Anthropic just opened a waitlist for Claude Duet, an AI agent that watches your work apps and taps you on the shoulder when something needs you. Here is what it does, what it can't do yet, and the simple rules I use before any AI gets to interrupt my day.
By Warren Schuitema, AI Systems Consultant, Matchless Marketing / Matchless AI
Published: October 24, 2026 | Last Updated: October 2026
Every two minutes.
That's how often the average worker gets pulled away by a meeting, an email, or a chat message, according to Microsoft's research. About 275 times a day.
Now picture what that looks like when you run a business from your kitchen table and the school app is on the same phone. Most days, whatever pings the loudest runs your schedule.
So, should you let a proactive AI agent like Claude Duet decide what needs your attention? Yes, with rules. A good agent can read the noise for you and bring you only the few things that need your call. But it should only bring you things. You still make the decisions. You set what it watches, when it can reach you, and what it must wait on. Skip those rules and you just traded one noisy inbox for a louder one.
That question got real this week. On October 21, 2026, Anthropic opened a waitlist for Claude Duet. TechCrunch called it "Anthropic's bet on agents." It's built for professionals and small business owners, and it doesn't wait for you to ask. It comes to you.
Key Takeaways
- Claude Duet is a new Anthropic app, in research preview for U.S. Claude Pro and Max subscribers, that watches your linked apps and brings you what needs your input.
- Duet shows you its planned step and waits for approval before it emails anyone, spends money, signs paperwork, or deletes data.
- The average worker in Microsoft's research gets interrupted every two minutes, about 275 times a day. A good agent should cut that number.
- Before you turn on any proactive agent, write three rules: what waits for a daily brief, what earns a same-day ping, and what stops everything.
- Test any agent the same way I test every system: does it give you back Time, Money, or Energy?
Your Day Is Already Run by Whatever Pings Loudest
You sit down to do the one thing that grows the business. Write the proposal. Call the warm lead. Then the email chime goes off. Then a text. Then the school app.
The problem is they all sound the same. A customer with a real problem and a coupon from a store you shopped at once get the same buzz. So you check every one, and every check costs you the thread you were holding.
I spent 16 years in manufacturing and supply chain. On a production floor, nobody treats every signal the same. A machine running a little warm and a machine on fire do not get the same alarm. One goes on the shift report. The other stops the line. That difference is the whole job of a good supervisor.
Most small business owners don't have a supervisor for their attention. They have a phone.
That's why the newest AI agents matter. For the last couple of years, AI waited for you to type a question. The new wave flips that. These agents watch your accounts in the background and come to you when they think something matters.
That sounds great. It also sounds like one more thing that buzzes.
So skip "Is this agent smart?" for a minute. The better question is "Who decides what's worth my attention?" If you don't answer that, the agent will answer it for you.
Here Is What Claude Duet Actually Does
What it is. TechCrunch reported on October 22, 2026, that Claude Duet is a new Anthropic app aimed at professionals and small businesses. Anthropic pitches it as an agent for work. TechCrunch noted it's weaker at personal errands like booking tickets. It runs on Claude Opus 5.5.
What it does. According to BetaNews, Duet watches the apps you link to it and picks up work without waiting for a prompt. It gives you a short morning rundown of the top items waiting across your accounts. Through the day, it keeps watching and contacts you only when something needs your input. You can also hand it tasks, like scheduling meetings or building a dashboard that pulls numbers from your tools.
Where it works. Duet runs on its own cloud computer with a built-in browser, so tasks keep going when your laptop is closed. It works on iPhone, Android, and the web, and it brings over your memory, skills, and app connections from Claude.
What it won't do without you. This is the part I care about most. BetaNews reported that before Duet emails or messages anyone, spends money, signs paperwork, or deletes data, it shows you the planned step and waits for your yes. Trading stocks and moving money between accounts are blocked outright, even if you say yes. Anthropic's own post put it simply: "the important decisions stay with you."
