S08, E04: Designing eCommerce for agentic discovery (with Rishabh Jain)

You know those channels your colleagues keep bragging about? The ones getting all the credit? Yeah, they might be doing squat. [music] Attribution makes every channel look like a hero, even when it's a zero. Incremental tells you who's actually [music] doing the work. It's like a lie detector for your marketing budget. Start using Incremental today. Get your demo at incremental.com. That's i n c r m n t a l.com. Mention that you came through the Mobile Dev [music] Memo podcast for a special 15% discount for the first 6 months.
The problem is that the distinction needs to be drawn between the competence [music] of the economists and the correctness of their analysis.
Welcome to the MDM podcast. I'm your host Eric Seufert. Today I am pleased to be joined by Rishab Jain. Rishab, I can't even count the number of times you've been on the podcast. Please introduce yourself, give some background

on who you are, but we do have a lot to talk about.
Totally. Yeah, so I'm Rishab. Before even starting this company, I've been a long-time listener of Eric, so it's so great to be able to come back with some frequency. So, thank you again for having me. The company we started 4 years ago is called Fermat Commerce, and what we do is we use consumer behavior as an underpinning and and how consumers make decisions for shopping by sort of making inferences based on how they actually shop on people's websites to then determine how to generate the appropriate intent-based personalized website for that person. So, that's the long story short.
Do you like that it's a meme on D2C Twitter that people don't know what Fermat does?
I will say two things about that. One is people only hate on success of some variety.
There you go. There you go. That's true.
Right? So, I think it started because we have been very successful at raising a lot of money. So, I remember when Scentbird our

series B, which was like we raised 45 million last year, that's when it sort of like picked up. It was like, "Wow, how does a company that no one knows what it does raising so much money?" And that's when you sort of like realize how the internet works, which is great. Like I have no I have no problem with that. The other side of it is there are components where in some weird way it has worked to our advantage because when I talk to people one-on-one and we're in in a sales cycle and they just see what we do, because they have this like preconceived notion of like, "Oh, it's going to be complex." But then it's like, "Oh, wow, like the magic is like you're able to create this inference on how people behave."
Right.
And that's like a not hard idea to understand. It actually sets me up for a very good sales motion. So, you know, it has worked out to our advantage. Yeah.
Okay, so I I didn't realize I think before I asked that it was kind of a sore spot. I I always interpreted as

being like tongue-in-cheek. I Whenever I see that, I kind of I hearken to the the Jimmy Kimmel show where he would say like, "And apologies to Matt Damon, we ran out of time." I kind of see it in the same vein. It's like kind of a friendly tongue-in-cheek Totally. jab. It's not It's not meant to be like an insult or something.
Oh, yeah, yeah, yeah. Most of the people who do it as like a friendly tongue-in-cheek. So,
Right, okay.
cuz I I know all of these people, by the way. They're all friends.
Okay. All right. I just want to make sure it's not like a sore spot that that you know, you genuinely
No. No, not at all.
All right. You are one of the shining lights of insight on e-commerce. You know, For Mot sits at the center of the e-com universe. Full disclosure, I'm an investor. A happy investor. I've got a number of people coming on the podcast in the next few weeks. I'm going to be interrogating agentic commerce and so I'm asking them all the same question at the start of the conversation. Am I wrong about agentic commerce being a mirage? Specifically, I mean, independent agents intermediating purchases across major retailers, e-commerce retailers. Let's put travel

aside. I'm distinguishing travel. I'm publishing on that today. I'm putting travel into a separate bucket. So, when I talk about agentic commerce, I mean agentic e-commerce. And then also specifically independent third-party agents, not platform-bundled agents. Am I wrong about agentic commerce being a mirage?
Okay.
[laughter]
So, right before we hit record, I said like we have to talk about Instinct and Muse. And the reason I said that is I think here's what I think. While it was ChatGPT and Anthropic and whatever else, I think that it was basically a crazy notion that someone would use like the ChatGPT UI and like shop on Amazon or shop on whatever else. Like versions of that have been tried so many times and it has like never worked, right? So, like Google Shopping originally and then Facebook shops and so on and so forth. I think there is a switch and you can

tell me now how it sort of plays into your agentic e-commerce thing more specifically, but with Muse, I use it to shop and complete the transaction on other sites. And I would not have expected that I would do that. Now, where of course it does not do that is on Amazon, which which I know we will also talk about, but there's a large series of things that I have started to purchase. So, for example, I used to only buy Marine Layer t-shirts. But I asked Muse, "Hey, I buy Marine Layer." It knows about my body type because I'm also using it as like a fitness coach. And so it was like, "What should I buy if I like Marine Layer that has a little bit more structure to it?" And it said, "Okay, buy this Buck Mason t-shirt." So, I said, "Okay, fine. Which one is the one that you would recommend? What size? And can you buy it for me?" And then it just did that. I did the same thing with jeans cuz I've only my whole life worn the exact same pair of Levi's.

