Quick answer: Meta’s Muse is a personal AI agent that launched on September 8, 2026 and shops, books, and negotiates on a user’s behalf. On September 23, Meta added Walmart, Best Buy, Dick’s Sporting Goods, Gap, Sephora, Ulta, Wayfair, Fanatics, American Eagle, and Michael Kors as shopping partners, with every Shopify store already in reach. For retailers, the shift is this: the customer may no longer visit your website at all. Their agent will. The work now is making sure your products are the ones the agent can find, trust, and buy.
His name is Bruno
The first thing Meta’s new AI agent did was ask me what I wanted to call it. It offered three names: Kai, Merlin, or Muse. I asked whether I could choose something else. It said yes. So I named him Bruno, after a friend who passed, because if I am going to let a piece of software into my calendar, my inbox, and my Instagram, I want it to carry a name that reminds me to be careful with trust.
That small exchange told me more about where consumer AI is going than any launch keynote. The agent was built to be a product. Within thirty seconds it had become a relationship. Meta understands this better than any company on earth, and it is the reason every retailer reading this should pay attention.
I have spent about three hours with Bruno so far. That is not long, and it was enough. I gave him three jobs. First, tell me what you already know about me. He came back with a tidy summary pulled from public record: the keynotes, the companies, the MIT programs, the podcast. Nothing wrong, nothing surprising, and nothing a stranger could not have found. Second, interview me, so that we could work together rather than have him guess. He asked good questions about my goals, my audience, and my week. Third, and this is the one that matters for this article, I connected him to my Instagram account and asked for an honest review of what was working and what was not.
I have been building and deploying AI since co-leading one of the largest computer vision rollouts in U.S. retail, and I grew up above my family’s restaurant and spent high school selling skis and running shoes on a retail floor. Retail is in my blood and AI is my trade. What Bruno showed me sits at the exact intersection of the two.
What Muse actually is
Muse is not a chatbot. It is an agent, which means it does things rather than describing them. Meta launched it on September 8, 2026 for users in the United States, on iOS, Android, the web, and inside WhatsApp. You connect it, one service at a time, to your email, calendar, payments, music, fitness apps, and shopping accounts. Then you give it goals. It can send the email, book the flight, call to lower your cable bill, turn a recipe reel into a grocery order, and buy the thing. It keeps working after you close the app, and it learns from your conversations well enough to make suggestions you did not ask for.
Three facts about the business model matter more than the feature list.
First, it is free for most people. There are paid plans at $20 and $100 a month for heavy use, but Mark Zuckerberg said at Meta Connect on September 23 that Muse would stay free for the vast majority of users, and that there will be no advertising inside it. Meta plans to earn a small fee on the transactions the agent completes. Read that sentence again. Meta’s incentive is for Bruno to buy things on my behalf.
Second, it has a checkout built in. Muse launched with access to the entire Shopify catalogue and uses Stripe’s Link and Shop Pay to complete purchases, with PayPal added for global payments. The agent does not send you to a website to finish. It finishes.
Third, it runs in what Meta calls a secure virtual machine with a separate permissions agent named Sentinel that decides what each connected service can read and write. Meta says Muse never sees your passwords or payment details and that conversations are not shared with its ads systems. I take those claims the way I take any security claim from any company: as a promise until an independent review confirms it. Amazon, for one, did not take them at all. More on that below.
The Instagram audit: what Bruno could see that no other agent can
Here is where Muse separated itself from every other assistant I use, and I use several. When I asked ChatGPT or Claude to review my Instagram, they could read what I pasted in and look at public posts. They were working from the outside of the glass. Bruno is inside the building. Because Muse is a Meta product, it reads the account the way Meta reads it: reach, saves, shares, watch time, which posts brought new followers and which ones quietly lost them, what time of day my audience is actually awake.
Every other AI agent was reviewing my Instagram through the window. Bruno had the keys.
