I checked my email on a normal Tuesday morning last month and counted 41 unread messages before I had finished my tea. Newsletters, a delivery update, two invoices, a thread from a client that had gone quiet, and a calendar invite I had already missed. Nothing in that list was hard. It was just a lot, and it was all waiting for me to go looking for it.
That is the exact problem behind the biggest AI shift of this autumn. For years, a chatbot sat there and waited. You typed, it answered, and the conversation ended. In the last two months, the three biggest AI companies have all shipped something different: an agent that comes to you first. It checks your apps, spots what needs attention, and keeps working on jobs after you close the app.
Meta went first with Muse on September 8, 2026. OpenAI followed with Dots on September 29. Anthropic opened its research preview of Claude Duet on October 21. On paper they sound identical. In practice they are aimed at different people, cost very different amounts, and are available in very different places. I spent the last week reading the launch documents, the pricing pages, and the early hands-on coverage to sort out which one, if any, is worth your attention right now.
One honest note before I start. I have not been able to try Duet hands-on yet, and that is not a small detail. It is a US-only waitlisted preview for paying Claude subscribers, and I am not in the US. The same goes for parts of Dots and Muse, which are also region-limited at launch. So this is not a hands-on review where I pretend I clicked every button. It is a careful comparison of what each company has actually shipped, what it costs, and where it falls short, so you can decide whether to join a waitlist, pay, or simply wait.
So what changed? Chatbots waited. These agents do not.
A normal chatbot is reactive. You ask a question, it answers, and then it forgets the job exists until you come back. A proactive agent flips that around in three specific ways.
First, it checks on its own. Duet looks across your connected email and calendar each morning and sends a brief of what matters most. After that it checks every few hours and stays quiet unless something needs you. Dots does something similar with what OpenAI calls proactive research, using read-only tools while idle to spot changes in your connected apps. Muse watches your inbox and other connected services and pings you when something changes, like a price drop or a reply you missed.
Second, it keeps working after you leave. All three run on their own computer in the cloud, not on your phone or laptop. You can close the app, turn your phone off, and the job keeps going. When it finishes, the result comes back as a document, a dashboard, a draft reply, or a message.
Third, it is supposed to ask before doing anything that matters. Every one of these companies makes the same promise in different words: nothing gets sent, paid, signed, or deleted without your approval. How strict that promise is, and how much control you actually get, is one of the real differences between them. I will come back to that, because it matters more than any feature list.
If you only remember one sentence from this article, make it this one. These are not smarter chatbots. They are assistants with permission to look at your stuff and act on it, and permission is the whole story.
Claude Duet: the work assistant that starts your day for you
Anthropic describes Duet in one line that I actually like: it comes to you with what needs attention, completes tasks you ask for, keeps working when you step away, and checks with you before it acts on your behalf. The launch coverage in this Duet report matches that description closely, including the warning that background work counts toward plan limits. You message it the way you would message a person, by text or voice, in a standalone app on iOS and Android or on the web.
The morning brief is the heart of it. Duet reads your email and calendar, picks out what most needs your attention, and can start the first steps, like drafting replies for you to review. Through the day it sends updates on the work it is doing and follows up on threads that have gone quiet. Give it a bigger job, like a research brief, a sales forecast, or a dashboard, and it works on it in the background and hands back a file you can edit. Finished work shows up in a Make tab.
Duet runs on Claude Opus 5.5, and you cannot pick another model in the preview. If model names like Opus 5.5 and GPT-6 Astra mean little to you yet, my explainer on Gemini 4 Argon shows how I break down a new model release, price and access first, hype last. It uses the same memories, skills, and connectors as regular Claude, so it can search past conversations instead of making you repeat yourself. Anthropic says it works with a large connector library, and early coverage points to Gmail and Slack among the examples, plus Stripe for checking unpaid invoices and drafting payment reminders.
Here is the catch, and it is a big one. Duet is a research preview, not a general launch. It is open to Claude Pro and Max subscribers in the United States, by waitlist, with invitations going out in waves. Team and Enterprise plans are not included yet. There is no separate free tier, and the preview is included at no extra cost but shares your existing plan limits. Background checks and big tasks draw on that same allowance, so Anthropic warns you could hit your limits sooner than with chat alone. Anthropic is giving each admitted user one limit reset, available until November 30, 2026, which tells you they expect people to burn through limits while testing.
