There’s been no shortage of announcements, partnerships, and new protocols in agentic commerce. But when it comes to real-world implementation at scale, we were still waiting for things to take off. Earlier this year, we decided to hit the pause button on the topic until something actually happened.

While progress has been gradual, the industry IS moving forward, and many of the practical challenges we’ve been discussing over the past year are coming into sharper focus.

In this episode of Payments on Fire®, Glenbrook’s Chris Uriarte and Drew Edmond revisit the state of agentic commerce, exploring emerging consumer interfaces like Instinct and Meta Muse, access to merchant catalogs and data, and the challenges of turning agent-driven shopping into actual transactions. They also share their firsthand experiences making an agentic purchase through Meta Muse, and the industry complexities that still need addressed.

 

 

 

 

Episode Transcript

Chris Uriarte: And welcome to Payments on Fire. I’m Chris Uriarte, a partner at Glenbrook, and it is great to have you back with us today. And we’ve got another episode of our Fanning the Flames series, where we give our esteemed industry guests a little bit of a break, and it’s just some banter between me and some of my colleagues here at Glenbrook.

And today joining me is my fellow partner here at Glenbrook, Drew Edmond. Hey, Drew. How are you?

Drew Edmond: Chris, it’s great to be here with you. Excited to talk about a very topical topic today.

Chris Uriarte: Very, topical today, and our listeners are, of course, very familiar with this topic as folks couldn’t stop talking about it for a couple years. But here we are at an interesting place today. We’re gonna be doing a little bit of status of what’s going on in the market with agentic commerce. And, it’s been an interesting year thus far.

And, here we are, I guess we’ve officially ended summer now, Drew. We’re at the start of fall, and we’ve kicked things off with the conference season. We’ve been really busy talking to a lot of folks at a few of the conferences that have popped up. And what we can say is that at the beginning of the year, of course, tons and tons of chatter about agentic, but agentic hasn’t really been a big topic of conversation, at least amongst payments people, over the last few months.

But I think things are changing, right? And we should probably share with our listeners that we initially had a number of conversations earlier this summer where we said we were basically gonna take a break from talking about agentic, and I think a lot of us would agree that this topic has consumed an incredible amount of bandwidth over the course of the last year or two.

And, to be quite frank, we’ve kind of been left in a holding pattern the past year when it comes to the implementation of anything really serious or at scale, right? So we, of course, have seen lots of different trials. We’ve seen lots of different press releases. We’ve seen lots of different partnerships from the typical players in this space, the Walmarts, the Targets, the Etsys, of course, big tech, Shopify, OpenAI, Meta, Google, and lots of others.

But I think it’s probably easy to agree, without too much controversy, right, that we’re still not yet at the point where consumers are regularly using agents to make purchases, at least successfully online, right? So maybe that’s a good place to start. I mean, would you say that I’m categorizing the state of play correctly?

Drew Edmond: Yeah, and I’m laughing a little bit because I think over, how long has it been now, a year and a half maybe of, talking about 1A and 1B with stablecoins for a long time. And even during that whole time period, we test things ourselves and we’ll talk about some of the things we’ve done lately. But we’ve been trying to test it for a long time and, even us, we’re payments nerds and we weren’t even able to get really great experiences or successfully complete an agentic transaction.

Chris Uriarte: We tried for months to get agentic transactions to work. We tried and we just couldn’t make it happen.

Drew Edmond: And if we can’t make it, no one’s gonna make it from a payments perspective.

Chris Uriarte: Right, exactly.

Drew Edmond: Not to toot our own horns, but we know how to make a purchase.

Chris Uriarte: Right.

Drew Edmond: But I think you’re right, and I think there has been maybe even to the level of frustration to a certain extent on this topic because, there’s so much hype, right? You go on LinkedIn or you go to conferences and you see the word agentic that’s plastered on every single business’ banner that they have at their booth to make sure that they’re not missing the wave of being an important component of this new important technology that’s gonna change the world when it comes to payments.

You have this explosion of consortia and protocols that were meant to start to build that foundation for what agentic commerce could be. And really this rush from providers to try to articulate how they’re on the forefront of enabling it, right? And I think no one was really clear on what it is. It was like, we know it’s generally something about an agent making a purchase for a human. That’s about as far as we all got as an industry, I think.

And the value proposition is emerging, right? I think the people generally get the concept, of course, right? It’s yeah, okay, a robot’s gonna go make a purchase for me. But why, and when, and how, and all these fundamental questions. And taking all of that from different perspectives, right? Like merchants think about that concept of “Well, what does that mean for me as a merchant?”

And the consumer is like, “Well, what are you actually doing for me that’s so hard for right now?” And so, I think as you go through all the stakeholders in the ecosystem, there still remains some questions to be answered around how this will and should evolve across kind of different use cases, different verticals, different domains. It’s an interesting time because we are starting to come out of it a little bit.

Chris Uriarte: So the question is, why now? Why today? Why take up our listeners’ very valuable time and energy talking about agentic here as we’re quickly approaching, believe it or not, the close of 2026. It hurts me to say that. And everybody we know is very, very busy and I would say that the answer to that is that we are actually seeing a little bit of progress and some interesting news in this area, right?

I would not necessarily chalk this progress up to being a big win for agentic commerce yet, but some things are definitely going on today, and they’re definitely serving as a proving ground for really how agentic should and I would say should not work as well. And, some of these things that we’re seeing are starting to reveal many of the practical complexities that you and I and the rest of our colleagues have been speaking about over the past two years, right? And quite frankly, some of this has been very, very interesting.

So I want us to hit on a few different topics today that covers this activity that’s been going on. First, I think there’s something that’s super interesting that’s starting to emerge, right? Let’s talk about the evolution of the actual consumer interface, the user interface itself, right?

