The Interface Moved on a Tuesday

In early September, Eric Allen, a retired tech executive in Carmel, California, asked Meta’s new agent Muse to go through the state’s unclaimed property records and look for money he’d forgotten about. Six minutes later it came back with about $730 in Amazon store credit the state had been holding for more than three years. Plus, a $272 refund from when he left Sprint. Plus, gift cards sitting in his Yahoo inbox.

He had it do the same for his wife. Another $800: closed bank accounts, health insurance claims, a possible class action payout.

Allen told Forbes the money won’t change his life. It was just the kind of chore that never makes it to the top of the list.

I did the same thing this weekend. Lowered my Verizon bill. Lowered a few others. Canceled a few. Well, I didn’t. My ChatGPT agent did. I watched it work through the calls with the customer service reps and get it all done while I worked on other projects I had going.

Now flip that story around. Every one of those dollars was sitting there because somebody was betting nobody would bother to look. A lot of businesses run on that bet. I mean a lot. Last week the market started asking how many.

Wall Street

Monday, September 21: Meta’s stock jumps 11% as Muse hits number one on Apple’s App Store.

Tuesday: Bloomberg reports the S&P 500 Financials Index down as much as 2.4%. Allstate and Charles Schwab off more than 5%. Expedia and Booking down close to 4%. Planet Fitness, of all things, down as much as 11%.

Goldman Sachs’ trading desk put a name on it and built a consumer inertia basket: AT&T, T-Mobile, Allstate, Progressive Insurance, Netflix, Paramount Skydance, Expedia and Booking. The common thread is money made on recurring bills, negotiable prices and add-ons, from customers who never pick up the phone to push back.

Citrini Research then asked the question I’d put in front of every healthcare board right now: how much do health insurers make simply because people won’t sit on hold for five hours to get coverage approved? If you think that’s hypothetical, ask my wife. I’m the one who works in healthcare, but she’s the one who navigates it for our family, and she’ll tell you the system is broken and takes way too much effort to get anything done. She would jump at the chance to hand an agent the tedious part of that job and get us what we should be getting under our plan.

How many people are using it

Fewer than the stock moves would suggest. The Information, citing Meta’s internal data, reported just over 500,000 people tried Muse in week one, 250,000 of them daily. The outside trackers are higher now, and they don’t agree with each other. Apptopia Inc. counted 2.8 million installs in the first 12 days and put U.S. daily mobile users at 642,000, versus 231,000 for ChatGPT at the same point (ChatGPT launched on iOS only, to be fair). Sensor Tower says 3.4 million downloads. Appfigures says about 2.3 million.

So do some simple math, using the people using it rather than the downloads. Apptopia puts it at roughly 650,000 daily users in the U.S. Say each of them claws back $200 a year. That’s low next to every example in the Forbes piece. It comes to $130 million. At $500 each, it’s $325 million.

Then think about what happens next. People love telling each other what they saved. You’re reading me tell you about mine. Every one of those conversations is an ad nobody paid for, and Sensor Tower already saw Muse’s daily users jump 27% the day of Meta’s Connect keynote. Get to 10 million active users at $200 and it’s $2 billion a year. Get to 100 million, which is small for a company that reaches billions of people, and it’s $20 billion.

None of that money comes out of thin air. It’s revenue somebody was booking, whether that’s a phone carrier or a subscription nobody remembered. Now it walks out the door.

And the money is the small part. Investors repriced whole sectors on a few million downloads. I don’t think that’s a bet on Muse. I think it’s a bet that the interface is moving (maybe they all read my book), and Muse is the first proof people can hold in their hands.

What moves when the interface moves

When I say the interface is moving, I mean something specific. In The Invisible Interface I call it a Personal Operating Layer, or POL: software that sits between what you want and what gets done, holding your memory, the permission to act for you, and reach across everything you use. You stop going to the apps, and the layer goes to them for you.