Anthropic is also unusually open about what Duet gets wrong. It says Duet might misread a thread, misjudge what is urgent, or miss something that arrives between check-ins. It is weaker at personal errands, like booking tickets, because it was built around work. If your business runs on Microsoft 365, Duet cannot connect to it yet, which rules it out for a lot of offices in the short term.
My read: Duet is the most focused of the three. It knows exactly who it is for, which is a busy professional drowning in email and follow-ups, and it does not pretend to be a general errand runner. But right now it is also the hardest to actually get. The timing is worth noting too. Anthropic filed to go public on October 26, 2026, in the same jittery AI market I wrote about when Alphabet lost $500 billion in a day, so expect a lot of big claims about these agents while the company is in the spotlight. Judge the product, not the valuation headlines.
ChatGPT Dots: the agent for people already deep inside ChatGPT
OpenAI launched Dots at its DevDay event on September 29, 2026. A dot is a personal agent that lives inside ChatGPT, has its own cloud computer and browser, and keeps working on your goals between conversations. You reach it through ChatGPT, voice, Slack, or Microsoft Teams, and it reports back in those same places.
The idea is close to Duet, but the shape is different. Where Duet is built around a morning brief and one ongoing conversation, a dot is built around ongoing goals. You give it something that takes days, like tracking a project, preparing a launch, or keeping an eye on a set of accounts, and it works on it in the background. When it is idle, it does proactive research with read-only tools, taking private notes to spot changes. Those tools cannot directly send messages, edit connected apps, or control a browser on their own. Follow-up actions have to pass the usual permission and safety checks.
Dots runs on GPT-6 Astra and connects to more than 4,000 apps through ChatGPT plugins, a setup covered in detail in this Dots explainer. That app count is the biggest practical advantage Dots has. If your work already lives across a pile of tools, a dot is more likely to reach them than a narrower agent.
The control system is also the most detailed of the three, at least on paper. For each kind of action you choose a rule: act without asking, act if pre-approved, ask first, or hand it back to you. A separate check called Auto-review looks at an email recipient and content before it goes out, and OpenAI says password changes and money transfers always come back to you. That is a good design. The question, as always with agent safety, is how it behaves on a tired Tuesday when you have approved a lot of things quickly and stopped reading carefully.
Price and access are where Dots gets expensive fast. Your first dot is included in the plan, but the plans that include one start at $100 a month. ChatGPT Pro costs $100, $200, or $500 a month, and Business Premium is $100 per user per month billed annually or $125 month to month. Free, Go, and Plus do not include Dots at launch. Pro users in the European Economic Area, the UK, and Switzerland cannot get one at all, while Business Premium users are eligible in all supported regions. Enterprise, Edu, and Healthcare workspaces can use Dots in beta, but only if an admin turns it on. You also cannot create a dot on a phone. Setup needs the desktop app or desktop web.
One more detail that surprised me: chatting with your dot does not count toward your ChatGPT usage limit, but tasks it runs in Codex or ChatGPT Work do use the usual allowances. So the headline “included” is true, but the heavy work still draws on a meter.
My read: Dots makes the most sense if you already pay for ChatGPT Pro or Business Premium and your work lives in a lot of connected apps. Paying $100 a month just to try a dot, when you could wait for a cheaper tier, is hard to justify for most beginners.
Meta Muse: the everyday errand runner with the lowest price
Meta launched Muse on September 8, 2026, and it is the odd one out in a good way. Where Duet is built for work and Dots is built for ongoing projects, Muse is built for everyday life. You give it a goal in plain words and it does the tasks for you: emailing, booking travel, filling in forms, shopping, and negotiating bills. It runs on Meta’s Muse Spark model, and you reach it through the Muse app on iOS and Android, on the web at muse.ai, on a Mac app, and inside WhatsApp, where talking to it works like messaging another person.
Muse for Small Business arrived on September 29, adding business skills and connectors for owners who want help planning content, improving ads, reviewing sales and expenses, and following up with customers. Meta says nothing publishes, sends, or spends without your approval, and it describes a separate guard that checks outbound actions before they go out. That guard is a sensible idea, but it has not been independently audited, so I would treat it as a promise to test, not a guarantee.