Jeans because it's such a pain to figure out what fit is going to actually fit me. Right? And so I was like, here's exactly my fit. Here's exactly what my body type is. Can you find me an equivalent pair that's like cooler than Levi's? And it was like able to do that. And on again on another site. And so I think that this is actually going to start to happen. Will it happen through to a platform? So like through to Amazon? I mean, Amazon is not going to let that happen, but it is for sure happening for independent brands. And I I I could not believe it. Like now the average person who I talked to has made a transaction on Meow's. And previously the average person I talked to there was no way they were going to make a transaction on ChatGPT.
Yeah, I think I think that's fair. I mean, I think the question is scale and penetration. And so, you know, I totally accept that if you are going directly to the brand

then there's it's win-win. Like there's no there's there's maybe a little bit of an incentive alignment problem there because I think the brand would like to get your email, they would like to have a relationship with you, but if the agent's doing it through your browser and, you know, creating an account and registering, they don't lose that. You know, the ChatGPT integration was was a non-viable from the start. It was just a non-starter because of Well, ChatGPT's got the consumer relationship. And then the issue is when you tried when you're talking about going through a platform, I just don't I don't see that happening. But that's okay. That's That's good context and that's that's really valuable insight.
Yeah, exactly. So So by the way, I agree actually. I agree that it will not happen on a platform. I think the scary thing is what is the scale that it will happen with independent brands because actually the incentives are getting aligned because of this browser use issue.
Yeah.
So the underlying insight that I had was the browser use solves two problems that the ChatGPT integration did not solve

and I think completely shifts the e-commerce landscape. The first problem it solves is it solves a trust problem for the consumer.
[snorts]
So, when you interact with ChatGPT, the problem is you don't actually know or trust whether or not the information you're getting is correct from the underlying brand. When you're using Muse especially, you're seeing the browser interaction happen, and Muse does this thing of putting you in the loop for the browser use. And so, you actually fully trust it. It's like seeing a house with a real estate agent. The agent is there doing the work for you, but you're sort of alongside the agent. And so, you have full trust and visibility, which is why you're comfortable making the transaction. And you are actually interacting directly with the website. So, yes, actually Buck Mason got my email because my agent said, "Hey, did you know that you have to pay for shipping, but if I give them your email, you don't?"
Yeah.
And so, the whole thing is is actually like very different I mean, by the way, I agree it will not happen with the platforms, but I think the scale is very

possible.
Well, yeah. I mean, it it it's then it's a question of like what percentage of e-com are we talking about? Cuz if we are talking about almost We're talking, you know, so let's say Amazon's 36, Walmart's call it 9 to 11. It's a lot, right? And then, you know, going going directly to Levi's, yeah, sure, do it. Like that's the same with travel, by the way. Go directly to Why would I go through an OTA? Why? It makes no sense. The airline would prefer that I not do that. I think if you given the choice. Like they're not losing anything. Right? And same with Levi's. Same with Levi's, right? And so, then the question becomes though, and this is also part of the argument. Well, agent says, "All right, I took 5% of that." Well, okay, Muse takes nothing yet. But at some point, the agent's going to say, "Well, I should be getting something for this. I'm driving transactions. I'm making commerce happen. Like it's worth something, right?" So, how much is it worth? Levi's is worth 4% to you? And maybe Levi's says, "Yeah, okay.

I guess so. Sure, take your 4%. And then, Muse says, "Well, but what if I sent the person to some Japanese jeans that cost $300? Then I take the 4% of that. And that's more money. So maybe I should direct people to the Japanese jeans. Like let's say they're both equally suitable to that person. They're both equally relevant. There's just a price differential. Well, maybe I'm going to nudge them in the direction of the higher price thing cuz I'm going to get more money. And then Levi's says, "Well, wait a second. Don't do that. I'm willing to give you more than 4%. You're sure Yeah, that was arbitrary. Let me tell you how much I'm willing to give. And then you you use that to decide. Don't just by default pick the higher price item or the lower price item cuz you're more likely to transact or whichever it is, right? You They're They'll find some whatever that is, whatever the optimum of that of that equation is. Let me just tell you how much I'm willing to pay and you decide on that basis." And now we've got an auction. And now we've got ads. Totally.
Oh, totally. By the way, the reason I like

using Muse as an example is because we should we should have ads.
Right.
Look, there was a podcast where the Instinct founder came on and and was talking about more of a take rate style.
I'm familiar. I'm familiar.
And we will see if that works. I am skeptical precisely because of what you described, right? And it's not like affiliate fee or transaction model-based things are are new. Like we we sort of know Yeah, it's exactly what you said. We know exactly what happens when when you do [laughter] that. But the reason I'm so optimistic about Muse in particular is because you can have higher confidence that some of these fundamental problems are solved a priori. So they're like browser use thing and seeing the browser use happen is only possible in Muse, it's not possible in Instinct, right? Because it's an actual application, it's not a texting interface. The probability in how you serve ads or things like that Meta just has better ways to monetize various ad surfaces off of those