Ranjan Roy put his finger on what that does to a business on the Big Technology Podcast. Think about how much revenue depends on customers never really reading their phone bill or forgetting to cancel. An agent reads the five-page PDF and shops the alternatives. He made the same point about Expedia. It owns no hotel rooms. Its business is comparing, then getting paid to send you somewhere. The agent does the comparing now.

Bloomberg Intelligence described Muse and Instinct, the text-message agent that raised money at a $2.5 billion valuation in August, as toll collectors. The purchase still happens; the layer takes a cut. Its analysts ran a scenario where agents handle 5% to 10% of travel, ride-hailing and delivery, and incumbents lose more than $5 billion in revenue.

It’s funny, most companies still think of the interface as the website or the app. It’s where the money gets made: the sponsored listing, the default option, the retention offer you take because canceling is a pain, the brand you pick because you know the logo. Put a layer on top that does the choosing, and those levers all get weaker at the same time. The apps underneath are still there. They’re just suppliers now, to whoever owns the layer.

In the book I asked which parts of your business depend on customers tolerating friction, and how fast a POL can remove it. When I wrote that, it was a question for a strategy offsite. Last Tuesday the market started answering it, one sector at a time.

Block the agent or let it in

Amazon picked its answer early. It started blocking Muse on September 20, after Meta refused to leave Amazon out. Amazon’s stated reasons are about security: the agent didn’t identify itself as a bot, and it appeared to capture customer credentials. Meta disputes that.

Fine. But follow the money. Amazon did more than $68 billion in advertising last year. That business works because a human being scrolls Amazon’s pages and sees sponsored products. An agent can now read the page, compare, and either buy or move on. It has no use for the sponsored slot.

And look, Amazon runs its own agent. It’s called Buy for Me, and it goes out to other retailers’ sites and buys on the shopper’s behalf. Meta has played the other side of this too. It rewrote WhatsApp’s business terms to keep general-purpose chatbots, ChatGPT included, off the service companies use to talk to customers. That took effect January 15.

So, everybody guards their own front door. Always have. The difference now is that customers are getting a new one, and it belongs to somebody else.

What does the agent do when your door is closed? It goes somewhere else. The day after Amazon blocked Muse, Shopify opened every store on its platform to it. Walmart and Target helped Google and Shopify build an open standard for agent checkout. If the shopper stays inside the agent, Amazon’s wall mostly means Amazon doesn’t get considered.

Then there’s Expedia’s week, which I think is the most important data point in this whole story. September 22: it announces a tie-up with Muse. September 23: the stock falls more than 7%. It did what everyone tells you to do, right? It partnered. And it still got hit, because the partner now sits between Expedia and the traveler.

Being open doesn’t mean you win, because an agent that can reach you can also reach the alternative that’s cheaper or ships tomorrow. A team of researchers tested how AI shopping agents choose, and the findings should worry every marketer. The agents piled demand onto a handful of products and ignored the rest, and a model update could reshuffle who won overnight. They also marked down anything tagged as sponsored. Your ad budget doesn’t buy much from a buyer that discounts ads.

The innovator’s dilemma, for everybody at once

Clayton Christensen’s innovator’s dilemma is a story about a single incumbent watching a disruptor come up in its own market. Everything the incumbent measures tells it to wait. By the time the numbers move, it’s late.

Most leadership teams know that story. The playbook assumes you can see the disruptor coming in your market. This one sits on top of every market at once. It hit insurers, banks, phone carriers, travel sites and a gym chain on the same Tuesday.

And it isn’t only consumer businesses. Gartner predicts that by 2028, 90% of B2B buying will be intermediated by AI agents, with more than $15 trillion of spend flowing through agent exchanges. A procurement agent reads your specs, pricing and delivery terms, compares them against everyone else’s, and builds the shortlist before a human ever gets on a call. It has no memory of the dinner your sales rep bought. If your product information is a PDF buried on a marketing site, you’re not on the list.

Expedia opened up and got punished anyway. Amazon blocked, and Amazon has the scale to make that stick for a while. I do not believe most companies do.

