After Meta Muse

It's 9:40 on a Wednesday night. A mother of two is packing for a weekend at the lake when the tent pole snaps in her hands. She doesn't open your app. She doesn't search your site. She picks up her phone and says, "I need a replacement pole by Friday." By the time she has finished packing the cooler, her assistant has compared six retailers, picked one, and is waiting for her to tap yes.

She never saw your homepage. She never opened your email. Her points balance never came up. Your brand had one chance to matter, and it came and went in a few seconds, in a conversation you weren't part of.

That Wednesday is already here. In the third week of September 2026, Amazon blocked Meta's Muse from shopping its site, calling it "an unauthorized AI agent." Within days, Walmart, Best Buy, Gap, Sephora, Ulta, Wayfair and Dick's Sporting Goods signed on as Muse partners. Those are opposite bets on who owns the shopper. And both camps are betting on the same thing: the shopper is about to send an assistant.

A few weeks ago I sat down with the lead researcher at one of the largest retailers in the country. She had downloaded Muse and had been living with it. She was worried about her customers, and about the direct relationship her company had spent decades building with them. Nobody wants an agent standing between them and the people they serve. She asked me what I thought.

I told her I wrote Human Context for exactly this problem. This paper is the retail version of that answer, because she isn't the only one asking.

Ecommerce grew about half a point of retail share a year for twenty years. The pandemic pulled two years of growth into one quarter, then the line settled and kept climbing. In Q2 2026, Americans bought $340.2 billion of goods online, 17.1% of all retail sales, up 12.2% from a year earlier.

The lesson for retailers is the pace. Each shift has arrived faster than the one before it, on top of habits the last one built. Assistants start where apps left off.

The App Store opened in 2008. Within a decade, the phone had become the place where most shopping starts. Mobile now drives 75% to 77% of ecommerce site traffic, and phones accounted for 44.6% of U.S. ecommerce sales in 2024. eMarketer projects close to half by 2027.

The app closed the gap between a thought and an action. Saved payment, one-tap reorder, push alerts and curbside pickup removed nearly every step between "we're out of diapers" and "done." Apps convert at roughly 3.5%, against about 2% for mobile web.

Shoppers kept browsing, too. They spend about 202 minutes a month in shopping apps against 11 minutes on mobile shopping sites. The app won both jobs: the two-tap task and the scroll for inspiration. What changed was the task. Once a purchase could be done in seconds, customers stopped tolerating friction in it.

That speed changed what shopping meant. For a growing share of purchases, shopping stopped being an outing and became a job to get done. Retailers responded by optimizing for it: faster checkout, smarter replenishment, loyalty points for the habit. The app trained customers to expect the functional job to be effortless.

And it trained them to stop noticing who did the job, as long as it got done.

Meta launched Muse on September 8, 2026. It reached 2.5 million downloads in 13 days, 5 million in 22, and the top spot in both U.S. app stores. Muse keeps a memory of the person, keeps working after the app closes, and asks for approval only at the final decision. It browses retail sites, prepares the checkout and pays with a single-use card.

Then came the split: Amazon out, Walmart and the rest in. Meta plans to take a small fee on every partner transaction. That is: Muse intends to sit between the shopper and the store, and to be paid for standing there.

The behavior was moving before Muse arrived:

So, what are you going to do about it? You can react to Muse. Or you can think two steps ahead (or more). It is precisely for this type of challenge that I created a workshop and workbook for retailers called Human Context for Retail.

In the workbook, a case study about Trail & Summit, the composite retailer in my Human Context for Retail workbook, helps you think about how you maintain engagement with customers through AI using real world examples and the principles of Human Context.

Be honest about what has kept customers coming back. Some of it is love. A lot of it is habit.

Retail has quietly lived on consumer inertia. It's the store you visit because it's on the way home. The reorder you never question. The loyalty tier you don't want to lose. The app that's already on your phone, with your card already saved. Each one is a reason to skip the choice altogether.

Investors started pricing that risk the week Amazon blocked Muse. Companies that depend on customers who forget to cancel, never compare prices and stay put because switching is a chore were sold off. Retail lives closer to that line than most retailers like to admit.