So when we first started talking about agentic commerce, a lot of the focus was really around the consumer’s interaction with kind of like these big AI surfaces: Chat GPT, Claude, Google Gemini, of course. But I think what we’re actually seeing is now a bit of a battle for the consumer interface, right? And a number of different options are starting to pop up as these agentic solutions are starting to come to market, right?

So notably, I think the big announcement that we’ve most recently seen has come from our friends at Meta, and that is, of course, the big release of Muse. So lots of industry, mainstream tech press all about Muse. Of course, Zuckerberg has got up there and loves to demonstrate stuff like this. So lot of big bets being placed on Muse at Meta. So maybe can you talk us through a little bit about what we’re starting to see with Muse?

Drew Edmond: Yeah, absolutely. And I think the interface point is really important because, I wrote a short LinkedIn post a few, I don’t know, several weeks ago at this point, that said basically, until the consumer interface for agentic commerce proceeds beyond just like a blank MS-DOS screen that existed in 1985, we’re not gonna see much progress, right? Because that’s just not a great experience.

Chris Uriarte: Right.

Drew Edmond: When we’re thinking about discovery and shopping and intent and how we as human beings interact and buy things and interact with brands and companies and things like that, certainly there’s a component to that, and we’ll get into that with muse, but it can’t be the only way that we do this. It’s just not good.

So when we look at Muse, we actually see, if you get the desktop app, you see multiple interfaces within that desktop app. You have different tabs on the side, and I see this as like product discovery to a certain extent. And it’s a bit reflective of a moment in time of where we are today, where we’re coming from in terms of what consumers are familiar with, in terms of how they interact with AI.

Because again, that blank Chat GPT prompt, that’s still something that we all use a lot. We still have questions that we need to get answers to. We have documents we wanna drop in. There’s, of course, we’re gonna continue to use that form factor.

But we also, in that sense though, I think it brings up this notion of user-generated intent. I have to come to this blank screen and say, I wanna figure out like, what’s my new bike? I gotta figure out which bike I want. I’m gonna ask it a bunch of questions. It’s gonna bring some things back to me, and maybe at some point I can figure out how to make a transaction through that process.

Chris Uriarte: And that intent, I’ll just add, that intent is not yet natural to all users, right? I think maybe we’re getting closer there where, these days, of course, if we have a complex question, we’re getting trained to go to Chat GPT or to Claude. But I would say that the regular consumer is not yet at a place where they say, “I wanna make a purchase of something. The first place I’m gonna go to is Claude,” or something along those lines. I think it’s changing a lot, but it still has to be something that the consumer has to learn, right?

Drew Edmond: Absolutely. Absolutely. And that kind of takes us to the next pieces of Muse, where the next tab down is the feed, right? And this is a very Meta-like idea, right? We’ve all grown up and used Facebook feeds and Instagram feeds, and we have our feeds. We know our feeds. So the notion of a feed is not new to us as consumers, and this is an evolution of that concept, and even more personalized because now it’s starting to interweave these topics that you might be interested in, where you have that with a Facebook where you’ve joined these groups and you’ve liked these things, and so it’s showing you things that, it’s your algorithm, right? It’s a version of your algorithm.

Now we’re getting to the world where the Muse can, with the knowledge of you, because it has that, it’s starting to build that, that knowledge of you, it can generate intent to you. It can say, it can come to you, and it’s connected to my email now. It saw some email that says, “You’re getting charged for this subscription.” And it can ask me, “Hey, have you used this subscription lately? Do you want me to cancel this for you? Do you want to downgrade maybe to the lower version because I did some research on my own as an agent and I know these other options might be better for you because I know how you operate as a human being.”

Or it knows that you have items sitting in a cart over here and it can go finish a purchase for you. Or more complicated stuff where like “Hey, you just moved and you need to register your car, and do you want me to set up a plan to do that and secure an appointment with a smog checker and pay them upfront?”

There’s so much more that it can do now ’cause it’s building this context layer of who you are your data sources.

Chris Uriarte: So this feed is a far, far ways away from 30 cat videos just scrolling across my screen every time I open up Instagram or something along those lines, right?

Drew Edmond: I think if you tell them, if you tell Muse that’s what you really want to see in there, I bet you can figure out how a way to get that back in there because I don’t want you to lose that.

Chris Uriarte: I’m not gonna try it. I’m not gonna try it, Drew, because we all know how the algorithm responds. It responds pretty quickly.

Drew Edmond: That’s right. That’s right. You’ll be awash in kittens. Yeah, that’s right. But yeah, so we have this feed notion, which is a really interesting kind of evolution of how we think about interacting with agents and what that means for just, like, how we interact with our devices and all sorts of things like that.

Then the next one down, you’ve got the ideas page, and I think that one’s also very interesting because it’s really there to describe how you can use Muse, right? It has all these different categories and actions that you can take and things like that. Where we’re in this world where we know that consumers don’t exactly know how we’re gonna do this, right?

And even the product managers and the PMs that are thinking about these experiences and the consumer experience, they kind of don’t either. They have some ideas for, maybe what people will latch onto, what they’ve seen some of their users do, and they’re gonna pull that user data and say, “Hey, we’ve seen other people do this. Maybe other people are gonna like that too.”

That’s another way that a Muse or an agent, a similar agent can provide intent, right? It’s surfacing things to you that maybe you wouldn’t have otherwise done at that moment in time, on that device, in that channel, that’s gonna change your behavior, and it’s gonna change it in different ways that we’re not gonna know about until things start to crystallize a bit more.

