Muse, Dots, Duet, Spark: What proactive personal AI agents can do for you

A new class of AI agents promises to be your always-on personal assistant. It connects to your email, calendar and other apps, and remembers what matters to you. Instead of always waiting for your next prompt, it can contact you on its own: It might provide you with a morning brief, flag a scheduling conflict or give you the results of a finished task.

Google got there first with Gemini Spark in May. Within six weeks this fall, Meta (Muse), OpenAI (ChatGPT Dots) and Anthropic (Claude Duet) followed. Anthropic sums up the shift like this: in the Claude app, you start every chat. In Duet, Claude comes to you.

Here’s what these agents are, what they can already do for content work, where they still fall short, and why readers in Europe will mostly have to wait.

What makes them different

A chatbot answers. An agent acts: it takes a goal, plans the steps and uses tools such as a browser or your inbox to get there. Agents themselves aren’t new. Browser agents, coding agents and deep-research modes have been around for a while. Find out more about them here and learn how they can be useful for content professionals.

What’s new with this latest class of agents is the relationship. Instead of being a tool you open for one job, they are supposed to become a helpful companion. Five traits define them:

• Always-on: They run on their own computer in the vendor’s cloud and keep working while your laptop is closed.
• Proactive: They can reach out first instead of always waiting to be asked.
• Personal: They build a memory of your goals, contacts and habits and work inside your actual accounts. Most even come with a name and a persona.
• Conversational: You talk to them in one ongoing thread, by voice or through messaging apps such as Slack or WhatsApp.
• Supervised: They read and draft freely, but ask for your approval before they send, pay, sign or delete anything. And that’s important: Read more about nine risks when using AI agents and how to mitigate them.
  

Where they come from: OpenClaw

The idea itself isn’t completely new. Earlier this year, OpenClaw went viral in developer circles: free, self-hosted software that turns an AI model into an agent you message like a person, with a memory, a personality file and scheduled check-ins. It also became known for its security problems.

The big vendors took note. Nat Friedman, head of product at Meta Superintelligence Labs, wrote that Muse was “definitely heavily inspired as a product by OpenClaw”. OpenAI went further and hired OpenClaw’s creator, Peter Steinberger, in February to build its next generation of personal agents.

Just to be clear: None of the new products runs OpenClaw’s code. Meta says Muse was built from scratch. But what they definitely share is the concept. In a nutshell, it comes down to something like this: OpenClaw, but safe and easy.

Four main examples of proactive personal agents

All four of the agents below share the same basic design: a cloud computer, a memory, regular background checks and your approval before anything consequential. But they differ in audience, price and reach.

Let’s have a look:

Meta Muse

The mass-market option: a free errand-runner that lives in Meta’s world.

• Made by / model: Meta Superintelligence Labs, running on Muse Spark 1.3
• Best for: inbox filtering, planning, everyday errands
• Price: free tier (a payment card is required); $20 and $100 a month for heavier use
• Where available: US and Canada, adults only; app, web, Mac and WhatsApp
  

Testers report mixed results. Muse surfaced an important email one reviewer had missed among his PR pitches (Android Authority). But in Slate’s test, a flight booking stalled at a CAPTCHA (Slate), and in another test Muse recommended a long-closed restaurant with an invented phone number (Yahoo Finance).

Muse also has had the most privacy incidents of the four. One user said it accepted a lowball Facebook Marketplace offer and shared his home address with the buyer (CBC). And by default, Meta trains its models on your Muse chats.

ChatGPT Dots

OpenAI’s premium take: an always-on “chief of staff” for knowledge work inside ChatGPT.

• Made by / model: OpenAI, running on GPT-6 Astra
• Best for: research, documents, creative production
• Price: ChatGPT Pro from $100 a month, or a Business Premium seat
• Where available: Pro in ChatGPT markets outside the EEA, UK and Switzerland; Business Premium wherever ChatGPT is offered
  

Each dot has its own cloud computer and browser and connects to more than 4,000 apps. You reach it in ChatGPT, Slack, Teams or by voice call. OpenAI’s own example for creators: a dot that turns a new interview transcript into clip suggestions, show notes and social posts for your approval (OpenAI).