The price is the reason Muse deserves attention. It is free with a usage limit, and the paid plans are Power at $20 a month and Maximum at $100 a month. Reports put the free allowance at 100 million tokens a week, Power at 500 million, and Maximum at 3 billion, though token counts mean very little until you know how much a real week of use burns. A payment card is required to sign up, even on the free tier, which is worth knowing before you start.
Availability is the trade-off. Muse is available to adults aged 18 and over in the United States and Canada only. It launched US-only and reached Canada on September 18. Meta has not announced dates for the UK, Europe, or anywhere else. That is a strange position for a product aimed at everyday consumers, and it means most of my readers cannot try it yet no matter how good it sounds.
There are two other wrinkles. Amazon blocked Muse from shopping on Amazon.com on September 20, 2026, which cuts out the biggest store online for a shopping agent. And an early Mac security flaw was patched within hours, which is reassuring in one way and a reminder in another: an agent with this much access is a bigger target than a chatbot, and its security history will matter more than its feature list.
My read: Muse is the most approachable of the three and the only one a normal person can try for free. Its weakness is not the product idea, it is the map. Two countries is a very small launch for an everyday assistant.
Price and access, side by side
This is the table I wish I had seen before reading ten launch posts. Prices are in USD, as listed at launch, and availability is as of late October 2026. All three companies are still rolling these out gradually, so check before you pay for anything.
| Maker | Anthropic | OpenAI | Meta |
| Launched | October 21, 2026 (research preview) | September 29, 2026 | September 8, 2026 |
| Model | Claude Opus 5.5 | GPT-6 Astra | Muse Spark |
| Starting price | Included with Claude Pro or Max (Pro from $20 a month) | Included with ChatGPT Pro from $100 a month, or Business Premium | Free with a usage limit, then $20 or $100 a month |
| Who can get it now | US Pro and Max subscribers, by waitlist | Pro outside the EEA, UK, and Switzerland; Business Premium in supported regions; Enterprise beta if an admin enables it | Adults 18 and over in the US and Canada |
| Where you reach it | Standalone iOS and Android app, web | ChatGPT, Slack, Microsoft Teams, voice | Muse app, web, Mac app, WhatsApp |
| Built for | Work briefs, follow-ups, documents | Ongoing projects across many apps | Everyday errands, shopping, travel, bills |
| Biggest limit | Waitlisted, US-only, no Microsoft 365 yet | $100 a month minimum, no phone setup | US and Canada only, Amazon blocked shopping |
Read that table twice before you fall for any one product. The cheapest agent you cannot use is not cheap. The most powerful agent on a plan you would never buy is not a bargain either.
The safety question nobody should skip
I keep coming back to permissions because they are the real product here. A chatbot that gives a bad answer wastes your time. An agent with access to your email that sends the wrong reply, pays the wrong invoice, or deletes the wrong file costs you something real.
All three companies handle this the same way at a high level. They ask before consequential actions, they let you set rules, and they keep credentials away from the agent itself. Dots has the most granular rules, Duet leans on a simple promise to check before it sends, pays, signs, or deletes, and Muse describes an approval guard on every outbound action.
Here is what I would actually do before connecting any of them, and I would do it in this order.
Start with one low-risk app, not your whole digital life. Email read access for a morning brief is a reasonable first step. Full access to your inbox, calendar, files, payments, and shopping in one go is not.
Turn on “ask first” for everything at the start, even if the setup screen offers to let it act alone. You can loosen the rules later for jobs you trust. You cannot unsend an email.
Check what happens between check-ins. Duet checks every few hours, not constantly. Dots does proactive research while idle but with read-only tools. If a message arrives at 9:05 and your agent checks at 11:00, it will be late. Do not build anything time-critical on an agent that checks on a schedule.
Keep a separate eye on money. Even with approval steps, I would not connect a payment method to an agent in its first month of public use. Drafting a payment reminder is useful. Moving money is a different job, and Dots is right to always hand money transfers back to the user.
Finally, remember that usage limits are a safety feature in disguise. Duet spending your Claude limits on background checks means a runaway agent costs you access, not just embarrassment. If you hit a limit early, that is information. Slow down and look at what it was doing.
None of this means you should avoid these agents. It means you should treat them like a new employee with a lot of keys, not like a smarter search box. Give keys slowly, check the work, and keep the important ones yourself for a while.