And then finally, just within the Muse world, we have goals, which I think is fascinating because they’re really trying to say, “Here are these major aspects of your life, your health, your relationships, your finances. What do you wanna be doing? How do you wanna be doing, how do you wanna be behaving differently in these worlds? Do you wanna be working out more and getting your steps in? Do you want to be ordering from these types of restaurants because they’re healthier? Let’s connect your finances. Are you saving for retirement? Connect your bank, connect your wealth management”, and those types of things.

Now, all of a sudden, it’s not just I go to a chat prompt and say, “I’m looking for a bike. Help me find a bike.” It’s, “I know all of your goals in life, I know what your interests are, I know what’s in your email inbox, I know what’s on your calendar.” That’s a pretty robust profile of a human being, and what that might unlock from a commerce perspective.

Chris Uriarte: So it might be a different route to that bike, right? So, in the early days where we’ve got product search and discovery, it’s like going to Chat GPT and saying, “I wanna buy a bike.” A sort of more updated version of that might be Muse understanding your health goals, right? And eventually getting to the point and saying, “Hey, I know you live in a–

Drew Edmond: You should buy a bike.

Chris Uriarte: Yeah, “I know you live in a warm climate environment. You’re trying to burn more calories, blah, blah, blah. Have you thought about buying a bike? And maybe these are the types of bikes that you should be looking at,” right? I did see those default categories, health and relationship and finances and things like that.

I wasn’t about to click on the, do you need help with your relationship button. I think that opens up a whole can of worms. I’m not sure if we’re ready for that yet, but it was interesting nonetheless.

Drew Edmond: Exactly. Exactly.

Chris Uriarte: Yeah, great.

Drew Edmond: But this is the context layer, right? This is where we can see how agents could start to do things. I think a lot of us have been reticent to think about or reticent to say that agents can be doing a lot of things autonomously for us, and I think there’s still a lot that has to be done to get to the point where we think an agent could go do things where we’re not really super involved. I think there’s still going to be a lot of human in the loop here.

Chris Uriarte: Yeah.

Drew Edmond: But you’re starting to see where this kind of amassing of data of you as an individual can at least unlock some unique experiences that to this date have not existed.

Chris Uriarte: Yeah. It’s quite interesting. And I know, we haven’t really touched upon payments or really agentic commerce yet, and I wanna talk a lot more about its agentic capabilities. But let’s get back to that and, we’ll talk about that in much more detail in just a bit. ‘Cause another service that we’ve seen arise, just in really has been getting a lot of headlines over the last couple months, has been this service that’s called Instinct, right?

So this has been getting a lot of press, and of course, the pocketbooks in Silicon Valley are wide open to fund companies like this. I think I recently read that they got another billion dollars worth of funding in just a few weeks ago, and I think that brings their valuation up to about $10 billion right now.

But instead of having a dedicated app with multiple interfaces and communication channels, right, Instinct, I believe, is entirely SMS and voice-based. Is that right? So tell us a little bit about how that works.

Drew Edmond: How I think about it is, it’s a little bit, they’ve come at it from a slightly different angle, right? I think with Muse, it’s a very Meta infused kind of visual experience, right? Like going back to the feed, right? They’re thinking about how individuals, how consumers interact with their applications and brands and things like that in their ecosystem and they’re drawing on that experience to design what they think the right consumer experience is to take advantage of the capabilities that these agents can do now.

With Instinct, I think how they think about it, and I’m speaking, I don’t know them, and I’m just speaking from my interactions with it, it’s very much like, Hey, I’ve got a concierge in my pocket. It’s in my contacts book. I could be, it’s not that I’m sitting at my desktop and I wanna do some research on something I wanna buy.

It’s that I’m standing in the Starbucks line waiting to get a coffee, and I’m like, “Oh crap, I forgot to get that reservation on Thursday for work, the work thing. I don’t even know what restaurant we can get and what’s still available.” I can just text or call my Instinct buddy and say, “We’re gonna be in Greenwich Village, 8:00 PM, five people. Just tell me what’s available and come back to me.” Right? You can do it on the fly, you can text it, you can do a quick phone call, and I think it’s more about kind of being there wherever you are, rather than being in a specific place you need to be to take an action.

Chris Uriarte: Yeah. Drew, I’m gonna date myself here because everything really comes full circle. I remember prior to smartphones, Google had a text interface, is you could text the short code G-O-O-G-L, and I remember for probably about three or four years, I was walking around New York City, and I would, you know, couldn’t remember where a bar or a restaurant was, and I would text and I would say, “Address for Frank’s East Village, New York City,” and it would text you right back.

And it was a simple interface. It got its job done. It was reliable and it was no frills. So, hey, here we are again, right?

Drew Edmond: Yeah, and I think it’s interesting seeing some of the discourse online these days. Some people that I think are skeptical still about this type of evolution, if you will, and kind of interface and what the capabilities are. It’s like, Well, how hard are things really? Is it really that hard for me to go off and get a reservation? Or is it really that hard for me to book a flight or whatever it might be.

I think my answer to that is probably, there’s a time and a place for certain things. There’s, from the Instinct perspective, it’s, to me, being able to do something like that while standing in line getting a coffee is easier than getting that coffee and now I gotta go sit in a chair and look through Resy and do it all myself. Oh, I didn’t find anything. Now I’m going to OpenTable, trying to do all this stuff manually.

It’s like, is it hard? No, I know I can do it, but to be able to have something that goes out and does a lot of this stuff and facilitates it for you behind the scenes while you’re doing everything else that you need to do in life, to me, that’s an improvement. Maybe those are basic use cases, but there’s gonna be other ones that emerge too.

Chris Uriarte: Yeah, for sure. This was actually a point, a very similar point that was made today by a reporter by The New York Times. Just by coincidence, the daily podcast from The New York Times is on this topic as well. It’s focused on muse and these interfaces. It’s a great episode that’s called Call My Agent, which I thought was pretty funny

And that’s a point that the reporter who had been testing Muse for a while has made as well, is that same question about can I just do this on my own, right? But yeah, I think things are changing for sure. I think what’s really interesting here is these agent-driven systems are essentially saying you don’t have to be living in a Chat GPT session to take advantage of these agentic capabilities, right?

