JEFF WEINSTEIN: Welcome. If you had asked me a few years ago what would come first, autonomous cars or AI helping us buy things online, I really would have said AI buying things online. 

But I took a driverless car to get here, and I still had to use my fingers to buy this clipboard online. I imagine everyone in this room, maybe even some of you right now, are using AI to research or answer questions. It's not a stretch that AI is going to help us purchase. As you saw in our keynote this morning, that's finally changing. AI is coming into commerce. But how will it actually work? What should businesses do to adapt to this new channel? 

That's why I'm so excited for our panel today, to hear from pioneering leaders that are bringing this science fiction of AI commerce into reality. To get started, David, cofounder of /dev/agents, what is an AI agent and why does it matter for commerce? 

DAVID SINGLETON: It's a great question. So at /dev/agents, we are building a product, which we call a next-gen operating system for consumers to work with AI agents. You obviously hear a lot about agents right now. I think 2025 is like the year of agents. And there are really two different definitions of what an agent is, and I think they're both kind of valid. 

One is anywhere that there is a particular function that has been done by a human and is being now performed by some kind of AI-enabled workflow, you'll find a lot of people calling those agents—good examples there would be customer support bots and so forth. 

And then there's a more technical definition, which is closer to the one that I think about every day, which is any time we have an AI system where a goal has been set for that system by a human or on behalf of a human, and it can actually operate either in a long workflow or maybe in a loop where it's perceiving the world—maybe by calling other tools, pulling in its own data sources—and it can take action, then it's an agent. Definitely large language models at the core, goal set for a human taking action. 

JEFF WEINSTEIN: And Traci, you're head of operations at Rye. How are brands and other ecommerce merchants adapting to this new powerful way of humans having AI agents? 

TRACI HIROKAWA YOUNG: Yeah, so to give you all a little bit of context before jumping into it, Rye is a universal selling API that essentially connects brands to platforms. So what Rye does is we allow you as a developer to be able to pull and display any products that are on Shopify or Amazon today and be able to submit orders back to them programmatically so that you don't have to worry about making individual brand relationships, managing inventory, figuring out shipping logistics, all of the plumbing. 

So all that being said, like you're saying there's two exciting sides to it. On the AI agent side, we're seeing a ton in the consumer buying space. And so, some of those early use cases that we're expecting to see really win there are those that are already established in the commerce space. So a good example of that would be corporate gifting. It's one of those sneaky big markets. It's actually going to hit a trillion dollars in the next few years. And it's also one of those things that I think all of us would love to be able to pass off to an agent and not do ourselves. 

On the brand side, there's a few things I want to call out. One is the support side. We've seen with a lot of our brands, we actually see that their support teams are up to 50% of their entire head count, which then if you're using AI agents to help your support team, you're either helping enable them or you're eliminating them entirely, so huge cost savings there. There's also worth calling out this idea of AAO, which I'm sure a lot of you have seen, the idea of AI agent optimization. It's essentially SEO now for AI agents. It's going to be a huge use case for brands, and the brands that are already working on it are the ones that are going to end up ahead. 

It's basically, can you align your product descriptions and data with what users are actually searching for. And you know, what I always like to tell brands is just go where the data is. What we've seen in the past is, you know, these big department stores and retailers, what they initially saw was for the first time, they can make this whole customer picture, and they can understand a lot more about the customer rather than just segmented to what you're selling. Now we're seeing the next level with AI agents. 

So not only are you seeing the entire customer details, but you're also getting data fed to you. So it's going to be a really exciting space in commerce. Commerce hasn't seen a ton of innovation disruption, so I'm very excited to see how AI agents push that forward. 

JEFF WEINSTEIN: I mean, ecom was already looking for personalization. That sounds like hyperpersonalization. 

TRACI HIROKAWA YOUNG: Yeah, exactly. 

JEFF WEINSTEIN: And this is obviously a new space, comes with a lot of hype, very exciting. AI is a huge theme at our whole conference. Paul, you're founder and CEO of Browserbase. Are these agents actually in production? What are the patterns you're seeing? 

PAUL KLEIN IV: AI is unleashed. AI is in production, agents are in production, and they're doing tasks for people online every single day. Now, if you think of AI in a lot of the work that we do every day, it happens in a web browser on a website. And I think where we've seen AI agents really meet the internet where it is, is actually using the tools that we all use. Now at Browserbase, we focus on running web browsers for AI agents. We power web browsing capabilities for large companies like Perplexity or Clay. 