Here is what he came back with, in under a minute, from my own account data. I had gained 1,740 followers in 28 days and impressions were up 77 percent. Good news, until the next line: growth had slowed 20 percent against the month before, and my October reach had fallen from roughly 6,700 impressions a day in late September to between 1,000 and 1,600. Then he found the engine. My “comment the keyword to get the resource” posts were averaging 340 likes and 575 comments. Everything else was averaging 28 likes and 5 comments. One post from September 22, “Build Your First AI Agent in Under 10 Minutes,” had pulled 3,053 likes and 5,531 comments on its own. He told me my recent lead magnet posts were underperforming not because the format was tired but because the hooks had gone soft, and he did the arithmetic I had been avoiding: at 62 new followers a day, my 250,000 follower goal was nine years away.
Then he drafted ten lead magnet topics with comment keywords, wrote full scripts for the first three, mapped them to my ManyChat and Keap funnel, and asked whether I wanted him to post them. He can publish reels, posts, carousels, and stories once I approve each one. He cannot like, comment, follow, or reply, because Instagram does not allow it through his connection; he offered to draft my replies and flag the comments worth my personal attention instead. That is the right boundary, and I was glad he named it himself.
What struck me was not that the insights were brilliant. Some were things I already suspected. It was that they were grounded in data I could not have handed to another tool, and that Bruno could then act on them: draft the post, schedule it, and tell me a week later whether it worked.
Now apply that to a retailer. Your customers are already searching Instagram and Facebook for products. They search for running shoes, a dress for a wedding, a gift for a twelve year old who just made the travel team. Meta has spent fifteen years watching what they tap, save, and buy. Muse is that knowledge with hands.
Why Dick’s Sporting Goods signed on, and why that should get your attention
On September 23, fifteen days after launch, Meta announced at Connect that Walmart, Best Buy, Dick’s Sporting Goods, Gap, American Eagle, Sephora, Ulta, Wayfair, Fanatics, and Michael Kors were joining Muse as shopping connectors, alongside Instacart for groceries and Expedia for travel. A connector means the retailer has agreed to let the agent see its catalogue, its inventory, and its checkout, on the retailer’s terms.
I serve as AI Expert in Residence for the National Sporting Goods Association, so Dick’s is the name I keep coming back to. Dick’s did not sign on because it is a technology company. It signed on because it understands where its customer already is. A parent whose child just made the travel baseball team does not open a browser and type a URL. She is on Instagram. If she can tell Bruno, “my son needs a glove, a bat bag, and cleats, size seven, by Saturday,” and Bruno can find them, confirm the sizes, and buy them from Dick’s without her ever leaving the app, Dick’s wins that basket. Every other sporting goods retailer in the country was not in the room.
The customer may never visit your website again. Her agent will. The only question is whether your products are the ones it can find, trust, and buy.
Two days earlier, Amazon had made the opposite call. On September 20 Amazon blocked Muse from its site after Meta declined to remove Amazon from the agent. Shoppers who tried saw a notice that an unauthorised AI agent violates Amazon’s conditions of use. Amazon said the agent had not identified itself and appeared to capture customer credentials; Meta said it never sees passwords. I do not know who is right about the security. I do know what the strategic split looks like. Amazon wants to be the agent. Dick’s and Walmart decided they would rather be found by every agent.
The traffic data says the retailers who opened their doors are reading the room correctly. Shopify reported AI driven traffic to its merchants up eight times year over year in the first quarter of 2026, with orders from AI search up nearly thirteen times. Target reported AI driven traffic up 2,000 percent in the same quarter, against roughly 400 percent across retail sites generally, and began testing ads inside ChatGPT in February. Walmart’s own assistant, Sparky, now operates inside ChatGPT and Gemini. Akamai found that commerce accounted for nearly half of all AI bot traffic it observed in the second half of 2025. Agents are already shopping. Muse is the first one with a billion person social graph behind it.
