What is agentic commerce? Stripe’s CRO of AI explains.

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

RANA EL KALIOUBY: I want to try an experiment. If you know me, you know I LOVE chocolate. But I am also very aware of the food I put into my body. So this is my everyday problem: How can I satisfy my chocolate craving but eat something healthy?

I want to try an experiment: How can AI help?

So I’m going into ChatGPT, starting a new conversation, and entering this prompt:

I’m looking for a chocolate bar that has no added sugar, gluten-free, and is packed with antioxidants and maybe some protein. Can you send some options?

OK, and here are the options. So the top three choices: I know Lilly’s, but I haven’t heard of Askinosie and Pascha. They look yummy, and I can buy them right now through ChatGPT.

This kind of AI-powered shopping is called agentic commerce. And my guest today argues that this is the future. Maia Josebachvili is chief revenue officer of AI at Stripe. On this episode, she makes the case for AI chatbot shopping. We talk about the risk of commodification and how AI can bring new customers to businesses of every size.

I’m Rana El Kaliouby, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

EL KALIOUBY: Hi, Maia. Welcome to Pioneers of AI. I am so excited for our conversation.

MAIA JOSEBACHVILI: Hey, Rana. Thank you so much for having me.

Copy LinkHow risk-taking shaped Maia’s path to building in AI

EL KALIOUBY: We are going to talk about AI for most of the conversation, but before we dig in, I want to talk about you being a very accomplished skydiver. I don’t think I will ever skydive, so that sounds too terrifying. What was that first jump like?

JOSEBACHVILI: I don’t know that many people grow up saying, “I want to be a skydiver.” I certainly didn’t think that. Then I did my first tandem jump, and I landed and said, “I want to be a skydiver.” I just knew.

There’s this feeling you get when you jump out and you have, call it, 60 seconds of free fall, where the whole world goes away and all you can do is think about what you’re doing in that one moment. There’s absolutely nothing like it.

EL KALIOUBY: Do you still skydive?

JOSEBACHVILI: No, that was Maia in her 20s. Definitely not now.

EL KALIOUBY: That’s great. You then went on to become a founder, and I feel like that’s a different type of adrenaline rush. But you still have to take a leap of faith. So what has your skydiving experience taught you about being a founder?

JOSEBACHVILI: My skydiving instructor, Oli, who was probably 40 at the time and living in Vermont — and when you’re in college, 40 is very, very wise — used to always say, when I’d tell him I was scared, “Well, that’s good. It’s a sign of intelligence. I’d be very concerned if you weren’t. But what could possibly go wrong?”

There’s something very funny about asking what could possibly go wrong in a skydive. But what was helpful was that you’d actually walk through all the things that could go wrong and figure out how you would handle them. I think taking a leap of faith as a founder or a leader, or experimenting with new forms of AI, you can go through the same process, which is really to walk through what could go wrong.

At the end of the day, it’s not that bad. So go ahead and jump.

Copy LinkWhat it means to lead revenue for AI at Stripe

EL KALIOUBY: Go ahead and jump. I love that. So you’re now at Stripe. Your current role is chief revenue officer of AI. That is a very new title. How did that come about? What does it mean? What are your responsibilities? We’re always very curious about the new jobs that AI is creating, and I believe this is one of them.

JOSEBACHVILI: Yeah, absolutely. At Stripe, we’re creating the economic infrastructure for AI, and that means we power the world’s fastest-growing and most ambitious AI companies on their growth journeys. In fact, if you take the Forbes AI 50 and look at anyone who takes online payments, they do so with Stripe.

So part of my job is partnering with all of those companies — OpenAI, Anthropic, Cursor, Replit — and working with them on their growth journeys. We also are powering the new agentic commerce wave that you’re seeing across the board. OpenAI released Instant Checkout. Microsoft Copilot has a solution. Google has a solution. We’re powering some of the world’s largest retailers on those platforms.

Copy LinkHow agentic commerce is changing the shopping journey

EL KALIOUBY: When you say agentic commerce, what do you actually mean by that?

JOSEBACHVILI: We’re referring to AI-enabled commerce in two forms. The first is autonomous agents completing purchases on behalf of consumers, and the second is humans making purchases within AI applications such as chat interfaces. The industry generally uses the term agentic commerce to cover both, and as we partner with our customers, they do as well. So we think about agentic commerce as the whole suite.

EL KALIOUBY: Of these two scenarios, which one is more common today? Can you give us an example?