What it admits. Anthropic says the agent "still might misread a thread or miscalibrate what's urgent." To save on usage, Duet checks your linked services every few hours, and its support pages warn that "something urgent can arrive between checks."
Who can get it. It's a research preview. It's open to U.S. subscribers on Claude Pro or Max through a waitlist, with invites going out in batches. Team and Enterprise accounts are left out for now. There's no extra charge during the preview, but Duet uses the same plan limits as your other Claude tools, and its background checks count against them.
Duet has company. BetaNews noted that Meta launched Muse on September 8 and OpenAI launched dots inside ChatGPT on September 29. Proactive agents are turning into a whole category.
So the tool is real, it's early, and it was built with a stop sign in front of the risky stuff. Good.
The Numbers Say Attention Is the Real Bottleneck
We're already interrupted constantly. Microsoft's 2025 Work Trend Index special report, "Breaking down the infinite workday," combined Microsoft 365 usage signals with a survey of 31,000 workers across 31 markets. It found the average employee is interrupted every two minutes by a meeting, email, or chat, which adds up to about 275 times a day. The average worker also gets 117 emails a day.
Any new tool that adds to that pile is working against you, no matter how smart it is.
Small businesses are already using AI. The U.S. Chamber of Commerce's 2025 Empowering Small Business report found that 58 percent of small businesses say they use generative AI. That's up from 40 percent in 2024 and 23 percent in 2023. Most of us already have AI in the building. The next step is deciding how much it does without being asked.
Agents are coming fast, and many projects will fail. Gartner predicts that at least 15 percent of day-to-day work decisions will be made on their own by AI agents by 2028, up from zero in 2024. Gartner also predicts that over 40 percent of agent projects will be canceled by the end of 2027, because of rising costs, unclear business value, or weak risk controls.
Many "agents" aren't really agents. Gartner also warned about "agent washing," where vendors rebrand old chatbots and tools as agents. Its estimate: only about 130 of the thousands of agent vendors are real.
Put those together. We're drowning in pings. Most of us already use AI. Agents will make more small calls for us. And a lot of agent projects will fail from fuzzy goals and loose controls.
The owners who win with proactive agents will be the ones who decided, ahead of time, what the agent is allowed to bring them.
A Good Agent Works Like a Shift Lead
Here's the reframe I want you to keep.
A smoke alarm screams the same for burned toast and a house fire. A good shift lead works differently. On a plant floor, the shift lead walks the line all day. Most of what they see never reaches the plant manager. Small stuff gets fixed or written on the handoff report. Bigger stuff gets a phone call. And a few things, a safety problem or a quality defect heading out the door, stop the line right now. In lean plants, that's what the andon cord is for. Anyone can pull it, and everyone knows what it means.
The shift lead's value is in the sorting. They protect the manager's attention for the calls only a manager can make.
That's the job a proactive AI agent should do for you. Your inbox already "notices" everything. What you need is something that sorts it and holds most of it back.
So before I let any agent reach me, I give it three buckets. I call them Brief, Ping, and Stop.
- Brief. Things that can wait for one daily summary. New leads from overnight. A supplier invoice due next week. A new review. The agent gathers these and hands them to me once, at a time I pick.
- Ping. Things that need me today, but not this minute. A client asking a question before a deadline. A payment that failed. A booking change for tomorrow. These earn one message, inside my working hours.
- Stop. The short list of things that justify breaking into my day or evening. A client in real trouble. A security alert on a business account. A family emergency through a channel I trust. This list should be tiny. If everything is a Stop, nothing is.
Notice what's missing from all three buckets. Decisions. The agent sorts, drafts, and gathers. I decide.
Duet already has the shape: a morning rundown, a nudge when something needs input, and a pause before anything that sends, spends, signs, or deletes. Brief, Ping, and Stop are how you fill it in for your business.
How My AI Workforce Decides When to Bother Me
I run my business this way. Matchless runs on a workforce of AI agents. They handle a lot of the day-to-day work: research, first drafts, notes, content prep, and tracking open items. This article came out of that same system. An agent researched it and drafted it, and it sits waiting for my review before it goes anywhere.