But in the future, I think AI is going to be as smart as us. It's going to be able to do what we do, which includes purchasing things online, and it may not always have, you know, a native integration for it. It may have to go use the tools that already exist, and we're really excited to see how AI can work like we do using the same things we do, which is often a web browser. 

JEFF WEINSTEIN: Do you have an example where you're like, oh my goodness, thank goodness we have this AI agent? 

PAUL KLEIN IV: All my examples are pretty lame, you know, because I love the boring stuff. I want to see AI go out and automate my, we were just talking about this, like my Delaware franchise tax filing as a founder. I want to see… 

DAVID SINGLETON: I would buy that one. 

PAUL KLEIN IV: Right? You know, I don't think it's going to be booking a flight for me necessarily. I think it's actually going to be doing this long fat tail of tasks where there might not be a MCP server for that task, you know, there might not be a native integration. So where I'm excited is seeing not only like platforms like Stripe be able to offer ways for AI to go out and buy things and be embedded within checkout portals, but also ways for companies that are building agents to use the best tools to really work where there might not be something like a paved path.

JEFF WEINSTEIN: And that sounds like, so much of life is just munging through a task that you could imagine, wow, I could program a robot to do this for me, but I just don't have the time. But David, you've gone all in on consumer agents. How are you thinking about how this is going to cross the chasm to consumers? 

DAVID SINGLETON: Yeah, so one of the things that we did very early on—I should say, before starting this company, I was the CTO at an economic infrastructure…  

JEFF WEINSTEIN: A small payments company, right? 

DAVID SINGLETON: Indeed, you might have heard of it. 

JEFF WEINSTEIN: Yes. 

DAVID SINGLETON: And started to see how we could apply agentic workflows inside of companies and realized, hey, this is coming for consumers as well. So early on in the company, we worked with a small group of users, gave them a very early version of the product, and looked at what they wanted to do with it. And Paul already mentioned booking a flight. Things like that, folks are definitely going to be able to do with agents and on our platform. 

But what we find people are much more excited about are the more mundane tasks, the things that you do many times throughout your day. That can include things like remembering to tell a person about a thing, right? And actually, a lot of what we saw from those data were folks wanting to do commerce. So people want to say, I want to buy a new vacuum cleaner. Well, that's, like, actually kind of a big task. You need to first go and research which vacuum cleaners to buy, and then you have to figure out where you can buy them from, find them at a decent price, and then get them delivered, right? That's a perfect kind of mundane workflow that an agent can do really well. 

And you know, I'm sure there are many folks that are selling stuff to people in the room. The thing that I think is going to be interesting for the industry right now is, it used to be that when you had automated flows and bots and so forth driving through your checkout flow, you knew that was likely a sign of a problem, fraud, something bad. But now we have good bots. We have actual human beings that want to be your customers that have employed this bot to go get something for them. 

They want to give you money. They are legit. And there's a lot that we all can do as an industry to kind of embrace that and figure out how to make it work really well for them. 

JEFF WEINSTEIN: In some ways, the fraud detection might be too good in that case. We need some way to communicate, hi, I'm actually Jeff Weinstein and I've delegated myself to this new agent. How are you seeing brands, are they accepting AI agents? What kind of consumer relationships do brands still want to have when buyers might be in a different modality? 

TRACI HIROKAWA YOUNG: We've certainly seen that there's a ton of triers, but there's really a shortage of buyers. And that's one of the keys here, and that's one of the things that brands are still trying to help their AI agents that they're partnering with figure out, is that they can get people top of funnel. So in a typical case, if you think of classic software, if you get people through the top of funnel, you can get them through, you can close on them. It's actually the opposite for a lot of these AI agents, and that's what brands are seeing, is that there's a ton of drop-off throughout the entire buying process. 

And so we've seen this increased emphasis on your entire checkout process being incredibly frictionless. So to give you a good example of that, there are things called link-outs occasionally, which is basically if you go to add a product to your cart or if you go to checkout, you're actually redirected to a new browser or a new app, wherever it is, just not native to wherever platform that you are. That conversion rate, we typically see around 0.001%–0.1% of a conversion to actual checkout. 

JEFF WEINSTEIN: That's low. 

TRACI HIROKAWA YOUNG: That's incredibly low. These AI agents are actually 100 to 1,000Xing that by just providing a native checkout experience. And so we're seeing that up to 1%–11%. 