interactions. Right? Like they they can be more creative. Just like they have been quite creative with WhatsApp as a fuel into their ad system, right? Or as a service. And so I I I just think that there is First of all, I think they should have ads. And then I also think that Meta's far more likely to produce the right type of ad.
Yeah, versus like an Instinct and who you know, we'll see where Instinct goes, but like it's there's a it's very difficult to build this infrastructure and they've got it. They already have it. Meta has it. So all they need to do is apply it. Like I just think like every time you walk through like I mean and and and I'm just like two in the weeds on this and I'm too online about this. And I'm I'm you know, having the same conversation over and over and over again. And it's the conversation always resolves to my sort of opponent cuz it's usually a fight or debate coming up with the concept of ads from first principles. And it's just as as I just took you through it. It's like, well, okay, but what if there's multiple products can be relevant? Like maybe then then then you

know, the agent wants to get paid and so they can compute an expected value and they nudge you towards the one with the higher expected value cuz they're going to get paid and they they and so so it's the 4% and then they'll just determine what the probability of conversion is and they'll come up with the expected value and they'll nudge you in the direction of whichever has a higher expected value because they're getting paid on that basis and actually maybe one of those products would have been willing to pay more than 4%. And the way that they would have quantified that how much they're willing to pay is not as a percentage but as that's dollar value and then they could just tell you how much I'm willing to pay and then you've got an auction that you got a bid you've got an auction you've got ads. And I think that's where we go and that's that's part of the agentic commerce is a mirage idea. It because I think the way that VCs are pushing this is like, no, we hate ads and what the what the agent's going to do is it knows me and it's going to find me the most relevant product, right? Well, at the end of the day they're not going to do that for free, right? They have to monetize especially the agents, right? If you're you know, Muse is giving you a virtual machine, right? Like the they there's a cost basis here. It's not it's not like an app where like the marginal cost of

someone another person using the app is zero. Well, no, there's actually a cost they've got to recover. So the monetization imperative is I would say even stronger. Maybe not with Meta and maybe not right this second, but you know, at some point it's Meta. They're going to monetize it. And I don't think they're going to try to come up with a new model. I think they're going to use the existing model, and I think it's going to be ads. And I think ChatGPT already walked us through this experience. Let's hover here because you hinted at this very fundamental distinction between ChatGPT and their instant checkout implementation and what you actually get with a standalone agent. Talk Talk to me about that. Walk me through why that's a fundamental distinction.
Yeah, so it's funny. I was so dogmatic
[laughter]
that like people will not give their credit card to like ChatGPT to go make a purchase. And I I remember telling merchants distinctly, like the reason is

they will do their research on ChatGPT, but they will come to your website to verify what they have learned. And so, like in your purchase journey, sure, there's discovery, there's consideration, then there's inside decision. There's this important step of verification of the decision that you have made, and then you transact. Right? And the problem was always that if you try to keep everything in a place that is outside of the brand's home, and you just kind of pull the checkout into that other place, how do you verify that? And as humans, we are so used to that verification being like oh, I am on whatever, marinlayer.com, and therefore I know that when I like I'm looking at this shirt, it is in fact the medium one, it is in fact the blue one, and that's what I'm going to get. And so, then I feel confident hitting pay. And what agents have done, or Muse in particular with browser use, is it's giving me the information the

way that I would have gotten it from the website. This one, I think no one else has caught up to yet. In fact, I was on a walk with a venture friend of ours yesterday who we who we both know, and he was talking about like how we should have a feed that goes to the agent, and I'm like, no. Like, the exact problem is that everyone thinks it should be a feed that goes to the agent. And exactly that is what what should not happen. Like, what should happen, and what is actually, ironically, the controversial idea, is that you have to have a website. And that website is visible by the end user as the agent is interacting with it and telling the end user what they are seeing, what information they are getting, so that way the trust and verification is happening through the agent because the user knows that it's going to the website. And that's what computer use and browser use has made so magical and why people are just dishing their credit card, even me.

Like, I literally would not have done it. And I do it all the time now because I see that happening. And in fact, what it is telling us is like, okay, the human used to navigate the website in a certain way, but now that it is agent intermediated, the way that they're going to navigate is going to change. And that is the thing people are not thinking about. It's not like, hey, you need to have structured data, and you need to have a feed. It's like, no, no, no. You need to have a fundamentally different navigation approach for an agent intermediated interaction that does not look like the ways a navigation looks like today, but instead helps the person finish the transaction as they're doing it with the agent while maintaining trust. And that, I think, is totally unexplored, and we're going to see tons of innovation there.
What do you think that looks like? Is that just I mean, cuz I, you know, when I've got the agent, you know, I use I use Codex for a lot of this. And so, the the window of the the browser usage is

pretty small. Is that just like bigger fonts, bigger I mean, it could be as simple as that, just being just a just a just some something that is magnified so that you actually see what's happening. Or is that like is it more in terms of like the architecture of the usage going like left to right or something instead of top to bottom? Just talk to me about what that looks like.
I'll give you a good example. So, with apparel, the way that you used to have to communicate sizing and fit information when a human was interacting with the website is you would do things like, "Okay, you would show a model image and then you would say something like, 'Model is 6'2", 240 lb, wears size large, whatever.' So, like that's one way you're communicating sizing information. Then you would have a size chart that you could click into. When is interacting with the page, you could just say like, you could give an enormous amount of information. So, it's like, "Hey, trying to figure out what size?