JOSEBACHVILI: Absolutely. We’re seeing a lot more pull right now on the second one, which is humans buying within chat interfaces. We’re really focused on helping everyone prepare for the agent era. So, as an example, buying a coffee table in the before times:

You would go to 10 different websites of the brands you knew. You’d go into the search and the filter and say, “I want a coffee table.” And you were very limited in the options. You could choose wood or metal, round or rectangular, and then you’d flip through and see what you could find.

EL KALIOUBY: Yep. And then you’d get your first 20 options, and then you have to click onto the next page, and so on.

JOSEBACHVILI: Exactly. Now you can take a picture of your living room, upload it to your favorite LLM surface, and say, “I’m looking for a warm, modern-vibe coffee table that fits nicely here.” What you’ll get back, if you prompt it, is a few different images of different types of coffee tables already situated in your living room that fit your aesthetic.

EL KALIOUBY: And then you click buy.

JOSEBACHVILI: Well, that’s where we’re all heading, right? A few retailers are live today, but we’re working on getting more merchants live. The vision is that Stripe will be the fastest on-ramp for brands to participate in agentic commerce.

But you can see how this is just a much more powerful shopping experience and better for the end consumer.

Copy LinkWhy AI shopping could be bigger than the mobile shift

EL KALIOUBY: You have actually said that AI is a much bigger shift for commerce than mobile was. Why do you think so?

JOSEBACHVILI: In mobile, you still had the purchase happening in the brand environment, and consumers still had to know where they were going, right?

EL KALIOUBY: Right.

JOSEBACHVILI: What agentic commerce really brings is the ability to bring commerce to where intent is happening. People are already using these LLMs to search, and now we can bring the different options right where the consumer is. I’m really excited about it for a lot of reasons, but it also democratizes the ability for brands to get in front of consumers. We have Etsy live in ChatGPT and Microsoft Copilot.

EL KALIOUBY: That is so cool.

JOSEBACHVILI: It’s so cool. You have these local artisans who may not always be able to get in front of the consumer.

But now, if someone is searching for a Pokémon Halloween costume, that might show up and give that local craftsperson access to consumers they didn’t have before.

Copy LinkHow brands can stay visible and in control in AI search

EL KALIOUBY: Very cool. On that topic, because AI is changing the way we discover information and find products, search engine optimization is really being reimagined, right? SEO was basically how Google surfaced products to the top. How is AI changing all that when you’re shopping inside these models?

JOSEBACHVILI: This is one of the big conversations happening right now, and brands are really thinking about how they show up where the consumer intent is. The product catalog really matters, and the nature of the product catalog really matters. How do you get this information to be really agent-legible?

EL KALIOUBY: That is a new term. What do you mean by that?

JOSEBACHVILI: You’ve got agents now trying to understand and make sense of all the different SKUs on your website, all the different SKUs in your product catalog. How do you make that really digestible for them so that when someone is searching for the warm coffee table, that shows up?

EL KALIOUBY: Right. For every product item, what is the information you have about the product, and can you make the agent access that information? That’s pretty cool.

JOSEBACHVILI: Yeah, it’s a real shift in the infrastructure that these retailers need in order to participate. And we obsess over what it takes for businesses to thrive in the agent era without giving up control.

That control is an important piece for them. We’re talking about all the benefits, but the concern we hear from a lot of merchants is that they still want to have a relationship with the customer. So something that was really important when we were designing the agentic commerce suite was how do we make sure that the merchants are still the merchant of record and they can still have the customer relationship? That’s been one of the big priorities as we’ve been building: How do we build the most merchant-friendly infrastructure so that the world’s leading brands can participate in agentic commerce?

EL KALIOUBY: We’re going to take a break. When we come back, the role of ads in our agentic future. Stay with us.

Copy LinkWhat ads and rapid AI growth signal about the next wave

EL KALIOUBY: OpenAI announced that they are rolling out advertisements on ChatGPT, and they’ve actually said that these ads won’t influence what products you see in answer to queries. Rather, they will just allow advertisers to advertise at the bottom of the ChatGPT interface. What do you make of all of this?

JOSEBACHVILI: I think it’s a natural extension, and it’s to be expected.

EL KALIOUBY: Do you think people should be allowed to pay more to show up more in answers to queries? I personally can see a world where that is fair game, but I also worry because if I’m asking ChatGPT for advice on supplements to take, I want the answer to be impartial. So I can see the pros and cons. What about you?

JOSEBACHVILI: I’m thinking about it because I think it’s a really good question. My personal opinion is that if it is going to show up, you need to know it was influenced by ads, and I think that might be a way to get around it. I also think it is really early days, and things are going to look very different in a year or two. A lot of companies are going to experiment and see how it all plays out.