JEFF WEINSTEIN: And is the takeaway there that brands should radically improve their payment flows? Or is it that brands should work on their own AI agents? What would you recommend if a brand were sitting in front of us? 

TRACI HIROKAWA YOUNG: Yeah, so brands, we've seen a ton of use cases where they are using their own AI agents for various internal use cases as well. But we've seen a lot of brands partner with these selling platforms. 

And so I'd really encourage these brands to think about the entire selling experience. A lot of AI agents are coming back to brands right now saying, look at all these people that are interested, not look at all these people who are buying. So I'd really encourage brands to really focus on that and make sure they're focusing on the entire experience. 

PAUL KLEIN IV: Yeah, just tagging on there, we've seen a lot of brands who are using browser automation to actually monitor, hey, is my partner listing the price the right way? I have 20 channel resellers. Are they listing it correctly? Does the SKU match the description? There is a lot more to agentic commerce than just buying. 

There's also agentic listing. If you have a lot of resellers, how do you manage all these campaigns? Generally you have people automating that, or sorry, people doing that work. You can automate that with an AI agent internally. 

JEFF WEINSTEIN: It's like infinity secret shoppers. 

PAUL KLEIN IV: Yeah. And I think shopping in itself is a giant research problem, right? I mean, we've all tried to buy something, like maybe a new laptop, and you're looking at 20 different websites. If you want to solve agentic commerce, you have to solve a research agent problem. You have to be able to evaluate many different choices, look at pricing. 

If you add it to cart, when is it going to arrive? Because you might need this laptop tomorrow. This kind of research agent use case for buying, I think, is really important, because it's not only just buying the thing, it's picking the thing that really makes agentic commerce happen. 

JEFF WEINSTEIN: And what do we feel is missing from this being an early great idea with a lot of potential to be completely mainstream? Let's start with David. 

DAVID SINGLETON: I think if you try to actually complete any of these use cases today, like if you go out to build a product to do this, first of all, there's a lack of kind of necessarily understanding in the folks you might want to partner with to deliver things that this is like actually going to be good business. 

So, you know, hopefully we're helping change that here. It's also, I think, important to deliver a more delightful experience to people than they could get if they actually went and did it themselves. So nothing worse than asking—I don't know how many people have tried using ChatGPT's operator to actually buy stuff. It does work sometimes, but it rarely works faster than I could have gone and done it myself. 

So how can we have a system that has enough understanding of the user in the first place to be able to go off and enough reliability, as Traci was talking about, to actually complete the use case, where it's more delightful than what we had before? 

JEFF WEINSTEIN: Anything technically you feel that you would sort of wish someone in the audience would magically invent? 

PAUL KLEIN IV: Yeah, for us at Browserbase, we think that the infrastructure is what's holding back a lot of this space. Models are getting better and better and better, but you need reliable infrastructure to build these agents on top of. I really want to see someone innovate on the UI side. 

I think you're totally right. It's fun to browse the web. I like going on certain websites. I like clicking those. I don't want to automate that away. I want to automate away the boring part of the web. And I think there's some really interesting opportunities in consumer, in UIs or generative UIs on top of agents that make transacting and working with an agent really fun. Hopefully, requests for startups, go build that. I'd invest. I'd love to see it happen. 

TRACI HIROKAWA YOUNG: One other thing to add really quickly is on the AI agent front, the thing that makes it really great is mass adoption and extremely good retention. And so those are two things to call out. There's so many AI agents that are popping up. There's so many great experiences to try. 

But if you have a bad experience one time, consumers are just dropping that entirely and moving on to the next one. And so, really excited to see that. I'm really hoping that we can find a really great innovator who can just see both of those things, which are incredibly hard challenges. 

JEFF WEINSTEIN: And where do you think that the value will kind of accumulate? Do you think that there'll be new ecommerce that pops up that's kind of AI native? Do you think that the current set of popular retailers will be able to adapt to this new technology? Do you think that the major technology companies will all have some new version of shopping? How do you think this will all shake out? 

PAUL KLEIN IV: Can I hop in? I really like the idea of cloud kitchens here, and there's all these DoorDash restaurants that don't have restaurants. They're just kitchens, and they have something listed on DoorDash. And I wonder what's the cloud ecommerce kitchen. Maybe there's just vendors that only have agent presence or agentic commerce presence, and they don't have a traditional storefront. 