They essentially wanna integrate with the channels that you use the most every single day, and probably more importantly, the ones that are probably already really critical to how you function every day, right? Social media feeds, SMS texting, WhatsApp texting, things along those lines. People live in these applications every day, and their brain is just trained to use them. So it’s really interesting to see how this is evolving.

Drew Edmond: I’ll tell you just a recent anecdotal experience I had that I thought was pretty cool. So I’ve been signed up for Instinct. I’ve given it access to my personal Gmail, which is a little bit scary, but it’s where we’re at these days. And I’ve just moved across the country and I’ve been dealing with movers and all these things and getting kind of updates via email from the moving company. And then the movers come, and I was giving it information, “Oh, some boxes arrived damaged. What’s the process for filing a claim?”

And it’s read every email interaction I’ve had with the moving company. It’s read every document that I’ve signed end to end, and it cropped out exactly the line that probably would’ve taken me eight minutes to find that says, “You can make a claim even after you’ve signed for the inventory,” and all these things. That just, it gave me peace of mind at the time. I didn’t have to spend the time doing it. I know it’s not, this isn’t a payments thing, but that was amazing for me. That was really cool.

Chris Uriarte: Yeah, that’s really cool. We’ll be keeping an eye on how these interfaces evolve for sure. So I wanna address another big topic that we’re seeing a lot of headlines about, and this is essentially the fight for agents to get access to merchants. Now, we’ve been talking about this for a very long time, but I’m starting to read things in The Wall Street Journal and The New York Times about this, so you know it’s really become a problem. This has been raised when it hits the mainstream press.

And what we’re really talking about here is the access to merchant websites, merchant data, their inventory selection, and of course, ultimately the ability for an agent to make a purchase directly on a merchant’s site without the consumer actually facilitating that, without the consumer actually visiting the site, right?

And I think what we’re seeing now is that this has become a challenge. This is really becoming a challenge because of all different concerns and policies and different behaviors that we’re seeing from different retailers out there. So maybe you could just take a step back and tell us about some of the models that we’re seeing out there right now.

Drew Edmond: Yeah, I think this is all a reaction. Every business has to interpret whatever change is happening in terms of the impact on their business and the impact on their relationship with customers and their economics and their revenue streams and all that. And this is no different. In fact, maybe it’s more acute than ever.

For example, we have someone like Amazon who has been, they’ve really been pretty closed, I would say, right? They’ve put the walls up in terms of the types of bots that they want allowed into their system, what actions those bots could do. They have a massive, robust ecosystem that they are trying to protect. They’ve done an amazing job over the decades of building what they have.

And now they have this new technology where this agent can come in and they have less control over it. They don’t have the visibility that they want to have. There’s just so many questions around if they just let the doors wide open and let the agents run rampant, what does that do to a over a trillion dollar company?

Chris Uriarte: Yeah, let me just interject and say they haven’t been shy about this either, right? If you recall, they sued Perplexity over this issue, brought it to court, right, and they’re still being incredibly aggressive about this closed-door policy, right?

Drew Edmond: Absolutely. Yeah and so everyone’s gonna have their philosophy and the guardrails that they wanna put up, how they wanna think about, how they interact with this new technology as it continues to evolve. So that’s kind of the Amazon side.

I think we have, we see folks like a Shopify who has also been on the forefront of experimentation and connection, I would say, in terms of, Hey, we have this kind of massive portfolio of small, medium-sized online businesses, or even pretty large businesses sometimes that use them to get their wares out into the world. And it’s on us as a platform to enable new channels and new technology if that is what these businesses want. We wanna make sure that if people are going onto any interface and interacting with an agent that is out there looking for items that could be brought to these customers, we need to make that available to them.

Because it’s not easy for them and we have this long tail that we’re trying to support, it’s gonna be really hard to get these millions and millions of merchants overall in the world on board unless the platforms that they use are enabling this technology. So they’ve really been on the forefront there, and I think that’s a critical component to surfacing the small guy, if you will, compared to maybe the behemoths like an Amazon and some of the other folks.

Payments is always gonna be chicken and egg. If we have all these consumers, how do we get these merchants on there? And, a Shopify and all these e-commerce platforms are gonna be a critical component of that.

Chris Uriarte: Yeah, for sure.

Drew Edmond: Yeah, and then I think just to close that one out, I think, then we have the larger retailers that are being maybe more direct and thoughtful about, “I want to make sure that as this continues to evolve, not only do I want to participate in it, in terms of doing what I need to do on my end from a data perspective and risk and all the things that you need to do as a merchant to handle a new channel like this, but I wanna be involved and help shape it,” right?

I think some of the big brands that we’ve seen come out, like a Wayfair and Walmart and Best Buy, these are massive companies that sell a lot of stuff. And we interact with them as consumers and individuals a quite regular basis as a whole. And so they should be a part of this. They should help shape what these experiences look like. They should help raise some of the questions when it comes to how these experiences should happen, to make it work, right? I mean, the technology platforms like the Muses, the Metas and the OpenAI’s, they want this stuff to work. They want people using their products for it, but the merchants want this to be a good experience too if it’s gonna happen.

Chris Uriarte: Yeah, absolutely. And I think Shopify is an interesting one, because, listen, we know that Amazon and Walmart are marketplaces as well, but we’ll put those off to the side because they’re very unique in the way that they operate as a marketplace. But you look at Shopify as a commerce platform and we’ve always talked about that the enablement of agentic across that platform could be incredibly powerful, right?