An assistant has no inertia. It compares every time, as carefully at 9:40 on a Wednesday night as at noon on a Saturday. The habits you have been counting on as loyalty are about to be tested, one request at a time. And the customers you keep will be the ones who prefer you.

In the workbook, Trail & Summit's membership program renews at a healthy 71%, yet the renewal moment itself rates At Risk on my Time Well Spent measure. Members renew out of habit and sunk cost, and the renewal runs as an autopay line item that earns no trust. The day a lower-friction competitor shows up, that renewal rate is exposed. Muse is that competitor.

The three forecasts were written before Muse launched, and they point in the same direction. Morgan Stanley and Bain put agents at 10% to 25% of U.S. ecommerce by 2030. Salesforce expects AI agents to drive one in five ecommerce visits this holiday season. (I think their numbers are low.)

The forecasts also agree on the order. Morgan Stanley sees grocery and consumer packaged goods as the largest early unlock. Bain expects spec-driven purchases, where price and availability decide, to move first, with apparel, travel and other considered purchases following as trust builds.

Think about the five types of jobs to be done. Agents take the functional jobs first: the restock, the reorder, the commodity at the best price. The purchases that carry emotional, social and aspirational weight move later and more slowly. And the systemic job, running the routines of a household, comes slowly as people integrate personal assistants into their daily lives.

A retailer can see the cart. Muse sees the reason for the cart.

Go back to the mother with the snapped tent pole. She didn't ask for a tent pole. She asked for a weekend at the lake to still happen. The agent compared every retailer on one question: who answers that situation best? The homepage hero, the seasonal email and the points balance never entered the decision.

That is a situational market: a recurring situation that creates predictable demand, counted by how many times it happens each year. The weekend restock. The seasonal gear swap. The holiday gift that has to be right. The weather emergency. Each one recurs, each one can be counted, and each one calls for its own format, message and place to buy.

Most retailers still plan by demographic segment and product category. Muse plans by situation, because that is what the person tells it. In my Trail & Summit case study, about a composite 210-store outdoor retailer, one "Weekend Warrior" needed a fast restock one week, an emergency repair the next, and gear for her sister's first trip the week after. Same segment. Three different situations. The segment predicted none of them.

In the workbook, Trail & Summit replaces its four demographic buckets with 21 situations that add up to 793 million occasions a year. Three of them carry most of the volume it already serves: the seasonal gear swap, the weekend trip restock and holiday gifting. The growth sits in situations it had never planned for, such as renting before buying, gear anxiety among first-timers, and the scramble before a weather emergency.

Every situation had a signal already sitting in the data. A search for consumables within 72 hours of a trailhead reservation meant a weekend restock. A severe-weather alert in a member's region meant emergency preparation. Nothing had been designed to act on either. The marketing stack knew who each member was and had no idea what situation any of them was in. Profile personalization finds the right person. Situational personalization finds the right moment. And Muse is built for the second.

As I said, an agent competes on the situation. A retailer that hasn't named and counted its situations is competing blind against a tool that sees them first.

A situation tells you when a customer shows up. A mode tells you the mindset they show up in: a set of feelings and behaviors a person gets into to get something done. Two people can be in the same situation and in different modes.

Muse reads the request, the calendar and the cart. It doesn't read the excitement, the anxiety or the group text in the next tab. So it gives all three shoppers the same efficient answer. The Excited Planner wanted a conversation with someone who knows the trail. The Anxious First-Timer wanted to hear she wasn't forgetting anything. The Social Organizer wanted a list her friends could check off.

Modes change inside the same person, sometimes within weeks. In early June, Jordan spent forty minutes on an outdoor retailer's app researching a first thru-hike: maps, packs, the right shoes. Jordan was an Excited Planner, ready to be inspired. Three weeks later Jordan was back on the same app with a twisted ankle and a bruised ego, wondering whether to try again. Same person, same app, opposite needs. Jordan needed someone to say that a setback is normal, and here is how people come back from it. A carousel of new gear said the opposite.

The app showed Jordan the same product grid both times. So would Muse. Neither one could tell that the person on the other end had gone from hope to doubt.

In the workbook, Jordan's story comes from Trail & Summit's own research with its members, which found five modes: the Excited Planner, the Anxious First-Timer, the Frustrated Post-Failure member, the Social Organizer and the Depleted Returner.