Two rules keep that workforce from running me.
Rule one: draft first, and I'm the only approval gate. Nothing sends, posts, invoices, books, buys, signs, or spends on my behalf until I say so. My agents can work fast on everything up to that line. Then they stop and wait.
Rule two: my agents respect my quiet hours. These are the hours when I don't want to be alerted, and I wrote them into my system as calendar guardrails. No notifications between 4:45 and 7:00 in the morning. None after 5 PM on weekdays. None on weekends. If something urgent comes up during those windows, it leads the next morning brief.
That second rule is Brief, Ping, and Stop in real life. The morning brief is the Brief bucket. The working day is when Pings are allowed. And the bar for breaking my quiet hours is set on purpose.
Does that mean I sometimes see something a few hours later than I could have? Yes. I'm fine with that trade. My evenings and weekends stay with my family.
That's Energy I get back. And I got it by telling my agents when I'm off limits.
My post from yesterday on whether you should give an AI agent the passwords to your business accounts covers what an agent should be allowed to touch. This post covers when it's allowed to reach you.
The Same Rule Works at the Kitchen Table
This idea started at home for me.
Our Digital Wall Calendar replaced a $50 a month subscription. But the bigger win was what it ended: the nightly "what does tomorrow look like?" conversation. Now the calendar is just there. Everyone sees the day at a glance. Nobody has to ask.
That's a Brief. One place, one look, no pinging.
Same with the Weekly Meal Planner. It replaced HelloFresh in our house and ended the Sunday planning spiral. The planner does the sorting ahead of time. We make the final call.
In both cases, the tool took a pile of small questions and handled them in one calm place. That's the bar I hold any agent to. Here's the test I use on every system:
Time. Does the agent cut the number of times you check things? If you still open your inbox 30 times a day "just in case," it isn't saving time yet.
Money. Does it catch the things that cost you, like a missed lead or a late invoice, without spending anything on its own? Duet can't move money between accounts, and it asks before it spends. Keep it that way in any tool you use.
Energy. Do you trust it enough to stop watching? An agent you have to babysit is just another tab.
Is Duet Worth Joining the Waitlist For?
Based on what's been reported so far, here's my take.
Join the waitlist if you already pay for Claude Pro or Max, you live in your email and calendar, and you're tired of being the human filter for your own business. The morning rundown alone could be worth it. There's no extra charge during the preview.
Wait if you need something that catches emergencies in real time. Duet checks in every few hours, and Anthropic itself says urgent things can land between checks. Keep your true Stop items on a channel you control, like a phone call or a text from a short list of people.
Go slow either way. It's a research preview, and Anthropic is still tuning its timing based on tester feedback. Start with a few linked apps.
If you don't use Claude, the same thinking applies to OpenAI's dots, Meta's Muse, or whatever lands next. The tools will change. Your rules shouldn't. For a wider look at what agents can handle today versus what's still hype, see my earlier post, Do AI Agents Actually Handle Your Everyday Tasks, or Is That Still Hype in 2026?
7 Steps to Set Up a Proactive AI Agent Without Losing Control
These work for Claude Duet and any agent that reaches out to you.
- Count your pings for one day. Before you add anything, tally every time your phone or computer pulls you away. You need a starting number to tell if the agent helps or hurts.
- Write your Stop list first. Grab a sheet of paper and write the three to five things that truly deserve to break into your day. Keep it short. Everything else goes in Brief or Ping by default.
- Pick your Brief time and your quiet hours. Choose one time for your daily summary, like 7:30 a.m. with coffee. Then set your quiet hours, when the agent can't alert you. Mine are 4:45 to 7 a.m., after 5 p.m. on weekdays, and all weekend.
- Link only two or three apps to start. Start with email and calendar. Leave your bank, store, and social accounts out for the first two weeks. Add more once you trust how it sorts.
- Tell the agent your rules in plain words. Duet lets you choose which topics it raises and how often it reaches out. Type your three lists right in. Something like: "Only message me during the day for client deadlines and failed payments. Everything else goes in my morning summary."