And I’m just using Shopify as an example, but you’ve got lots of other commerce platforms that are out there, is the ability to take millions of very small and medium-sized merchants and quote-unquote, agentic-enable them, and doing it in a way that just opens millions of merchants up to the agentic world could be just incredibly powerful here.So this is why I know that, this is why Shopify has so much invested in this, right? Why they’re so interested in this.

Really interesting. Yeah, I guess a question that goes through my mind, though, is that if consumers really do wanna be shopping with these agents, if they do find value in agentic commerce, why does somebody like Amazon wanna block the agents? I mean, isn’t it just more traffic to their site? I mean, don’t they wanna go with what the trends are? Don’t they want consumers delivered to their doorstep? I mean, what’s the motivations here?

Drew Edmond: I think that if they were given confidence that unlocking this would only lead to net new revenue, they’d say, “Great, let’s do it.” But I think the reality is with someone like Amazon, think of the man-hours, the people hours that have gone into constructing the consumer experience that they have on their front page, their personalized recommendation engine, how they, you know, recommendations within what you’re searching for. All of a sudden, my interaction with Amazon is I’m trying to buy some socks and Muse is showing me a picture of some socks that happen to be on Amazon, right? That’s it.

And what they want is you’re on amazon.com, you’ve typed in socks. It shows you the ones that they decide that they want to highlight as the priority socks for Drew, because they know who Drew is and they think that I’ll buy these ones. And if I don’t like those, here’s some other great ones that you have a choice of. By the way, last time you were here, you were looking at this thing, and I want to make sure that you don’t forget about that, so here’s this thing as well. That’s a very, that’s an experience that they’ve they continuously optimize over time, right?

And now all of a sudden, this agent’s doing that. My agent isn’t necessarily looking at all that stuff and, caring about, “Oh, by the way, let me bring back… You just asked for socks, but I’m also gonna bring back this basketball that you looked at two months ago on Amazon.” That’s probably not gonna happen, ’cause that’s not a great experience within that experience.

Chris Uriarte: Right. Exactly. I mean, we’re talking about essentially a loss of control, right? And retailers like to be in control of the consumer experience. I think control is really a huge point here. There’s really a very big difference between letting some rogue agent that I don’t have a relationship with scrape my website and own that entire consumer experience versus deliberately integrating with a known agent, right?

If I’m Walmart or I’m Best Buy, presumably I’d rather give the agent an API or some sort of connector where I can control the catalog, I can control the inventory, I can control how it’s presented, the pricing, what that ultimate experience looks like in conjunction with the agent. And probably also put some very important requirements or restrictions in place as to what the agent can or cannot do.

And of course, I’m not addressing the commercial aspects of this, which is a whole episode in and of itself. Who’s gonna make money from all this? How does that work? I’m also not addressing the whole risk and fraud side of this as well, which is another thing that we know, this is another reason why merchants have been hesitant just to open up their door to other agents. Again, another discussion for probably an entire episode as we continue to evolve. These are big, hairy, complex issues here, right?

So listen, I think now is a good time for us to share some battle stories, battle scars maybe, from the field, from some of our recent testing here. And with all the headlines that have been associated with Meta’s release of Muse, we of course couldn’t help ourselves, so Drew and I decided to give its agentic commerce features a try.

So Drew, why don’t you go first and share your experience? So the first big question is, did it ultimately work?

Drew Edmond: That’s actually funny because, like I said, I did just move across the country, and I don’t have a key to my mailbox yet, so I don’t even know if it’s here yet.

Chris Uriarte: Okay.

Drew Edmond: Yeah, but I’m assuming it’s sitting at the post office. But yeah, so it did technically work, to the point where my money was taken from me. So for all intents and purposes, the merchant has been paid as far as I know, and I’ve been charged. I did attempt to buy some delightful, low ankle white socks. To no one’s surprise, there was nothing available on Amazon when I tried to do that via Muse, and so it surfaced, direct, to the merchant. I went straight to hanes.com.

Chris Uriarte: Okay. Good old Hanes. Okay. All right.

Drew Edmond: Good ol’ Hanes, tried and true. One thing that jumped out to me immediately, you know, given, I think many of us are probably Amazon Prime subscribers, and I said, Man, just for this pack of socks I gotta pay, I think it was $8 in shipping, just for 1 pack of socks.

I’m like, wow. So that alone, it’s like, okay, this is already a kind of a strange experience for those that are used to at least your commodity regular purchase things being shipped to you for free in 30 seconds from Amazon, right?

Chris Uriarte: Yeah.

Drew Edmond: And then, as you’re going through the process, it gives you the opportunity to expand out the browser that all these actions are being taken on. So you can see the agent actually doing what it does. It’s not very fast, I will say. I would say, you see it slowly clicking on a field, guest checkout, going through the guest checkout field by field, entering the information that it knows about you. If it doesn’t know about you, it goes back to you and asks questions to make sure that it can fill those in, and it goes back and fills those in.

So when it comes to conversion and speed of conversion and speed of payment, we’re starting from a pretty low point, I would say. I think the payment aspect is really interesting. With Muse, they’re using Stripe Link. We know today that they are partnering with ShopPay and PayPal, and I’m sure they’ll continue to think about what’s the right way to expose payment credentials in the right way for folks. So that’s gonna continue to expand. But my typical experience with Stripe Link is you’re on a website that uses Stripe and has Stripe Link available, it’s pretty much a one-click checkout at that point, right?

Chris Uriarte: Yeah.

Drew Edmond: Talking about we’re filling all this stuff out that Stripe Link should already have and that Muse maybe already has, and it’s doing all this stuff from scratch. It’s using a virtual card on my behalf to make that particular transaction occur, and then it’s going back and charging one of my funding sources that I have associated with Stripe Link.