Go here." And now what happens is it's like, "Can you ask the person what brands they already buy and what size they buy from those brands and how it fits them?" And it can present comparative information. It can present composite information. It can basically custom tailor, like it can tell you how well tailored it's going to be for you. So, for example, a thing that no apparel merchant does is they never send the factory specs directly onto the website. So, like if you're an apparel company, you have two different things in your system. You have a spec sheet, which you send to the factory in order to make the garment. And then you have the size guide, which is what you put on the website. But that's actually entirely unnecessary. You could put the spec sheet on the website. And the agent could understand how to read the spec sheet because now you as the human would not know how to read that spec sheet, but that is totally untrue about an agent.

And so, an agent would read the spec sheet and you would just say like, "Hey, here's the spec sheet." And then now because it's agent intermediated, you would just say like, "Hey, here's like the questions I need to answer in order to tell you exactly how this will fit your body." And those are the types of things that I think today do not exist in any website interaction, right? But they will start to exist is there will be presentation of information that yes a human could read it, but when an agent intermediated interaction is happening, it will be a very rich interaction. And the human will be like, oh wow, yeah, like I get it. Like not only do I now know how it fits me, but I have learned how to read a spec sheet because this agent just sort of walked me through it.
Right, yeah. That's that's interesting. So it's almost like an agent.md for like fits.
That's for apparel. That's for example for apparel. Yeah.
Yeah, yeah, it's interesting. Yeah, so I I I'm I'm rewriting this stasis library that I that I built a couple years ago and essentially it's for like cohort analysis. And like the idea is like there's just this API and

you you have like data sources and stuff. And it does cohort analysis on the data for forecasting. And like as I'm rebuilding it and I've been working on it, you know, on and off for like last few months, I'm like, this is not the way humans code anymore. No one codes. I don't code. Like why am I doing this? Like I I I So what I did was like there are like capabilities and you describe those in an MD file. And so the way someone's going to interact with this is through Codex or Claude. They're going to natural language. They're never going to need to know the API. The API is totally irrelevant to them. And the reality is if you detach from that idea of you're sort of like limited by the ability of the person to understand and to remember essentially the API, if that's not a constraint anymore, then what you really need to do is just describe like what is possible with this and then tell the agent, okay, if you need to write a connector for Snowflake, this is what the connector should look like. And this is this is the fields that I'm

expecting and this is a You could be much more complex than you what you would present present to a human. And then this is how that pulls the data in is ingested and then flows through to a forecast. Like what I'm providing you is your forecast ability. And so like And then you think about like, well, okay, if I'm going to have a tutorial, the reality is the person's like, okay, I spin up a forecast on this data set or whatever. And like that's what they're going to say. And the agent needs to understand all the steps a human never does. It doesn't matter to them. And so like it's interesting to think about all these ways that really we're designing these interfaces to be consumed by agents, to be navigated by agents, and you can add a lot more complexity and nuance when you do that than if you're expecting a human to interface with it. Because like if you take me to like a jeans website and I've got to understand how to translate the sizing into like pair of jeans that I'm familiar with, I'm just probably not going to complete that transaction. Forget this, I'll go to the store and try them on.

But it literally
Yeah.
That that is what I did, by the way. So so cuz I don't know how to do Why would I know how to do this? And so I asked Muse to fit help me figure it out. I was like, "Can you figure it out? Cuz apparently you're supposed to buy Abercrombie these days. Like Abercrombie is hot, is what I'm told."
Interesting.
I mean, what do I know? Yeah, but
[laughter]
that's what Muse told me. So so I was like, "Okay, what what size should I buy from Abercrombie?" I mean, we'll see if it was correct because it has not shown up yet.
Yeah.
But I'll report back.
[laughter]
But it is not the same size that I buy from Levi's, I will tell you that. Right? So it's like three sizes bigger.
Yeah. Yeah, well and and now from what I've observed on on campus when I was at school, like the style is to wear these like really baggy clothes. And it like to you know, you you look almost like a hobo. It's like that's
[laughter]
That's that's the style. That's the cut. I mean, they should fit. They fit, but they're just

they have a larger cut, I guess. I don't know.
That's a separate issue. My wife says I wear like normal fitting jeans. She's like, "Rishab, the style is to wear loose-fitting jeans." And I'm like, "I'm just not going to do that."
Yeah, yeah.
Like that's not going to happen with me.
[laughter] [gasps]
Anyway.
I talked to you about consumer preferences with an agent. Are these things that the agent can like sort of like intuitively learn? Is something that, you know, you you So you've you've argued that commerce involves this nuanced journey. It's not just matching a catalog to a prompt, right? That's the kind of premise of Fermat. So, if I discover what I want by encountering alternatives, what should the agent optimize for and and how how do we recognize when it's narrowed my choices prematurely and when it got things wrong and it incorrectly disqualified some items? How do we do that? How do we test for that?
Let me just start with a slightly different premise, which is I think that a consumer articulates a set of preferences that are incomplete. So, like whenever I articulate something, I actually don't fully know what I want and then the act of shopping