What makes this different from every innovator’s dilemma case I know is the timing. Your competitors and your distributors are making this call at the same moment you are, and each of their choices changes yours.

The layer isn’t finished yet

One reason this is hard to call: the layer itself only half works. Muse and Instinct can act for you. Neither is a real POL yet.

Start with memory. Meta’s launch post explains how to make Muse forget things. It says nothing about taking what Muse learned about you to another provider. That’s cognitive rent in the making.

Permission is shakier. During Instinct’s private-access period, testers reported an email it sent without approval, instructions it followed from a planted email, and an inbox it kept summarizing hours after they’d disconnected it. Meta built a separate Sentinel agent to police Muse, and Tarek Sheasha of Meta Superintelligence Labs still wrote that “prompt injection remains an open problem in the industry.” Then Thomson Reuters reported Meta had quietly tested routing some Muse phone calls to human contractors.

And reach? Amazon just showed you it gets negotiated site by site, and that the agent simply routes around whoever says no.

Meanwhile the field is getting crowded. Google is testing its own agent. OpenAI may show its hand at DevDay on September 29: code spotted in ChatGPT points to an always-on agent called “o.” To be clear, that’s a leak. OpenAI hasn’t announced anything.

I think the unfinished part matters most. The market is already repricing incumbents on a layer that only half works. If it moves stocks like this while half working, I don’t want to be the company still figuring out its answer when it works.

Two questions for your next board meeting

If you run a company, these agents are coming to you whether you planned for it or not. My guess is most boards haven’t discussed it yet, which means the answer is being set by whoever configures the bot rules on the website.

Two questions: How much of our revenue depends on customers not looking closely? And when an agent shows up to look, do we want it to find us?

Harry Glorikian is the author of The Invisible Interface: How AI Turns Intentions Into Actions—And Who Wins ( Ideapress Publishing / Simon & Schuster , June 2026). He is General Partner at Scientia Ventures , an Affiliate Researcher at the MIT Media Lab , and host of The Harry Glorikian Show.

Sources

• Forbes, Sept. 25, 2026: AI agents helping people recover money (Eric Allen)

• Bloomberg via Yahoo Finance, Sept. 22, 2026: Meta’s Muse drags down stocks that depend on consumer inertia

• The Information via Techmeme: Muse passes 500,000 users in first week

• TechCrunch, Sept. 21, 2026: Muse outpacing ChatGPT’s early mobile launch (Apptopia)

• TechCrunch, Sept. 25, 2026: Meta puts its muscle behind Muse (Sensor Tower, Appfigures)

• GeekWire: Amazon blocks Meta’s Muse

• Bloomberg, Sept. 21, 2026: Amazon blocks Meta’s Muse from its retail site

• Search Engine Watch: Why Amazon is blocking Meta’s Muse (Buy for Me)

• Forbes, Sept. 21, 2026: Amazon blocks Muse; WhatsApp business terms

• TheStreet via Yahoo Finance, Sept. 23, 2026: Amazon blocks Muse, Shopify opens the door

• Allouah, Besbes, Figueroa, Kanoria and Kumar, “What Is Your AI Agent Buying?” arXiv:2508.02630

• Gartner, Oct. 21, 2025: Top strategic predictions for 2026 and beyond

•    Big Technology Podcast, Friday edition, with Alex Kantrowitz and Ranjan Roy

• StockStory via Yahoo Finance: Bloomberg Intelligence agent scenario

• 24/7 Wall St. via Yahoo Finance, Sept. 23, 2026: Travel booking stocks tumble

• Meta: Introducing Muse

• TechCrunch, Aug. 24, 2026: Instinct raises privacy and security concerns

• TechCrunch, Aug. 26, 2026: Instinct raises $350M at a $2.5B valuation

• Campaign US: Amazon blocks Muse (Sheasha quote)

• Reuters via BNN Bloomberg, Sept. 22, 2026: Meta tests a human concierge for Muse

• Spyglass: Why didn’t Google build Muse?

• TestingCatalog: OpenAI “o” always-on agent (leak, unconfirmed)