And so I checked my card statement, my American Express card statement after the transaction occurred, and the statement descriptor was Stripe rather than Hanes. It was that way for days as a kind of pending transaction at a different amount than I thought I was paying for that particular transaction, ’cause they still needed to figure out maybe taxes and some other fees at the time. But they wanted to make sure my card had the funds available at the moment in time so I think they just added a number.

And I tried to break it a little bit to be fair, ’cause the happy path is great, but the sad path is a little bit more interesting from a payments perspective.

Chris Uriarte: Sure.

Drew Edmond: So I went into the browser while it was doing its thing, and I just erased one of the address lines, and it just spun its wheels. Just spun its wheels for a while. It was trying to confirm the payment. It didn’t really know what was happening. It figured it out on its own to give it it due credit. But it did take a while. It reinserted my address on the wrong address line.

Chris Uriarte: Okay. Oh.

Drew Edmond: Then it reinserted the address on the right address line, but it didn’t delete the one on the second one. So it was, so I ended up having to go in there, clear things out, and do a manual process to complete it. But, so I would say, at day zero here, right? Inning zero, day zero, minute zero of really trying to figure out how this stuff is gonna work

Chris Uriarte: Okay. All right. Well, let me contrast that with my experience. So where you went out of nowhere and you said, “I just wanna buy some socks,” and you tried to break it, I tried to spoon feed it and give it a use case that hopefully would, I was too nice to it, right?

So I had a very specific item that I needed to buy. It’s a piece of a part for a piece of scuba diving equipment. And I know that this part is only sold by two merchants online in the US, and both those merchants happen to be Shopify merchants. Their store is online with Shopify. So we know that Meta has a relationship with Shopify. We know that, Muse will tell you that it could use Stripe Link or Shop Pay as a wallet to make a purchase for agentic. So I figured that would be a great way to force the happy path for it.

So what I did was I gave it the specific part number. It came back right away, and it recommended one of those two, well, one of those two merchants who I’ve bought from many times before. Shout out to the folks at Cave Country Dive Shop in High Springs, Florida. Great guys.

So I said, “Yep, we’re gonna go ahead and make that purchase.” I probably purchase something from them maybe once a month. So I have a long history with these folks. It brought me through the process of linking my Link wallet. That was very straightforward. Had a pop-up SMS text. My wallet was linked.

But interestingly enough, I said, “Well, it’s a Shopify merchant, why can’t I use Shop Pay? I use Shop Pay all the time as well.” And it came back, and it just said that Shop Pay was not available for my account. So I’m not quite sure what that’s about. But so here we are with a Shopify merchant using Link while we have Shop Pay available. So I’m not quite sure what’s going on there.

Like you, it was a little bit slow. If I was gonna do this myself on the merchant website, it probably would’ve been done in about one minute. This probably took five or six minutes to complete. And it wasn’t without a little bit of trouble. So when it was about to check out, it told me that it had chosen expedited shipping even though there was free shipping available.

It explicitly knew that free shipping was available and told me about that, but said that it chose expedited shipping because, quote-unquote, “speed of delivery.” I told it that speed of delivery was really important for dive gear. So I found that to be very interesting because I never told it that.

And I actually went back to it, and I said, “Well, where did you hear that? I don’t think I ever told you that.” And it was honest with me, and it came back, and it said, “I’m sorry, I made that up. You didn’t actually tell me that.” So I think what’s really interesting here is, we’ve been talking about hallucinations, hallucinations in large language models, these hallucinations make their way into the agentic commerce experience as well.

So I thought that was really interesting. So I went back to it and I said, “No, I didn’t tell you that. We want to use the free shipping,” ’cause I know I’ll still get it in two days on the free shipping. It went back, it changed it. First it told me my payment was declined for the one with expedited shipping, even though I know it didn’t actually send it out for authorization. So there was a little bit of confusion there.

But then it went, it completed it. It was successful. I went on the merchant site. It was there under my account in my order history. It looked like just any other order and, unlike you, Drew, who still don’t have your socks, is in about three business days, that part that I ordered showed up at home.

No issues. It was slow. It was a little bit clunky, but you know, at the end of the day, it worked, right? And I think what starts to get really interesting here when we look under the covers from a payments perspective, there’s some weird stuff going on, and you’ve highlighted a few of them, right?

So the first thing is on the payment credential itself. I think what we saw, right, is that Stripe Link is generating a one-time use card that’s actually being presented to the merchant. So that’s different from a regular use of Stripe Link at the merchant, right? If you go to a merchant and you have a card on file, the merchant is gonna see that card that you have on file in the Link wallet. But in this case, I think both of us saw a one-time use card that was, used to purchase at, so I think that’s interesting.

Drew Edmond: That’s right, I don’t think anyone wants that.

Chris Uriarte: Yeah, I don’t think anybody wants that either. That causes some confusion. So when I went into my order history, it said, “Oh yeah, it was purchased with a Mastercard ending in 0506.” And of course, I don’t own a Mastercard ending in 0506. So we could all imagine if there was a customer service issue and I had to call them up, and they asked for my card number, that might be an issue there.

The other thing that you talked about was the descriptor. While it was pending, you just saw Stripe Link. You didn’t see anything that related to Hanes. Ultimately, I think you saw Hanes on the final descriptor, right, as I saw with the dive shop, with the folks at Paragon Dive Shop, I saw that there.

But a couple weird things, right? So we saw in both cases the MCC was Stripe’s MCC. Like it was like computer software was the purchase, right? And I’m buying a dive part and you’re buying socks, definitely not computer software. The other thing that we saw was that on the descriptor, it was actually Stripe’s address and phone number. It was not hanes.com address and phone number. And I think what makes this even a little bit more disturbing/funny is we tried to call the Stripe phone number that was on the descriptor, and it went to a voicemail box that said, “The mailbox is full. Have a nice day.”