actually fills in the rest of my preference set. So, I have a certain set of things that that I think I want and then some of those adjust and then some of those are not preconceived that I are choices that I do need to make that get filled in when I am shopping. So, the act of shopping actually fills in the rest of your preference set. There are two parts. What is the agent interaction and then what is the merchant opportunity? The merchant opportunity is for those preferences that I am willing to change, how do I give you confidence in making a different choice? And for those preferences that are open, how do I identify that those are open and how do I give you information to fill in those open preferences? So, I'll I'll use the same example that we were using. So, for for this jeans shopping thing, right? Like what I actually want to make sure of is that it will fit sort of straight to leg and it will fit my waist in a normal way,

not like doof something funny and like sit low or whatever, you know? I'm not trying to be cool. I'm trying to be 40. So, it's like, you know, it should look like that. And so, [laughter] and so, those are the set preferences. And then the non-set preferences are like, okay, what colors are cool now? What What do the pockets look like? All all that sort of stuff, right? Is it Does it have a fade on it? Does it not have a fade? Whatever. And those are I think where the merchant has an opportunity and the agent has an opportunity because I think what the agent will learn is hey, what this person is trying to do is they have a shopping pattern where they buy the same thing over and over again. And all they're trying to do is be comfortable in that same thing that they're buying over and over again, but he has particularly asked me he's trying to do something that like is trendy. So the agent will learn that and say like hey, what is trendy right now is dark wash. The merchant opportunity in that context

is okay, let me make sure that I can serve the fact to to the agent very simply like what is on trend, what is not on trend, color availability, sizing information. So that way that can be fit very quickly and then I can influence the decision to happen because I have filled in the information that the agent is looking for, which is why I sort of gave this idea of like a very large amount of information living on the website in a place where the agent can navigate to it. Right? So if there is a way for the website to communicate like hey, dark wash is what is on trend right now, when I am watching the agent browse that and say like dark wash like hot on trend, just restocked, whatever, then it becomes apparent to me that it is fulfilling those criteria that the agent is helping me set. So like I I think that's the way that you one set what are the things that you prefer, make sure that the agent understands

those preferences and then if it starts to do something totally crazy like hey, here's like leopard print because you said you you don't mind any color, I think that the nice thing about agents is like the corrective mechanism is so fast. It's like hey, you got this totally wrong. Like I am just I'm looking for things that are going to make me look a little bit more modern but not stray too far. It self-corrects, right? And so that's like roughly how I see the interaction happening and that's why I think that the data availability on the website is so important.
Well, and the availability elsewhere, too, right? So, I mean, if they have access to my email I mean, some of these preferences can be learned from other sources of data, right? So, you got access to my email and and there may be some like latent information there that informs that I don't like leopard print, right? Or you could infer incorrectly that like, "Oh, I've got, you know, PETA emails, so I love animals and therefore I might like leopard print." I mean, but again, to to your point, like I think this this workflow that you're

describing, it does sound like you believe that this agentic commerce future is I'm watching the agent operate on my behalf and there's always a human in the loop and I Okay, I nudged it away from leopard print, it learns that, therefore that's corrected going forward, but like we're not in the land of I want to have a delivery on my front porch in a week that's got a whole new wardrobe. Here's my credit card, make it happen, agent. Like so you so you're kind of in this like human in a loop land and this is just kind of taking away all the clicking and all the the mundane stuff. And also, there's the benefit of the website now communicates directly with the agent in ways that are much more meaningful and carry a lot more information that you just never could do if there was a person navigating.
The reason I think that this is the sort of magical human in the loop moment is the agent workflows that are the most popular, so like coding, are very human in the loop. And I think that that is the thing that was missing in the venture speak

version, which is like, "Oh, things will just show up automatically." It's like there's no product or engineering person who that is any good that lets the agent run wild without getting the output and then saying some the next thing, saying the next thing, saying the next thing, saying the next thing. And that was the thing that needed to be replicated for commerce. And we are In my opinion, I think the Muse instantiation is the one that is the most It is It is the clearest first step in the right direction.
Yeah. The VC-pilled takes are just seem so sloppy and illogical. Like I mean I just feel like people aren't thinking clearly or I mean I get I mean I get it. Look, but the job of a VC is to be excited about every new thing because you'll be right one out of a hundred times. But if your fund is big enough, you only need to be right one out of a
a 10x on the fund.
Yeah, 10x on the [laughter] fund, yeah. But it's just some of this is just so

sloppy and illogical and it's like it just feels like people's people have been deranged. It's like a mania or something. Like I yesterday I was responding to Paul Graham. I said this is like religious zeal that you're talking. He just like takes for granted this idea that this is how everyone will shop going forward. And so Amazon just killed itself and will be disrupted by a startup that creates commerce for agents. Like is it come on? That's not that's deranged. That's just irrational thinking.
Yeah, I mean I think that the most interesting thing is I think that there will be someone who says here is how I will kill Amazon with agents and now they will get funded with a very large sum of money.
Yeah.
Because that tweet was sent out. Right. Like like like I actually part of me is not even sure that they believe it.
Mhm.
No one thinks
[laughter]
that like Amazon's magic thing is just that search bar. It's like what everybody does it because of the delivery infrastructure and like there's I mean there's just so