The most important finding was an empty column. Trail & Summit had zero designed responses for the Depleted Returner, in any situation. That column sat right on top of the people joining the category fastest, the ones coming back after an injury or a long time away. The signals were already in the data. An 11 p.m. visit two days before a reserved campsite already signaled urgency. A search for "replacement" instead of "best" already signaled the mode. The mode had never been the design brief.

Those are emotional, social and aspirational jobs, and they sit inside modes. Muse doesn't support modes, which means it has no loyalty to the feeling.

A retailer that designs for the mode becomes the name a shopper gives the assistant next time. But the mode only makes sense inside a situation. Retailers that skip the situational work end up designing moods with no moment attached, and the agent wins the moment anyway.

Apps made shopping functional. Muse will make much of life functional. The calendar, the inbox, the reservations, the reorders: an assistant that remembers everything and works while you sleep takes them off your plate.

That sounds like a threat to retail. It is also the best chance in twenty years for retailers to matter more to their customers.

People aren't ready to hand everything over. Half of consumers are not comfortable letting AI handle a purchase end to end, according to Bain. Seventy-seven percent of shoppers still name stores as their top holiday destination, according to Salesforce. Even McKinsey, the most bullish forecaster, says agents have to preserve "the emotional, brand-driven experiences that build trust and loyalty."

People hire products for more than the task. In my research, every purchase carries five kinds of jobs, and the functional one is only the floor.

This is a volume argument as well as a loyalty argument. Assistants follow instructions. A shopper can tell Muse which brand, which store, which price. A retailer that does the emotional, social, aspirational and systemic jobs well becomes the name in that instruction. The person says "get it from them," and the agent obeys.

So the retailer that only competes on the functional job gets compared on price by a machine. And the retailer that does the other four gets requested by name.

The systemic job is different from the other four. It is the one Muse is built to take.

A life system is the set of tools, habits and relationships a person uses to keep a part of life working: health, family, money, home, community. To most people they are just how the week runs: the calendar, the shared grocery list, the running app, the budget, the group chat that plans the camping trip.

Muse wants to sit on top of all of it. It remembers everything, works after the app closes, and asks only for the final yes. That is a bid to become the life system itself, with every brand underneath it a supplier.

A retailer can't win that fight head on. It can earn a role inside a life system. In my Trail & Summit case study, the retailer had real standing in physical health, family togetherness and travel, emerging standing in financial management, and none in friendship. That map told the team where it could credibly show up and where it would be ignored. Standing is earned one domain at a time.

The retailer that holds a real role becomes the part of the system the assistant defers to: the gear expert the shopper trusts, the store the family already plans around. Muse can run the errands. It still needs someone to send them to.

Trust decides who that is. When the researcher I met tried Muse, she wasn't impressed. It seemed more interested in serving up ad-like recommendations drawn from her social feeds than in helping her get anything done. That's par for the course for a company that earns nearly all its revenue from advertising. Muse collects 31 of the 35 data types Apple tracks, against an average of 13 for other apps, and only 8% of consumers say they trust Meta with their passwords. Many people will hesitate to hand Muse the intimate details of their lives.

Don't mistake that for safety. OpenAI launched its own agent, Dots, on September 29, and companies with better reputations for privacy are right behind it. Meta's trust problem buys retailers time, and time runs out. The retailer that uses that time to be clear about what it knows, use it only to help, and protect it will be doing an emotional job the next assistants will have to earn from scratch.

The order matters. Modes only pay off once you know the situations they sit inside, and neither pays off if an agent can't read your catalog.

In the workbook, Trail & Summit checks ten touchpoints against the time value each should deliver, and eight of them fail. The biggest misses are its two defining moments, membership renewal and first-purchase onboarding, which run as background processes: an autopay line item and a shipping confirmation. Those are the moments an assistant will happily automate if the retailer won't make them matter.

Retailers that start at step 4 build experiences on guesses. And retailers that stop at step 2 hand the relationship to the agent.

Every retail leadership team should be able to answer three questions before the holidays.

I wrote Trail & Summit before Muse launched. Read it now and it reads like a rehearsal. Each challenge in this paper has a module that works through it on a retailer's own operations, with a blank worksheet for your team at the end.

Stone Mantel offers three ways in, from a weekend read to a working session with your team.