- Keep sending, spending, and signing behind your yes. Let the agent draft the reply, prep the invoice, or fill in the form. You hit send. Duet already asks before it emails, spends, signs, or deletes. Don't turn that off in any tool you use.
- Grade it after two weeks. Count your pings again for one day. Compare it to your first number. Then ask: did it give me back Time, Money, or Energy? If the answer is yes, add one more app. If not, cut its access back or turn it off.
Frequently Asked Questions
What is Claude Duet?
Claude Duet is a proactive AI agent app from Anthropic, opened as a research preview on October 21, 2026. It watches your linked work apps, sends a short morning rundown, and contacts you when something needs your input. It runs in the cloud on iPhone, Android, and the web, and it asks before it sends, spends, signs, or deletes.
Who can use Claude Duet right now?
Claude Duet is available through a waitlist to U.S. subscribers on Claude Pro or Max plans, with invites sent in batches. Team and Enterprise accounts are not included yet. There is no extra charge during the preview, but Duet uses the same plan limits as your other Claude tools, and its background checks count against them.
Can a proactive AI agent make decisions for my business?
A proactive AI agent can sort information, draft replies, and prepare tasks, but you should keep decisions for yourself. Claude Duet shows its planned step and waits for approval before emailing anyone, spending money, signing paperwork, or deleting data. It blocks stock trades and transfers between accounts entirely. Treat any agent as a helper that prepares, while you approve.
Will an AI agent catch urgent problems the moment they happen?
Not always. Claude Duet checks your linked services every few hours to save on usage, and its support pages warn that something urgent can arrive between checks. Anthropic also says it may misjudge what is urgent. Keep true emergencies on a channel you control, like a phone call or text from a short list of trusted people.
What is the difference between an AI assistant and a proactive AI agent?
An AI assistant waits for you to ask a question or give a task. A proactive AI agent watches your accounts in the background and comes to you when it thinks something needs attention. That makes proactive agents more useful as filters, and it means you need clear rules about what they watch and when they reach you.
Let the Agent Watch. You Decide.
Back to that number. Every two minutes.
Most of those pings aren't bad on their own. The trouble is that nothing stands between them and you. Your phone treats a coupon exactly like a client in trouble, and it makes you do the sorting all day.
That's the promise of a tool like Claude Duet. Something can finally stand in that gap. It can read the pile, hold back the noise, and bring you the handful of things that need you.
But only if you tell it what "need" means.
So write your Stop list. Pick your quiet hours. Let the agent build the Brief and hold the Pings for working hours. Keep every send, every purchase, and every signature behind your own yes.
Then run the same test I run on everything. Did it give you back Time, Money, or Energy?
If it did, keep it.
If it just found a new way to buzz, take its voice away. That's A Better Way.
Keep Learning
- Should You Give an AI Agent the Passwords to Your Business Accounts? What 1Password for Agents and the bank crackdown mean, and how to decide what an agent can touch.
- Do AI Agents Actually Handle Your Everyday Tasks, or Is That Still Hype in 2026? A clear look at what AI can take off your plate today and what still needs you.
- What the AI Labs' Agent Security Crisis Means for Your Business How to audit what your AI agents can reach and where to add a human check.
- Will AI Shopping Agents Send Customers to Your Small Business, or Straight to Amazon? What AI agents look for when they shop, and how to get picked.
About the Author
Warren Schuitema is an AI Systems Consultant with Matchless Marketing and Matchless AI. He spent 16 years in manufacturing and supply chain, where he learned that not every signal deserves the same alarm. He runs his business day to day with a workforce of AI agents that draft first, respect his quiet hours, and wait for his approval before anything sends, spends, or signs. At home, his Weekly Meal Planner replaced HelloFresh and his Digital Wall Calendar replaced a $50 a month subscription. He helps small business owners and parent entrepreneurs build practical AI systems that give back Time, Money, or Energy, and he is a partner in the Goal Achievers Summit.