much other stuff that is going into what makes Amazon Amazon at this point. And just the sort of like search and navigation is not what makes Amazon Amazon. Like that that is not it. And so but I do think that putting it in the universe
Right.
means something interesting will happen. Because now someone is going to fund someone at like a hundred million dollars seed round and it is for sure not going to kill Amazon, but something interesting will happen because of it. And I I I actually like my less cynical view is like I think it is intentional.
Yeah.
I I I think that that's like a huge part of the game.
So so I think it sounds like what you're saying is look, we'll just shake the tree. Like I don't necessarily believe what I'm saying, but I believe it's disruptive in the sense that it'll make funding happen. And funding when funding happens, things get built. And when things get built, markets get disrupted. And when markets get disrupted, people make money. And that's Is that kind of Is that the Is that the general idea?

Precisely. And you have to say something big and provocative enough
Right.
that it can draw capital.
Right.
Uh other otherwise, it's like so hard to draw the capital, right? Like my best equivalent of this is like I mean, do I think Elon is sufficiently execution-oriented that we will actually produce a colony on Mars? Like he is incredible. So it's possible that we actually get there. Yeah. But I think that even if we did not get there, the total amount of good that will have happened because of that unbelievably ambitious flag is so big that it will have been worth it.
I mean, this could be a whole podcast episode. So I So as you know, I live in Austin. Austin is the first market for the Cybertruck. You know how transformational that is? So first of all, I I very rarely drive anyway. I usually take now robotaxis or cybercabs. Like but but when I do drive, I don't drive. My car drives me. I have a Tesla with FSD. The first time you do that, your life

changes. It's one of the only products I've ever experienced where I truly thought that forever my life has been changed. Like like I can't think of anything else that I've ever interacted with that gave me such a profound sense of change. And so when I do drive, it's FSD. I mean, like from the driveway to the driveway. It's I'm not I mean, unless there's you know, some navigation thing where it's it's taking me like a path that I know it's under construction or something. But like when you walk around Austin, you see cybercabs, robotaxis, and Waymos. It's got to be one out of every three cars. And then people tell me, "There's no impact on the economy of AI. What are you talking about? This LLM's don't change I'm like are you kidding me? You don't know what you're talking about. I don't drive my car anymore. When I go outside, a third of the cars I see on the road are autonomously being driven. Like this is an awesome phenomenon, too. But you go to SF, you see a lot of Waymos.
Yeah.
If you have not experienced it, I can maybe understand why you're just sort of like in the dark

about the impact. But yes, no, AI in this broad very broad-based use of the word has dramatically changed dramatically altered society. It's It's just inaccurate to say otherwise. And then, you know, you look at the people there's there's individuals driving that change. And like to your point about Muse, Zuck. I mean, I just read this piece today about he kind of drove the release of Muse. And there are individuals in the society that are catalyzing that change, which is just like fascinating to think about.
Let's get real. Half the time, you're not sure if your campaigns [music] work. You just kind of hope they do. Attribution numbers look good, but what if those conversions would have happened anyway? [music] Incremental helps you find out for real. No tracking people, no guessing games. [music] Just clear answers on which campaigns actually move the needle. If you want to stop guessing and start [music] knowing, go to Incremental. Because good marketing isn't about getting clicks, it's about getting results. Check them out at incremental.com. That's i n c r m n t a

l dot com, or click the link in the show description.
All right, so we've got the ad, the landing page, and the offer. These these things can all adapt together as essentially this cohesive unit, right? And that's the essence of Vermont, right? How does that change in the era of the agent? And should a merchant still kind of think about creative and conversion as as two separate ideas? Or is this all just this one kind of unified concept at this point?
Okay, I have always felt that it is all one unified concept. And that is sort of like you said, that is like the sort of premise of Vermont, is that when someone sees a piece of creative that brings them to the site, The fact that the site does not care about what creative sent someone there or what is the intent based on that creative or what is that's just like a ridiculous reality. And and the reason that it's ridiculous is is because you can just see by definition the conversion rate is

different when you come from a meta DPA ad versus a Google PLA ad. It's nominally the same ad format, right? It's like a catalog ad. It's like in both cases you push your catalog out and you see the item, you click on it and you get taken to the PDP. But obviously they are totally different forms of intent because one is search intent and the other one is discovery when in between watching cat videos on Instagram. So the obviousness of that statement is is like glaring, right? And so to me they have always been one thing. Like the content, the ad, the channel that drives that shapes the intent of the person, that should be carried through the shopping journey. You need to think about the whole journey as one and you need to watch the whole journey as one. I think the beautiful thing about it being agent intermediated and and meta here has this like unbelievable opportunity to like intermingle muse and meta ads and we will see how that continues to get