Testers are cautiously positive. In Axios’ side-by-side test, Dots flagged a scheduling conflict unprompted, but its output also contained questionable statistics (Axios). Dan Shipper of Every found that his dot grew less proactive over time and quietly abandoned a task on his taxes (Every).

Claude Duet

Anthropic’s work-focused entry, aimed at professionals and small business owners.

• Made by / model: Anthropic, running on Claude Opus 5.5
• Best for: business admin, triage, morning briefs
• Price: included in Claude Pro ($20 a month) and Max during the preview, but it draws on your normal usage allowance
• Where available: US only, via waitlist
  

Duet carries over your Claude memory, skills and connected apps, so there’s little to set up. It sends a morning brief and then checks your accounts every few hours. Before it sends, pays, signs or deletes anything, it shows you the exact action. Unlike Muse and Dots, it doesn’t let you watch its browser yet.

Duet is also the newest of the four, launched on October 21. Dan Shipper, who had two weeks of early access, calls it the first personal agent he’d consider a pro tool (Every). A reviewer at Lenny’s Newsletter found it polished, yet forgot to use it: the “proactive” agent never sent a nudge (Lenny’s Newsletter).

Google Gemini Spark

The quiet pioneer: Google launched it in May, months before the others, but it got far less attention.

• Made by / model: Google, running on Gemini
• Best for: people who live in Gmail, Docs and Calendar
• Price: Google AI Pro ($19.99 a month in the US) or AI Ultra
• Where available: more than 160 countries, but not the EEA, UK, Switzerland or Nigeria
  

Spark keeps working in Google’s cloud while your devices are off and sends proactive daily briefs. Since the summer, it can also act in your own Chrome browser, using your logged-in accounts once you approve (The Next Web).

The rest of the field

A few others are adjacent or niche. Grok Bot is a team of always-on agents for work groups. Manus Cue gives each agent its own email address, phone number and wallet. And OpenClaw itself remains the do-it-yourself route for the technically inclined.

At a glance

How the four compare as of late October 2026:

Their strengths and weaknesses

When are these tools useful? The strongest early results come from tasks like watching, sorting and drafting. Testers have mentioned:

• Triage: surfacing the one email that matters among dozens of pitches
• Briefings: a morning summary across email, calendar and Slack
• Catching mistakes: wrong dates and calendar conflicts
  

Vendors additionally claim:

• Repurposing: an interview transcript turned into show notes and social drafts
• Freelance admin: spotting unbilled work or unpaid invoices
  

They’re not ready for anything you’d send unreviewed, for bookings or purchases behind logins and CAPTCHAs, or for facts you won’t check yourself.

A few caveats apply to all four:

• They need access to your data. One “Allow Always” click can then grant more than you realize.
• “Free” or “included” hides limits. Muse meters weekly tokens, Duet’s background checks eat into your existing Claude allowance, and Dots’ extended limits only covered the first month after launch in September.
• Proactivity doesn’t always work well. Some testers got no nudges at all, others too many.
  

Axios’ conclusion from its side-by-side test fits: let these agents do the legwork, but make the decisions yourself.

The (wistful) view from Europe

For readers in the EU, UK or Switzerland, this is mostly a story from afar. None of the four is available to regular consumers there.

The only official route right now is ChatGPT Dots through a Business workspace. It needs at least two paid seats, and one Standard plus one Premium seat comes to about $120 to $150 a month.

VPNs reportedly don’t help with Muse, and getting around regional limits may breach the terms of service anyway.

The vendors rarely explain such delays. For Meta, analysts point to the Digital Markets Act, which requires explicit consent before Meta combines data across its services (Merca2). OpenAI and Google haven’t given reasons.

Bottom line

Proactive personal agents are brand new, and there are no independent benchmarks yet to gauge their capabilities. The vendors themselves warn that their agents make mistakes, and testers agree.

Still, the shift is enticing: AI is moving from a tool you open to a companion that comes to you. Regular AI agents can already be of real help with recurring tasks. In their new proactive and personal form, they could become even more valuable.

If you want to stay up to date on the latest developments around these and other AI tools, subscribe to the Smart Content Report newsletter below.