Claude Duet, dots or Muse: which one should you use for work?

Welcome back.

Claude Duet, dots and Muse all promise to keep working while you're busy with something else. This week, I'm looking at which of these personal agent is actually worth using for work.

Before we dive in, I'm hosting a free live Q&A with Chatbase this Thursday on how AI agents are changing customer support. Grab your spot here.

IN THIS EDITION

• Byte-size news: updates on Office Agent and dots, plus how agents recover from interruptions.

• Main read: which personal AI agent should you use for work?

• Your turn: a quick test to see if you should stick to your current AI agent

■ Byte-size News

1. Office Agent moves out of Copilot

Microsoft removed Office Agent entry points from Copilot on November 2. Existing files and tasks weren't moved over, but you can still find them in the standalone app. If your team used it, make sure they know where their work lives.

MICROSOFT · NOV 2

2. A date for your IT team: November 18

OpenAI's Enterprise dots preview ends November 18. Access then becomes available by default unless an admin opts out, and billing begins for enabled dots. Ask who owns that decision before the preview becomes a recurring bill.

OPENAI · NOV 18

3. Microsoft shows how agents recover interrupted work

Microsoft's November 5 Foundry update explains how hosted agents can restart after interruptions. Apps still need to save progress and avoid repeating actions. If an agent sends requests or updates records, ask what happens when it retries.

MICROSOFT · NOV 5

How AI Agents Change Support: Insights from Chatbase

Customers are starting to send AI assistants to contact businesses for them.

So what should a support team do when an assistant asks to change an order or get into an account?

On Thursday, November 12, at 2 PM ET, I'm joining Clint Kruger from Chatbase for a free live Q&A and a short demo.

We'll get into what Chatbase is seeing, how to check who an assistant represents, what it should be allowed to do, and how support teams can get ready.

■ Haroon-in-the-loop

Claude Duet, dots, or Muse: which one should you use for work?

All three promise to handle work for you.

The problem is that each one connects to different apps, so it can miss the details that matter. A polished answer built on old or wrong information still leaves the real work to you.

Take prep for a client call. The agent needs to catch the latest email and any calendar clash before writing your brief.

I dug into how all three hold up in real work. Here's a quick look at each one and where I'd use it.

• Claude Duet costs nothing extra for now if you're on Claude Pro, Max, or Ultra. It just counts toward your usual Claude limits.

• dots is included with ChatGPT Pro and Business Premium. Eligible Enterprise workspaces start getting billed for it on November 18.

• Muse is mostly free for everyday use, Meta says, and subscription plans cover more.

What each one can reach

An agent can only work with the apps you connect it to. Here's what each one can reach:

• Duet: It carries over the memory, skills, and connectors already set up in Claude, including Gmail, Google Calendar, Google Drive, Slack, and HubSpot. Microsoft 365 isn't available yet, and Slack works with limits during the preview. You also can't watch or take over Duet's cloud browser.

• dots: A dot has its own cloud computer and connects to apps through plugins. OpenAI says that it includes more than 4,000 apps, and you can message it in ChatGPT, Slack, or Teams. Access to your own laptop is optional and off by default. And being signed in to a site on your laptop doesn't sign your dot in.

• Muse: It lets you choose which apps it connects to, and what it can do with each. For email, that includes whether it can only read or also send.

Note: Check with your company before connecting any of these agents to work accounts. Disconnecting an app doesn't always clear what the agent already saw. With dots, for example, you'll want to check its memory settings and delete anything you don't want it to keep.

What it notices on its own

What sets these agents apart from a regular chatbot is that they can tell you something's off before you ask. Here's what each one has caught so far:

• Duet: It catches the small mistakes that throw off a workday. Early testers say it spotted a showing booked on the wrong day, and it noticed a client call was right before a flight and offered to move it. Those examples come from Anthropic's own testers, though.

• dots: It connects things across all your apps. When Axios asked it to plan a trip to a game, it caught that the game had moved up a day and that the new date clashed with something already on the calendar. Another user got a heads-up when someone replied about a group project, even though she'd never asked it to watch that thread.

• Muse: In the same test, it also caught the moved game and noticed tickets weren't on sale yet. It missed the calendar clash, though.

tl;dr: dots is the best at flagging problems on its own.

How far it gets with the work

Catching a problem is only half the job. You still need the agent to finish the work.

• dots: This is the one that gets closest to finished work. A reviewer at The Verge talked through a website redesign on a quick call, then walked away. Her dot built the next version and put it live, and she got a ping when it was ready. But it works best on projects you fully control. It slows down when a website asks it to prove it's human.

• Duet: It's strongest when the work lives in documents. An early review says it can edit docs, slides, and dashboards, and keep them up to date. On open tasks, it gets most of the way and then hands them back. An early user said it couldn't take a single task from start to finish, but it was good at nudging him to wrap things up.

• Muse: It's good at research and planning, but it has trouble finishing the job. In one test, it planned the whole shopping trip, then got stuck at the store's sign-in page right before checkout.

dots is the most likely to hand you work that's actually done. Duet gets close on docs and slides, but I'd still expect to finish most tasks myself. And Muse is fine for planning. Just expect to handle the final step on your own.

Approvals and control

Before an agent emails a client or spends money, you want to see what it's about to do. All three let you check first, but each one handles it differently.

• Duet: It asks every time. Before it sends a message, pays for something, signs a document, or deletes anything, it shows you exactly what it's about to do and waits for your go-ahead.

• dots: It gives you the most control. You set rules for what it's allowed to do, and for each kind of action you decide whether it needs your OK. OpenAI's help guide is clear that asking for a draft doesn't mean it's allowed to send it.

• Muse: It shows you a card to approve before it buys or sends anything. It also keeps a log, so you can see what it's working on and what it's already done.

Whichever one you pick, I'd have it ask before it sends or pays for anything.

What happens to your data

Each one handles your data a little differently, so check the settings before you connect anything.

• Muse: It uses your activity to train Meta's AI by default. You can turn that off in Data controls. It can also pick up your interests from a connected Instagram account.

• dots: On Business, Enterprise, and Edu plans, OpenAI says it doesn't train on your workspace content by default. On personal plans, you control that setting.

• Duet: It follows the settings already on your Claude account.

Which I'd use for which job

Here's where I'd start.

• Duet: Start here for meeting prep if you already work in Claude and can get into the preview. Expect a strong nudge, and plan on finishing some of the work yourself.

• dots: Start here for a proposal, a deck or anything else that keeps changing, especially if your team already uses ChatGPT, Slack or Teams. Set its rules before you connect anything, and check the November 18 billing change.

• Muse: It's the easiest one to try for personal admin. Connect only the apps you need, turn off training if you don't want it, and keep work accounts off it for now.

■ Your turn

You don't need all three agents to try this. Start with the one you already have.

Give it only what it needs for one client meeting: the account notes, the last meeting notes, the latest email, and the calendar.

Before you run it, write down what a good brief has to catch, like a recent change on the account or a meeting clash.

Then ask:

“Prepare me for this client meeting. Tell me what's changed since we last spoke, and flag any scheduling conflicts. Draft a reply for my review. Link each important point to its source, and tell me anything you couldn't confirm. Don't send anything.”

After that, look at how much you still had to fix. Check what it missed and what it got wrong, and compare that with how long the brief would take you to write yourself.

To try another agent, give it the same material and the same prompt.

The one I'd keep is the one that consistently leaves me better prepared with less work to redo.

■ Signing off

That's it for this week.

Which agent are you using most, and what does it do well? Hit reply. I'd especially like to hear what you still have to do yourself.

Until next time,
Haroon