intermingled. But the beautiful thing about agents is that the opportunity to be more adaptive on the website surface because you're interacting with an agent who like I said can like navigate the page very quickly, consume a lot of information very quickly, present it in a way in which the consumer wants to interact with it very quickly means that you can actually drive greater adaptation based on where the person is coming from if it's an agent intermediated interaction. I'll give you like a toy example. Like a toy example is let's just say I see a piece of content on meta and it's like someone wearing an outfit and I'm like, "Oh, I'm going to click on that." And somehow meta figures out how to make the next part of my flow agent intermediated. So instead of just an in-app browser, it is an agent intermediated in-app browser. Now what happens is all of a sudden it's like not only am I doing just like

curious discovery, but the agent in this world could say like, "Hey, what made you curious about this?" This is again a toy example. How how this shows up we will figure out, but let's just say it could figure it could ask you nominally, "What made you interested in clicking on this? Is there something in particular you're looking for?" And it's like, "Yeah, you know, I just always wear the same t-shirt. This one looked interesting to me because it seems like it'll cover a little bit of my gut. What like, you know, will it actually do that?" And now what needs to happen is the merchant needs to again have the information on the sites the agent can answer that type of question. And then you will sort of navigate the site with the agent, and it becomes like a fun agent intermediated shopping experience. I think that that is why a company like I or at least my view is a company like ours has a very interesting opportunity is because agent intermediated interactions means

you can think of these journeys both from a more personalized perspective and a more continuous perspective. And you don't need to have the funnel be split up into different parts. And I again I I just don't believe the funnel was ever split up into different parts, and I think that that's what caused friction when you had different channels all sending people to the same place.
Yeah, that's that's fascinating. That's that's a very profound idea, I think. So I mean I think so let's let's assume that where we're headed is ads and agents. And the the agentic workflow really is fully human mediated cuz I I do feel like it will be. So I'm I'm in my agent. Agent's helping me with discovery. A lot of the discovery is ads, sponsored products. And so that's creative, right? And so the agent can pull a lot of information out of the creative that is imperceptible to me. And so the way that it navigates that page is going to be informed by that data that those those themes or whatever that it pulled out of that

that creative that I that are they're imperceptible to me or they're or they're, you know, incomprehensible to me, but the agent knows something about that. The fact that I clicked on that on that creative, the agent learns something about what I want when we hit the landing page. And so then I let the agent the agent takes back over. I clicked. I chose which I wanted to which product I wanted to like sort of investigate. But now the agent takes the wheel and is navigating the page. And so that page needs to adapt to the knowledge that it knows the agent has. That's very fascinating, that idea.
Yeah, yeah. And this is, by the way, just the natural quote unquote natural agentic extension of how Meta ads work today. So it's like So so let's just pretend I I clicked on that ad first. And then a thousand people clicked on that ad. Meta has learned for all thousand of those people, what did they do next with their agent? What were they looking for? What were they trying So now when you show up and you're the thousand and first person to show up alongside your

agent, Meta's like, "Hey, you know, people really like this is what they are thinking when they come here. Isn't that what you're thinking, Eric?" And you're like, and then, you know, all the people who don't know how ads work is like, "Oh my god, they're listening to me."
Yeah.
But you know, you and I are just like, "Okay, this is just a data thing. They have the data. They understand exactly what I'm looking for. And now they're going to do that thing all the way through the purchase journey. And it's it is going to be amazing. It's going to be amazing. Yeah."
And so that that is kind of like that further supports this idea that like the whole landing page gets rewired for agentic use. But not not just because, you know, just agents will respond to the the form controls differently or whatever. Just they have more information. They're coming in with more information.
Yeah, exactly. Exactly.
And more intent because you gave it to them because you clicked the ad. It was an and it was an ad and will be an ad and you clicked it. And that's what that's the one you clicked, not the other one.
Precisely.
And therefore it knows something about you that you wouldn't know that you that

that would would not inform your purchase decision. Yeah, that's that's really fascinating.
All right, you you you you you predicted on LinkedIn that there will be this new cohort of brands that grows through AI chat traffic. So just as you know, Facebook ads gave birth to this whole I I I have claimed that it gave birth to DC category. It would DC would not exist before
Yeah, to Shopify.
To Shopify, yeah. Made Shopify a a company. There will be this cohort of brands that emerges as a result of, you know, AI chat traffic and agentic discovery. So talk to me about that.
Yeah. I was definitely wrong on the chat GPT part. And so, you know, if I were to like Keith Rabois my previous prediction, it would basically be but I'm going to be right about the Muse part of that prediction. You know, it's like I think like what what I what I and this I this part I mean genuinely um and I'm I'm just joking about Keith and Keith is like a phenomenal investor, right? So I'm I'm I'm being a little tongue-in-cheek about it, but the thing

that I thought is true is that anytime there's a new thing that has a lot of attention that's a new format is going to lead to new ways of doing discovery in some way, shape or form. And it just supports a new type of ad advertiser in that context because it's so hard for existing advertisers to catch up at the pace that you need to catch up to in order to take advantage of that new type of ad inventory. And so, my my actual view here is like we, you know, I I don't know if it'll be chat GPT, but now that I have seen Muse, I don't know if it'll be chat GPT. And so, maybe it will just continue to be meta who like wins this as well as a consequence, but the brand underneath will be the one who can support this interaction because I'll I'll tell you a funny story about my actual jeans shopping experience. My Muse agent thought that the size for dark wash was not available,

