ISSUE 27 · ME+MACHINE
If you've ever used Facebook Marketplace, you know it can be a slog. A place where buying a table can become a part-time job. Find a listing. Work out whether the price is fair. Ask if it is still available. Wait. Message again. Haggle. Get ghosted. Then repeat.
Over the weekend, I handed much of that busywork to Meta’s new personal AI agent, Muse. Muse is kind of like OpenClaw, but plugged into my Meta life.
I told it I was looking for a table, it found listings, researched prices, directly handled the back-and-forth with multiple sellers, and reported back to me when it made significant progress.
Muse has its own virtual computer, meaning somewhere to actually do work. It can open a browser, work with files, make a document, and keep going even after you close the app. You can watch what it is doing in the browser and take over if needed.
It remembers context across conversations, and can work on a schedule or when something relevant triggers it. It keeps goals and an activity log, so you can see what it is doing.
All of that is cool, but none of it is unique. OpenClaw, Grok bots and ChatGPT can all use tools and carry out tasks. So what makes Muse different?
The answer is Meta.
What makes Muse different is what it already knows before I even ask. Meta’s apps already contain years of clues about what catches my attention: the things I save, the creators and communities I follow, and the products I consider.
If Muse can use that context, it could make judgments based on my actual taste and social world instead of making me reconstruct them in a prompt. Then it can use the same kinds of tools other agents have to act on those judgments.
Meta has not explained exactly how much of that information Muse can use, but they have confirmed that Muse can begin with something saved on Instagram.
So, say I save a Reel of a dining room because I love the table, even though I couldn’t tell you what the style is called (for the record, it's mid-century modern). Muse could use my saved Reels as a visual reference, combine it with the room measurements and budget I had already given it, search Marketplace for similar tables, rule out the ones that would not fit or are overpriced, and message sellers about the best matches.
Other agents could do all of that too, but I would first have to find the Reel, upload it, and provide the relevant context. With Muse, the Reel, the Marketplace listings and the seller conversations could all become part of the same task.
That broader potential is why Millie Yang argues that Muse could outperform ChatGPT and Claude for mainstream users by turning Instagram’s niche, creator-driven content into real-time search and personalized feeds.
Marketplace gave me an early glimpse of what that could feel like: I asked for help in a place I already use, and Muse handled enough of the work to get me to a result. By the end of the weekend, Muse had bought me a dining table, chairs, a work desk, and a monitor.
THE 30-SECOND VERSION
Muse took over my Facebook Marketplace busywork: finding listings, checking prices, messaging sellers, and keeping the search moving.
By Sunday, Muse had helped me buy a desk, monitor, and dining set.
Those agent abilities aren’t unique. Muse’s advantage is Meta’s data—what I save, follow, and consider—so it may understand my taste without making me explain everything first.
Meta hasn’t revealed exactly how much of that context Muse can access, and some activity may indirectly influence ads.
Muse could easily become the top AI agent for mainstream users
The real table stakes for personal AI may be knowing which table I want before I do.
me+machine.
__
**Meta says Muse conversations and data in its cloud computer are NOT shared with Meta’s advertising systems. But also, it does say activity Muse performs elsewhere, including browsing and Marketplace work, may indirectly influence the ads you see.