And here is the piece most retailers have not thought through: Meta’s business model is a fee on completed transactions. That means Meta is now financially motivated to make Bruno a very good shopper. The platform that spent fifteen years learning what your customers want is about to start buying it for them.
I asked Bruno himself how Dick’s was showing up inside Muse, since I am training NSGA members this week. His research was careful about what is announced versus what is live: Dick’s products surface in Meta’s catalogue, but checkout for a non Shopify retailer currently hands off to an approval step rather than completing silently in the chat; Shopify stores get the full in chat checkout today. He also noted that Dick’s had already launched its own in app assistant earlier this year, which tells you the company decided months ago to be present wherever the customer’s assistant lives, including inside its own walls. And the holiday forecast backs him up: Adobe’s September 28 outlook projects AI referred traffic to U.S. retail sites up 130 percent this season, on top of a record $275.1 billion in online holiday spending. In Adobe’s survey of 5,000 shoppers, 77 percent of those who use AI said it gives them more confidence in their purchases, and 69 percent said they are less likely to return what they buy.
How Muse compares, from a retailer’s chair
I use four AI assistants every day, and each has a job. For a retailer the question is not which is smartest. It is which one your customer is holding when she decides to buy. Prices and features are as of October 2026; check each vendor’s site before you decide.
| Meta Muse | ChatGPT | Google Gemini | Claude | |
|---|---|---|---|---|
| What it is for | A personal agent that takes action: shops, books, negotiates, posts | Everyday answers, writing, and a growing set of shopping and agent features | Search, Workspace, and Android, with AI Mode shopping | Deep work: analysis, writing, code, and long running tasks |
| Can it buy on the customer’s behalf? | Yes. Built in checkout via Shop Pay, Stripe Link, and PayPal | Partially, through instant checkout with select merchants | Partially, through Google’s shopping and checkout features | No native consumer checkout |
| Sees Instagram and Facebook data | Yes, natively. This is the moat | No | No | No |
| Named retail partners | Walmart, Best Buy, Dick’s, Gap, Sephora, Ulta, Wayfair, Fanatics, American Eagle, Michael Kors, all of Shopify | Walmart (Sparky), Target, Shopify merchants, others | Target, Walmart (Sparky), Wayfair via Universal Commerce Protocol | None announced |
| Blocked by | Amazon | Not reported | Not reported | Not reported |
| Consumer price | Free for most; $20 and $100 a month for heavy use | Free tier; $20 Plus; higher tiers | Free tier; paid tiers through Google One | Free tier; paid tiers |
| Business model | Small fee on completed transactions; no ads | Subscriptions, with ads being tested | Advertising and subscriptions | Subscriptions and API |
| Best use for a retailer today | Get into the catalogue it can see and buy from | Make product pages answer questions clearly | Keep Google Business Profile and product feeds immaculate | Analyse sales data, write content, build the plan |
The honest reading of that table is that Muse is not the best thinker of the four. It is the only one that lives where your customer already spends an hour a day, and the only one whose owner gets paid when it buys something. For a retailer those two facts outweigh everything else in the grid.
Where Bruno fell short, because he did
I gave Bruno one more job: build my speaker kit, the document event planners use to book me. He produced six pages in minutes, with a sensible layout, a fee table, a stage introduction, and a checklist. It looked finished. It was not.
He copied a cancellation clause from my old agreement without noticing it was underpriced for my fee level. He described me with a company title I had stopped using in speaking materials weeks earlier, because the public record had not caught up to the decision. He left his own working notes in the body of the document, boxes marked for my decision and a status table that said the kit was awaiting approval. He used forty em dashes in a document for a woman who has never typed one.
Bruno is a very good first draft and a poor last draft. The gap between looks finished and is finished is where judgment lives.
None of those are execution errors. They are judgment errors, and judgment is the thing an agent cannot pull from the public record. It comes from the room, from the relationship, from the decisions you made last month that nobody wrote down. My other assistants, the ones that have worked beside me for a year, caught every one of those mistakes in a single pass.