EL KALIOUBY: I’m sure that’s going to be true because things are moving so fast. We had Suresh Kumar, the CTO of Walmart, on the show, and he talked about how they’re integrated within ChatGPT. So now you can be shopping, and then Walmart products could show up. Does Stripe power all of that?

JOSEBACHVILI: We are under the hood on those cases today, in some way, shape, or form. Different integration shapes, but we’re under the hood of the Instant Checkout experiences that are live today.

EL KALIOUBY: So you power a lot of the top AI companies out there. What are some of the trends that you’re seeing?

JOSEBACHVILI: Two really major ones, and they both have to do with growth. The first is that the pace these companies are growing at is just unprecedented. We’ve indexed the top 100 AI companies on Stripe, and the ones that hit $30 million in ARR did it in about 18 months. By comparison, when we look at the SaaS companies from 2018 that got there, it took them five years. And even within that, you have Lovable getting to $100 million in eight months, Manus, the general-purpose agent, going from zero to $100 million in eight months, and Higgsfield — if they’re not on your radar — getting to $200 million in nine months.

EL KALIOUBY: What do they do?

JOSEBACHVILI: Video generation. So that’s all at the high end. But even the companies we see hitting $1 million in ARR, AI startups are scaling four times faster than their peers. AI companies are global from Day 1. Another stat: The median AI startup, when we look at the top 100, is operating in 79 countries by its second year.

EL KALIOUBY: Wow.

JOSEBACHVILI: That’s about twice the internationalization of equally promising earlier SaaS companies.

EL KALIOUBY: It’s incredible. So they’re getting to revenue much faster than traditional SaaS companies. They’re global from Day 1. And I am really curious about different monetization schemes. We’re tracking that very closely as an early-stage fund as well, and I find that very interesting.

JOSEBACHVILI: It really seems like right now this hybrid subscription-plus-overage and usage-based billing is starting to be the norm. And I’m sure when we talk again in six months, we’ll hear new things, but the rate of change is really impressive.

EL KALIOUBY: I’m thinking about some of the AI companies. How can you tell if a company is going to explode in terms of revenue and growth?

JOSEBACHVILI: The journey looks different for different companies.

But there are certain signals you can see when they’re about to pop. It might just happen at different times in their trajectory.

Copy LinkHow retailers are adapting to protect brand identity and customer ownership

EL KALIOUBY: That’s interesting. So back to your providing this infrastructure: You mentioned how online retailers really want to still own the relationship with their customer. How are you enabling that, and have you gotten any pushback from some retailers? It sounds like you’re working a lot with AI companies that are at the forefront of innovation and are really reimagining what commerce looks like. But what about traditional online retailers? What’s the feedback been so far?

JOSEBACHVILI: The shift in the last seven months is pretty staggering, too. We have a customer advisory board with 20 of our largest merchants, many of them retailers. Last summer, when we were talking to them about agentic commerce, there was a lot of wait-and-see thinking. “I’m not sure.”

Fast forward, and 19 out of 20 of them had already started implementing an agentic strategy. Everyone is somewhere on the spectrum from experimentation to “I’m all in,” but everyone is jumping in.

EL KALIOUBY: So much of retail at the moment is driven by brands really investing a lot of time, effort, and money into that brand identity, right? Whether it’s a logo, the colors, or the sounds.

What does that look like in an agentic commerce world, and how do we prevent this world where everything gets commoditized? I’m buying my pair of snow boots, and it doesn’t really matter what brand it is.

JOSEBACHVILI: It’s really top of mind for us. We obsess over what it takes for businesses to thrive in the agent era without giving up control, and that means a couple things. It means they get to stay the merchant of record and control the whole post-purchase experience.

And their brand has to show up in that interface.

When we designed ACP, the agentic commerce protocol that we co-authored with OpenAI, and our agentic commerce suite, a core tenet of both was that the brand had to show up in that discovery. This is something we heard loud and clear from our merchants. Again, we’re building the most merchant-friendly infrastructure so that businesses can thrive in the agent era.

Having their brand show up and having control of the customer relationship were two of the most important tenets.

Copy LinkHow Stripe is fighting fraud in the era of AI agents

EL KALIOUBY: So let’s talk about some of the challenges, or maybe the things to look out for. I want to start with fraud in this agentic commerce world. How do we prevent fraud, and what are you doing about that?

JOSEBACHVILI: We spent decades preventing bots from buying on our websites. And now we’re saying, actually, we want the good bots. Please come and buy all the things. One of the things we’ve launched is the shared payment token. When someone is putting in their information within a chat experience, we have what we call the shared payment token, which basically wraps that credential and gives it to the merchant.

EL KALIOUBY: OK.