Drew Edmond: That’s not surprising.

Chris Uriarte: It’s not the best experience, and there’s a lot of clunkiness to be worked out, right? This is definitely a lot of clunkiness. I think there’s a lot of potential issues that could pop up with this architecture, but we know we have a long ways to go, right?

The descriptors aren’t great. The phone number is not the merchant phone number. The credential that’s being used to purchase the goods at the merchant is not the card holder’s card number at the end of the day. All of this, of course, can lead to customer service issue, chargeback problems, things along those lines.

And I think this is a particular challenge when you’re trying to build, I, guess what we would categorize this as an all-purpose agent, right? So Muse is, we would say, like an all-purpose agent that’s trying to serve any shopper, trying to get it to work at any retailer that it can, right? So you have to create this generic model that works for everybody, as opposed to having a very, tight proprietary integration or hopefully in the future, some sort of standards-based integration with a retailer.

But, you know, we’re seeing a couple different models here. What are some of the things that you’re seeing out there?

Drew Edmond: Yeah, I think you touched on one being that all-purpose agent, and I think it starts to raise some questions about how the industry adapts to that. You mentioned standards, which I think will certainly be some component of it, right? In terms of how are gonna do this in a repeatable fashion that doesn’t require bilateral integrations from every retailer into every agent that exists.

Chris Uriarte: Right.

Drew Edmond: Are there going to be new layers that emerge, almost agentic gateways that can help streamline some of this stuff for folks? That’s one for sure. There’s also, I would say, a almost like a vertical or a use case-focused approach here where think about grocery, for example, right?

Chris Uriarte: Yeah,

Drew Edmond: Is it that I want Instinct to replicate everything that maybe I’ve done with Instacart, or does Instacart own the grocery agentic experience because they already have the nodes with all of the endpoints and they understand how people interact with groceries, and maybe they can provide a better experience?

Is it that Instacart has a partnership with a Muse or an Instinct to make that better so that you don’t have to, so that each agent doesn’t have to build bespoke interfaces and experiences across every domain exists in the world? A lot of interesting questions that emerge out of that.

Chris Uriarte: Instacart almost serves as a vertically specific aggregator that then could perhaps interface with different agentic platforms. Yeah.

Drew Edmond: I don’t see why not. Yeah, I think that would be really interesting.

Chris Uriarte: Yeah.

Drew Edmond: And you can take that concept and say “All right, let’s go to some other use cases and verticals and see how that, might apply,” right? I think, ’cause there’s these different components. If you think about retail, is it that I have this all-purpose agent that is really good at retail? Or does my all-purpose agent interact with a really good retail agent that’s for Drew. So I think that’s going to emerge, and people will probably experiment with different models in that space to see what ends up working best.

And then you just have the, I would say maybe more of the walled garden approach where, and this is already in existence today, I would say this is almost an extension or an evolution of kind of personalized recommendation engines for a place that you’ve already spent a lot of time on, right?

Going back to the Amazon or a Walmart example. You got Rufus and Sparky and all the dogs that are out there helping us within the ecosystem of a marketplace or something like that.

Chris Uriarte: The dogs are always very well-trained at walmart and Amazon to help us there. And yeah, I would put them more in the category as like shopping assistants, right? Yeah, they’re agents. Yeah, there’s headlines associated with these that are taking advantage of the agentic buzzword right now. But at the end of the day, they’re like shopping assistants.

They’re getting a lot better. We were talking about this with a bunch of merchants and service providers at the MRC board meeting. We were doing a workshop in San Diego a couple weeks ago, and I have always been using, the last two years, this what I call the Julie Ferguson birthday party test, where you know, she should be able to go to a agent and say, “I’m hosting a kid’s birthday party. What’s all the stuff that I need?” And it should be able to figure all that out, and you should just hit buy. Two years ago, that was really…

Drew Edmond: He likes dinosaurs, get me all the dinosaur stuff.

Chris Uriarte: Exactly. Two years ago, that was really difficult, and we’ve come back, we revisited a couple times, and everybody in the room this time said: “Oh, you know what? This stuff is actually getting better.” Like the Walmart agent Sparky and the Amazon agent Rufus is actually getting a lot better than it was just six months or a year ago.

So, I think they’re special purpose, but they’re a little bit different from the scenarios that we were talking about with Muse, right? I mean, do you think that any of these are gonna win out? Are they gonna coexist? Like, how, do we look at this?

Drew Edmond: I think that there probably is no one winner. I think I’m very curious to watch kind of the evolution of middle example of kind of the vertical use cases and how they interact. I think everyone that is willing to connect their data sources is going to, will ultimately have an all-purpose agent, right?

I think that whether it’s in two years or in 20 years, the reality is this technology’s rapidly evolving so fast, and we’re already starting to see some of the cool things that it can do, that it’s just gonna be a part of your life, is that you have an agent that knows everything about you so that it can help you navigate your world. So all-purpose agent will win to the extent that I think everyone will have some semblance of that.

Now, when it comes to crafting great experiences for various use cases and domains, that’s where, can something be everything, versus do we still have specialized labor in this world where people craft better experiences than other folks because they know a lot about it, and therefore there’s some combination of those things together.

And then there’s probably still going to be times where I know I wanna go to Amazon. I don’t need my, maybe for whatever reason, I want to go to a particular brand. I want to have a brand experience and I don’t think we’re gonna abstract ourselves as humans from everything out there.