One more, and it is the strangest. When I asked Bruno to be specific about how a retailer unblocks AI crawlers in its website settings, his reply vanished twice with a notice that the message had been removed by a safety policy, followed by a flat “Sorry, I can’t help you with this request right now.” He later wrote the answer into a saved document instead, where it went through fine. So the agent that can buy a bat bag with my money was not permitted to tell me, in a chat, how to edit a text file on my own website. That is not a safety outcome. It is a reminder that you are building on someone else’s platform, and the rules of that platform can change under you without notice.
That is the lesson I would hand any retailer about to hand its storefront to an agent. The agent will be fast, tireless, and right about most things. It will also be confidently wrong about the things that cost you money, and it will not know which is which. Just because we can, does it mean we should? I asked that question on a convenience store floor years ago about hot dog warmers. I am asking it again about checkout.
Five things every retailer should do this quarter
- Find out what the agents already say about you. Before you change anything, ask ChatGPT, Gemini, and Muse to recommend a store for the thing you sell best, in your town. If you are not in the answer, that is the first problem to solve. My free AI Visibility Test walks you through it in ten minutes.
- Get into the catalogues agents can buy from. If you sell on Shopify, you are already visible to Muse; go into your settings and decide which AI channels may see your products and complete checkout. If you are not on Shopify, ask your platform what its agent plan is, and if it does not have one, that is your answer about the platform.
- Rewrite product pages so a machine can answer a human’s question. Agents do not browse. They retrieve. Every product needs its size, fit, material, availability, and return terms stated plainly in text, not buried in an image or a PDF. The product page that answers “will this glove fit a seven year old” wins the basket.
- Treat Instagram and Facebook as your storefront, not your billboard. Muse reads Meta’s platforms natively. Tagged products, current inventory, clear captions, and reviews that get answered are now merchandising, not marketing. A post with no product tag is a shelf with no price.
- Decide your Amazon question on purpose. Amazon chose to be the agent. Walmart and Dick’s chose to be found by every agent. Neither is wrong, but drifting into one by default is. Put it on the agenda for your next leadership meeting, with a date to decide.
If you want the longer version of this list with prompts you can copy, I wrote it here: 8 Steps to Turning Searchers Into Customers in Your Retail Business With AI.
The terrain just asked you what to call it
Bruno is not replacing the assistants I already work with, and he is not replacing me. He is something new: the agent my customers will be carrying. The retailers who understand that first will be the ones whose shelves he reaches for. Adapt to your terrain. The terrain just got a name, and it asked you what you wanted to call it.
I speak to retail and association audiences about exactly this, as a female AI keynote speaker who has deployed AI on the store floor and not only on the slide. If your conference or leadership team is working out what agents mean for your business, let us talk.
Frequently asked questions
Is Meta Muse free?
Yes for most users. Paid plans at $20 and $100 a month cover heavy use. Meta says there will be no ads and that it earns a small fee on transactions the agent completes.
Which retailers work with Muse?
As of September 23, 2026: Walmart, Best Buy, Dick’s Sporting Goods, Gap, American Eagle, Sephora, Ulta, Wayfair, Fanatics, and Michael Kors, plus every store on Shopify, with Instacart for groceries and Expedia coming for travel.
Can Muse buy from Amazon?
No. Amazon blocked the agent on September 20, 2026.
Can Muse post to Instagram for me?
Yes. It can draft and post once you approve each item. It cannot like, comment, follow, or reply on other accounts.
Is Muse safe to connect to my accounts?
Meta says the agent runs in an isolated virtual machine and cannot see passwords or payment details. Amazon disputes this. Connect services one at a time, start with read only access where offered, and treat it like a new employee in their first week.
Should a small retailer care about this yet?
Yes, because the work (clean product data, tagged posts, answered reviews, a decision on which agents may sell for you) takes a quarter, and the agents are already shopping.