JOSEBACHVILI: The merchant can retain control. They are the merchant of record, and they have a relationship with the customer.

In this, we also package a risk score.

EL KALIOUBY: How risky is that buyer?

JOSEBACHVILI: Exactly.

EL KALIOUBY: But how would you come up with this risk score? What is it based on?

JOSEBACHVILI: How we built our fraud and risk models has completely changed. Historically, Stripe used specialized models. We had authorization optimization, one for fraud, one for disputes.

It worked well. But recent breakthroughs in AI have shown us that a generalized foundation model can outperform the narrow ones.

EL KALIOUBY: Interesting.

JOSEBACHVILI: We created our payments foundation model, and it’s trained on tens of billions of transactions across our network. Much like an LLM understands language, our payments foundation model understands payments.

EL KALIOUBY: That is fascinating. And then it’s able to basically flag whether this is a payment that looks OK or is fraudulent.

JOSEBACHVILI: Yeah. What we’ve seen is that as we’ve used this new payments foundation model, we’ve reduced card testing for businesses on Stripe by 80 percent over the last two years. Just by using this model, we were able to detect that class of attacks 59 percent of the time before, and 97 percent almost immediately after.

EL KALIOUBY: What are some examples of fraudulent behaviors that AI companies are experiencing?

JOSEBACHVILI: In the same way that we’re seeing new monetization models and different growth, we’re also seeing really new fraud patterns emerge with these AI companies. One example is free-trial abuse. What we see is bad actors chaining accounts together, triggering AI workflow loads, and then leaving startups with massive compute bills. In traditional SaaS, a stolen trial didn’t matter that much because there wasn’t much marginal cost. But with AI, when you have real compute costs, these can be existential for those AI startups. We’ve seen some AI companies actually be forced to turn off free trials altogether because of this. We’ve since invested in Stripe Radar, our fraud detection tool, so that we can detect and block this behavior and help AI companies let legitimate users go through while stopping the abusive ones.

EL KALIOUBY: More with Maia after a short break. Stay with us.

Copy LinkWhy AI adoption at work depends on culture not just tools

EL KALIOUBY: So I want to talk next about how AI is changing the way we do work. I would love for you to take us inside Stripe. One of my investment theses is that AI is shifting value creation, specifically through this idea of agentic AI and AI co-workers that get work done on your behalf. So it’s not just creating a tool that helps you be more efficient, it’s also taking on end-to-end tasks on your behalf. Does Stripe deploy some of these workflows, and how has that changed the way people work, especially the culture of the organization?

JOSEBACHVILI: I love that you mentioned culture, because one of the most important things we’ve learned is that AI adoption is as much a culture shift as it is a technology shift. What we’re seeing is that the companies that benefit most are the ones that build a habit of experimentation.

So at Stripe, we focus on making it easy for every team to safely interact with a range of models, and for those models to safely interact with our internal tools and systems.

I’ll give you a good example here. Local payment methods are really important for businesses for conversion, reach, and trust. What I mean by that is customers in Germany prefer to pay with Giropay. In Brazil, Pix dominates. In India, you have to have UPI.

So being able to offer these really matters. Historically, each new payment integration required a lot of manual engineering effort. But if you take a step back, most integrations really follow a simple pattern, which is that you have two APIs and you need to make sure they understand each other. So we fed an LLM our integration documents alongside the payment method’s documentation, and we prompted it to generate the integration code. After a few iterations, it worked end to end. We now have a new pan-European payment method whose integration was almost entirely written by an LLM. Before these tools, that would have taken us two months. This one took us two weeks, and I’m pretty sure we’re going to be able to bring it down to one to two days as those patterns standardize.

EL KALIOUBY: That’s incredible from a productivity perspective. How do you bring your organization along to allow people to build these things?

JOSEBACHVILI: I think that goes back to what you were just saying about culture. We really try to build a culture of experimentation and encouragement, celebrate all of this, and encourage people to do it. We have an internal agent builder where anyone can go in and build an agent. I was playing around this weekend with building my own agent for a couple of things. Just making it really easy for any Striper to experiment safely is a really important component of this.

Copy LinkWhat the future of AI shopping could look like beyond chat

EL KALIOUBY: All right, a few more questions for you. What’s one use case in AI for shopping that we haven’t seen yet that you would like to see?

JOSEBACHVILI: I am really curious about how local services will end up showing up. What I mean by that is, our puppy is sick this week — he’ll be OK, thank you — but we had to call our vet, who couldn’t see him, so we had to call a bunch of different vets. It’s not shopping in the traditional sense,

EL KALIOUBY: But still, you’re shopping for a service, right?