So there’s still going to be this desire to have some sort of connection to companies that you care about, to companies that are doing things that you like, that have the design aesthetic that you like, that have the ethical standards that you like, whatever it might be, right, that you still want to have some sort of interface. So there’s, I think, all three in some fashion will likely exist,

Chris Uriarte: Yeah, I would agree with you on that. But listen, clunky or not, any of our listeners can go and they could make an agentic purchase right now. It might not be smooth, but you know, we’ve demonstrated that you could do it. But we talked about a few sort of snafus in that process here.

So of course, this big question that’s been looming over the past couple years has been: What happens when things go wrong? What happens when the agent makes a mistake? Who’s liable for this? How are these mistakes handled? And I think we’re starting to see parts of the industry pick this up and trying to address this. So what are you seeing out there? What’s some of the talking points that we’re hearing?

Drew Edmond: Yeah. Well, first I just want to say, I think part of it is we do often, I think, as an industry, look to the card networks, especially Visa and Mastercard, to weigh in on this. I think we still need them to, at some point, to figure out from a dispute reason code perspective and how do we assign liability to these card transactions.

That being said, these aren’t always going to be card transactions. The US is not the only place that’s going to have agentic commerce. There’s going to be other payment methods out there. So other networks and other payment methods also have to figure out how they want to handle these. And this is a really challenging problem. There’s a reason why we don’t have an answer yet. It’s because it’s really hard.

And I think we saw it in your use case, right? The agent went off, the agent that you chose to work with went off and chose expedited shipping. You didn’t ask it to do expedited shipping, and if you hadn’t caught it and you just click okay, and maybe you didn’t read it, and then you see that on your statement, you’re like, “I didn’t pay for expedited shipping. I didn’t want expedited shipping. Do I file a chargeback now?”

I mean, we know consumers are going to do that. Even if they shouldn’t, they will, right? In that scenario, well, is Muse responsible? Are you responsible because you didn’t read it? Is the merchant responsible because they didn’t verify something? I don’t know how to answer that question.

Chris Uriarte: Yeah.

Drew Edmond: I think the reality is, and maybe this leads to, I think, from the merchant perspective, constant growing concern over how consumers use chargebacks to interact and to get refunds and those types of things, which then leads to fines and VAMP and all these complications on the chargeback side.

Well, now we have agentic-friendly fraud, if you will. You have a Stripe link saying the agent is acting on behalf of the consumer. You see Instinct saying, “We’re not responsible for anything that’s unintended from an agentic action. You have to go figure it out on your own.”

You’ve got an American Express saying, “Well, the cardholder isn’t liable if the agent did something different than what your intent is.” But I could talk for another hour about how hard it is to determine what actual intent is.

Chris Uriarte: Yeah. How do we determine that, right? How do we determine that? Yeah. Yeah. Yeah.

Drew Edmond: I think it’s gonna be one of the hardest problems to solve for the ecosystem that I don’t think there’s an easy answer to it. So that’s where we still are on that front.

Chris Uriarte: So I think in the Western world, right, where we’ve made it clear how mature we are here right now. But unsurprisingly, right, there are some places where we’re seeing some agentic activity happen at scale. And of course, China, some parts of Asia, comes into focus right away. Just very quickly, anything interesting going on there that you wanna highlight?

Drew Edmond: I think it’s a fascinating juxtaposition, if you will, in terms of the different kind of Eastern versus hemisphere models, not very often I get to use juxtaposition on a podcast, but I’m going for it.

The reality is that in Asia, they live in a world of super apps, right? And that means that you have a centralized platform that the consumers and that the merchants are on. When you have that, the ability to make decisions and make policies and get everybody using the same kind of technology and expectations around how this is supposed to use is just much easier

We live in a much more fragmented world in the West, where we have different card networks, different payment method types, different providers. Everything’s very fragmented, and there’s no kind of clear centralized leader on how this is going to work, right? It’s just the reality of the situation. And so they’ve been able to roll things out much more quickly because it’s a centralized decision-making body that can do these things and get the ecosystem on board in that type of way.

I think some interesting things that are emerging that can still inform us. Given we have such different models, it’s hard to say, “Well, because they did that, we can do this.” It’s like, well, maybe it’s different enough that we can’t, and that we’re gonna have to think about it a little bit differently. But some interesting things around, like agentic identity and conversations that are happening in Asia and in India about kind of registration for agents, right? I think there’s a major component about on the merchant site, you know, how do I know that this is a good agent? There’s some protocols out there and things like that, but we may need some sort of registry to be able to manage this ’cause it’s just so complicated.

Chris Uriarte: Yeah, it’s very complicated. I mean, this has all been a lot to digest. To wrap us up here, are we making positive progress? Is this just more confusing? I mean, it sounds to me like we’re making some progress. We’re heading in the right direction, but we still got a long way to go, right? How do we think about this?

Drew Edmond: I think the reality is that agentic commerce is here. It’s still very small, but it’s here. The experiences that we had were clunky and not perfect, but what does that mean for a year from now, from two years from now, five years from now? Like this, it’s only going to get better. We’re gonna figure out some of these hard problems.

So to me, the agentic commerce era has actually arrived. don’t know how we see it reflected in the numbers and how merchants start seeing it reflected in their numbers, it’s gonna, it may still take some time ’cause we’re gonna still need to iron out some of the kinks in this process.

But I think we’re a lot further along than we were when we originally had kind of instant checkout and Chat GPT that didn’t work so well.

Chris Uriarte: I can assure you this is not the last that we’re going to be talking about this topic for sure. It’s great talking to you about it. I think, Hey, I think I’m going to see you in person tomorrow in beautiful Monterey, California. We’ll all be together here at Glenbrook celebrating Glenbrook’s 25th anniversary.

So I really want to thank all of our clients and all of our listeners for putting your trust in us over the course of the time. And we just hit our 300th episode here at Payments on Fire. It’s really incredible how that has come up so quickly. So again, thanks for listening. Until next time, do great work and have a great day.