to evaluate AI tools without falling for the demo, for testing an agent on your own work

If you want help setting agent rules for your marketing team, our services page explains how we work.

I run marketing at [company]. We're preparing for staff using personal AI agents
that can act in our tools on their own judgement.

Attached is a list of the tools my team uses, with the actions each person
takes in each one. Publishing a social post is one example. Changing a
HubSpot list is another.

For every action, draft one approval rule using one of these options:
- the agent may do it alone
- the agent prepares it for a person to approve
- the agent must never do it
Name the role who approves.

Give me a table with the columns: Tool | Action | Agent may... | Approver |
Reason.

Flag any action where you're unsure of the risk instead of guessing.

Set this up as a routine that runs every Monday for two weeks.

Each Monday I'll paste in the performance brief an AI agent wrote for my
marketing team, plus the raw numbers it was based on.

Score the brief against this checklist:
- Did it flag the changes a marketing lead would act on?
- Did it miss anything important in the numbers?
- Is every figure traceable to a source?
- Did it raise anything that didn't need attention?

Give me a table with the columns: Date | Flagged correctly | Missed |
Untraceable figures | False alarms | Score out of 10.

After the second week, tell me in two sentences whether the brief has earned
enough trust to stop checking it line by line.

I gave two AI agents the same brief: [paste the brief].

Here is what the first agent produced: [paste output A].
Here is what the second agent produced: [paste output B].

Compare them. List what each flagged and what each left out. Mark any claim
that isn't supported by the brief or the data.

Give me a table with the columns: Item | Agent A | Agent B | Which is right?

End with the three differences that matter most for a marketing team. Don't
pick a winner on style alone. Say what you couldn't check
rather than guessing.