JOSEBACHVILI: Yeah. So how do you make that sort of discovery and booking really easy? We work with users like Jobber, which is a platform for local service providers, whether you need a plumber or something like that. It sounds so much easier to be able to go to a chat service and say, “I’m looking for a painter,” and be able to do it all within that. So I think that’s not coming tomorrow, but it could be a really interesting use case.

EL KALIOUBY: I would personally love the use case where you’re brainstorming gifts for people, and hopefully you want it to be creative. Then it surfaces who does that out there, maybe through the Etsy integration. I don’t know, maybe it exists already.

JOSEBACHVILI: I love that too. Honestly, I did all my holiday shopping through chat surfaces — not the full transaction, but the “here’s the person and here’s what they like, what do you think would be good gifts for them?” part. So as soon as we have more merchants live, I think that’s going to be a really seamless experience.

EL KALIOUBY: A lot of the way we are interacting with AI today is through a chat interface, right? There’s a little bit of voice and conversation, but it’s mostly chat. One of my investment theses is that we’re going to start to see more AI-native devices and interfaces, whether it’s glasses or maybe it’s an embedded device or a wearable or whatnot.

What does agentic commerce look like in this world, and do we even know?

JOSEBACHVILI: It is really early days. I think it’s very clear that there will be other modalities. We feel very confident about that. Exactly how those play out is still to be seen. There are a couple of startups doing different things. We see some browser buying happening, but right now the effort I’m seeing across both the AI platforms and the merchants is really on the chat-based experience.

And I think there’s still a lot to figure out there. My personal hypothesis is that we’re going to see a lot more development there in the near term. Then, as that starts to get more built out, we’ll start to see the other modalities expand.

EL KALIOUBY: So, Maia, do you still shop in person? And do you think AI is going to change the in-person shopping experience at all?

JOSEBACHVILI: I have to admit, I’m one of those people who probably stopped shopping in person as soon as the delivery experience got really good with online shopping. But my personal hypothesis is, look, some people will always love to shop in person and need to feel and touch. But we’re seeing a lot more AI startups come out where you can see an image of yourself and see what the clothes look like on you. Gemini is even starting to get really good at texture — how does fleece sit on you versus silk, which is very cool.

So I do think we’re going to see, as that technology gets better, more people opting to shop virtually in a more immersive format than what we were seeing before.

EL KALIOUBY: Very cool. I can’t wait for that. OK, final question: What’s one way AI has changed your everyday life?

JOSEBACHVILI: I have a whole new commute ritual.

EL KALIOUBY: Ooh.

JOSEBACHVILI: I alternate between Claude and Gemini and ChatGPT to compare, but my drive home is me going back and forth, batting around different ideas and getting perspectives. It’s really been some of my best thinking and workshopping time.

EL KALIOUBY: I have to say, I love that use case because driving time is not necessarily downtime, but wouldn’t it be so productive if you had a conversation with an intelligent bot in the car and asked it to take notes and brainstorm and whatnot? But I will confess:

I did that once, and I got so into it because the interface isn’t quite there yet. I actually got stopped by a cop, so I haven’t done that again. But I love the use case, and I do think it will get better. That was a few months ago, and maybe it’s gotten way better.

JOSEBACHVILI: I think it’ll keep getting better. What’s yours?

EL KALIOUBY: I consult AI often as ChatGPT, but I’m increasingly using Gemini before I need to get any work done. When I’m brainstorming ideas for the podcast, when I’m brainstorming ideas for vacations with my kids, I go to AI first, which is really interesting. That’s a fundamental behavior change.

JOSEBACHVILI: It’s a great thought partner. It’s a great thought partner.

If you haven’t tried it for furniture or decoration shopping, I also highly recommend that one. I now have burnt orange and sage green throw pillows. I’m not a designer, so I wouldn’t have thought to do that, but ChatGPT told me it would look good, and sure enough —

EL KALIOUBY: There we go. I love that. I’m going to try. Maia, thank you so much for joining us. This was great.

JOSEBACHVILI: Thanks for having me.

EL KALIOUBY: It was fascinating speaking with Maia. AI is reshaping the way we shop, and I love that it has the potential to bring new customers to small businesses.

But I’m curious what the future of agentic commerce will look like. The chatbot is a compelling form factor given how ubiquitous it is, so integrating shopping there seems like a no-brainer.

But laptops and smartphones aren’t AI-native. What will agentic commerce look like as we discover new AI-native interfaces? How will it work in glasses or wearable pins? Like Maia said, this is still early days. There’s a lot of opportunity there, and as an investor, I’m keeping tabs.

Thank you so much for listening. We’ll be back with a new episode next week for Earth Day!