and so they said like, "Hey, you should just buy the medium wash instead." And it was because of the way the merchant was sending the information to Meta and the rate at which they were doing refreshes, basically. And so, like what it tells me is whoever are the brands who invest the most quickly, and those tend to be younger brands in understanding what is necessary for taking advantage of this way that people are going to shop, that is the sort of a new wave of brands that will emerge. I also don't know if Shopify will be the best supporting infrastructure for that. And so, there's potentially another opportunity, which is like the Shopify format of purchasing, because Shopify seems like it is continuing to invest in things like sending the feed and UBT and actually not allowing agents doing the check You You know what I mean? It's like a kind of funky stance that they have. And so, I think both there's a new wave of

brands and there's potentially an opportunity to build the agent-native version of Shopify that doesn't need to be precious about about checkout and therefore can let the agent do it.
What concretely like what do you think that looks like? What So, like what I guess what what's your advice? So, let's say that you've got some plucky D2C founder in the audience, which I mean probably do. What and they they're like, "I want to launch this new I want to be on the at the forefront of this. I want to be one of these companies that emerges to take advantage of this opportunity." What what should they do? Is it Is that just a function of like what we've already been talking about, which is like this Agents MD idea for here's how you interpret everything about my product and like and adapt it to your knowledge of the user. Is that it? Or is Are there other things? Just like what what types of products we can make? Is it Is Are there certain categories that were better suited for this?
Yeah. I guess I'll share it since if someone is going to spend this amount of time listening to us, then here's the gem that you got. The actual answer to that question is

you would have to approach it the same way a growth hacker would approach any new system. So so the way the way that a simple way you would think about this is you would construct a page and then you would tell Muse to go to that page or tell Instinct to go to that page and help you shop. And then you would just keep iterating on that, keep iterating on that until you figure out the the thing. So you would take this approach of like, okay, then you would do multiple pages, then you would do a full website. And you would sort of growth hack the Muse interaction from the end state out and then you would build the brand around that and then that's how you would sort of go to market. Right? And then you would you would just keep figuring that out and you would know the system better than anyone else because that's how you will have done it. So is that it looks very similar to every other era like the sort of thing to do is actually exactly the thing that every growth hacker did in every other era of every new thing, right? So it's like Google search SEO, how did people do it? They just sort of like

you know, you iteratively figure it out and then you just ride that wave. And that is what the plucky D2C founder should do.
Got it. It's just emulate the end user who will be navigating these pages with their agent. Like similar to what I was talking about with the Python library. Like I'm not building this for the person that I'm building this for the agent. So how does my agent use it? And if they use it correctly then I've done a good job and if they don't then I've I've done a poor job.
Yeah.
Do you think there's brands that are better able to capitalize on that? I'm I'm thinking about like, you know, you'd think about brands where there's just more information maybe and so like the agent knows more going in or can better interpret the distribution. Is it is that like would you go out like is this make a maybe there's like a red like the red ocean is the right strategy? I'm trying to think through like the product categories that that succeed here.
[laughter]
You know the irony here is I would go to the thing that you started with around it being ironic around around travel and I would think

about what are the products that people want to buy when they travel or that they should buy when they want to travel because they're using these agents to book travel as the first thing. And there is just no way that I believe that again there's not a way to growth hack the fact that it's like everybody is using Meows to travel. How do I influence Meows to ask to get you to think about what you need to buy when you go on that trip? And how do I insert myself into that category? So sunscreen or you know you know like the things like this are like things that immediately jump out to me, but what I would do is I would think about what is the category that people are already using this for a lot. And then what are the accessories to that category? That is my opinion of what is going to be the thing that happens first. It's just like what are people chatting about anyways with these agents?
Yeah. Yeah, I I think you're right. So before we started recording you said that you're excited because you think this just changes commerce where you had maybe thought for the last few years that commerce was fairly static. And I

think you're right. I mean I think there just are and and like in this in I think these changes will be maybe like in in ways that are difficult to predict now cuz I mean the agent agentic anything does really change fundamentally like the scope of knowledge about a task. And and so you do kind of think about like there's just like these strictures that just don't apply when it's a machine doing it. And so it's really interesting to think through like these these just fundamental transformations and how might the process change. This was fascinating, Rishab. Thank you so much. It's always a pleasure to have you on and to learn from you. How can people learn more about Fermat?
Yeah, so you can find me on LinkedIn or Twitter. I'm just Rishab Jain or just find us at our website fermatcommerce.com. And yeah, Eric, these are always so fun. So thank you again for letting me come back on.
Of course. And for what it's worth I've I've always thought what Fermat does is

like very straightforward. I I I don't understand why people don't understand it. It's it's [laughter] It seems like very straightforward. I don't I don't get the confusion. But, Rishabh, have a great weekend.
You too.
[music]