Where they fall short, honestly
Every launch post talks about what these agents can do. The limits are more useful, because they decide whether the product fits your life.
Duet can misread a thread or misjudge urgency, and Anthropic says so itself. It can miss things between check-ins. It is weaker at personal errands. It does not connect to Microsoft 365 yet. Its background work eats your plan limits. If you are outside the US, none of that matters yet because you cannot get in.
Dots costs at least $100 a month at the individual level, which puts it out of reach for most beginners and small businesses. It cannot be set up on a phone. Pro users in the UK, Europe, and Switzerland are excluded. Extra dots have no published price, so you cannot plan for a team setup. And because it is only a month old, the early reports are still thin. The product will change, and early adopters will be the ones finding the rough edges.
Muse is the most limited by geography. Two countries, adults only, no announced dates elsewhere. Amazon blocking shopping is a real blow for an errand agent. Its security story has not been independently audited, and a Mac flaw in its first weeks shows why that matters. The free tier sounds generous in tokens, but token allowances are hard to translate into real use until you have tried it for a week.
There is also a shared weakness that none of the companies talk about much. These agents are good at work that looks like other work. A follow-up email, a dashboard, a price check, a booking form. They are much less proven at judgment calls: which client is actually angry, which invoice is actually wrong, which “urgent” message can wait until Monday. That judgment is still your job, and handing it over too early is how small mistakes become expensive ones.
What this means if you are outside the US, UK, or Europe
A lot of tech coverage writes as if everyone lives in California. Most readers do not, and for this particular story that changes the advice completely.
If you are in the US and already pay for Claude Pro or Max, joining the Duet waitlist costs nothing extra and is worth doing, if only to learn how the product works. If you are in the US or Canada and want to try a proactive agent for free, Muse is the only real option right now, and its WhatsApp access makes it easy to test without changing how you work.
If you are in the UK or Europe, Dots on a Pro plan is closed to you at launch, Duet is US-only, and Muse has no announced date. Business Premium is the one door that is open for Dots in supported regions, and that is a business purchase, not a casual trial. For everyone else, the honest advice is to wait and watch. These products are weeks old. Prices, regions, and limits will all move in the next few months, and the second version of each will be better than the first.
That waiting time is not wasted. You can do the useful prep now without paying anyone. Clean up your inbox rules, decide which apps an agent should never touch, and write down the three repetitive jobs you would actually hand over first. When an agent does reach your country, you will know exactly what to test it on instead of connecting everything and hoping.
Who should try what, and what I would do
If you run a small business or a busy freelance practice, live in the US, and already pay for Claude, Duet is the one I would watch closest. The morning brief and follow-up chasing solve a real, daily pain, and the work-first design fits that life. I would still start with ask-first rules and one connected app.
If your work already lives inside ChatGPT and across dozens of connected apps, and you already pay for Pro or Business Premium, Dots is the natural fit. The app coverage and the rule system are genuinely strong. I would not upgrade to Pro just for a dot, though. Wait until a cheaper tier gets access or until you have a project big enough to justify $100 a month.
If you are in the US or Canada and just curious, Muse is the easy test. It is free to start, it lives in WhatsApp, and everyday errands are a forgiving place to learn how agents behave. I would keep payments disconnected at first and treat its shopping help as suggestions, not decisions.
If you are anywhere else, or you are on a tight budget, my recommendation is simple: do not pay, do not join a waitlist you cannot use, and do not reshape your workflow around a product that has not reached you. Something useful today, for free, is a better use of your time, and my roundup of 10 free AI tools that save hours every week is still the place I would start. Read the coverage, note which agent fits the way you actually work, and revisit in a few months. The companies are competing hard, and that competition usually means wider access and lower prices, not the other way around.
For myself, I am in the waiting group, and I am fine with that. The idea of an assistant that brings me the five things that matter each morning, instead of making me dig through 41 emails, is genuinely appealing. But I want to see a second round of safety reports, a wider rollout, and real pricing for extra capacity before I hand any of these agents the keys to my inbox. When Duet, Dots, or Muse reaches more countries with a plan I can actually buy, I will test it properly and update this article with what I found.
Until then, the smartest move is also the simplest one. Keep your chatbot for questions. Let an agent earn its way into your apps one permission at a time, and judge it on the work it hands back, not on the launch video.