That's going to be really interesting to see what happens. And in the end, the value goes to the user, the buyer. They're getting more choice, maybe better prices, because they don't have to maintain that storefront. So there's some stuff you can see in food delivery that might carry over to agents. 

JEFF WEINSTEIN: I mean, when I use a search engine today, and it at the bottom says, oh you're on page one of infinity search results, it’s like, ah, thank you very much. I'll just go through infinity search results to find what I want. 

I often want much more personalization when I'm finding things. And I find that I want to apply the models of AI to all the products and services, both consumer and business, that exist without kind of paging through each and tapping through each. As fun as it is to close 100 tabs at the end of the day, I do want much more personalization in my search and discovery and purchase. 

DAVID SINGLETON: I think something else that's exciting and gives a lot of potential for more commerce to happen is, think about all the times that you actually intended to buy something, especially gifting is a great use case here. So, times that you intended to buy something, but you never quite got to it because it was just too much of a schlep to figure it out. I actually feel like I'm a bad friend a lot, where I'm like, oh, I'm going to buy someone a gift for their birthday. 

JEFF WEINSTEIN: Mine was two days ago, David. I did not receive it. 

DAVID SINGLETON: I'm sorry, so you see, I'm a bad friend. I missed sending Jeff his present because it was just too much of a schlep. But the point is, if I have a system where I'm keeping track that I wanted to buy him something and have an agent that will pick out a good gift and make it a single click. 

JEFF WEINSTEIN: It's the same time every year. 

DAVID SINGLETON: That seems like something we could do even without AI. Anyway, the point is, there are all of these proactive use cases where reducing the barrier to completing a transaction will make it much more likely that we have more of them. And I get pretty excited about that. 

TRACI HIROKAWA YOUNG: Yeah, I think also with opening up commerce, there's a huge opportunity for brands as well. So what we saw with Amazon and their initial mass adoption there was everybody was expecting that people would become brand agnostic. They would just choose whatever was first. What we actually saw in reality was that people did still tend to go to the brand names that they recognize. 

I think this is the next level of that opportunity, where brands can come in again, and if you can match what consumers are looking for, the brand name can matter a little bit less. 

JEFF WEINSTEIN: Do either of you have any questions for each other? 

PAUL KLEIN IV: I was going to ask, I mean, that sounds so asynchronous. Like, the asynchronous nature of doing work is, it feels like a lot of work is moving from synchronous, like clicking these tabs, to asynchronous. I'm curious how you guys have thought about the design and the UIs of that. 

DAVID SINGLETON: Absolutely. I mean, this is kind of the core of what we're trying to figure out, which is in a world where, as a person—and we're really thinking about people, not people working inside of companies—you have the ability to kind of tap a pretty intelligent friend on the shoulder and say, hey, could you get this thing done for me, and then you go off around the rest of your day. You do need some mechanism for them to say, hey, yes, I figured this out, but I'm not sure if you would prefer to have the Dyson vacuum or the whatever other brand, the Shark vacuum. They need a moment to come back and tap you on the shoulder. So I think we're going to see a real explosion of innovation in how people work with computers as we build these interfaces. And that is the exact core of what we're trying to invent at our company. 

PAUL KLEIN IV: Cool. 

JEFF WEINSTEIN: I had a really odd experience the other day. At Stripe, we’re building tools to help third parties be able to monetize their services in any new place. So we were trying this experiment where—Cursor, you've heard a lot about today, this very popular coding editor. And there's that moment where you need a third-party service as you're writing.

So we wrote this example application called emojigenerator.com, which doesn't exist. And it's a third-party service we made up that its job is to draw emojis. And so we put it into our editor and said, okay, emoji generator through the LM, please draw me a snake. It drew a snake. And then we put in, please draw me a cow, and it drew a cow. Then we made it a paid service, right? 

So we said, okay, the next call to the emoji generator is going to cost 10 cents. We hit save. We ran, okay, please draw me a cat. The LLM calls the emoji generator service. The emoji generator service says, I cost 10 cents. And the LLM comes back and beautifully says, hi, the emoji generator service costs 10 cents, here's a way to buy it. And it also said, or I could draw it for you, which was in the first instance of us at Stripe playing with this new technology to make embedded commerce, there was this really amazing moment, like, wow, this thing is pretty smart. 

I wasn't expecting that moment. So how do you think that control is going to work? It was such a powerful experience, but it really opened my eyes to the fact that we are sort of living amongst an AI. And it could be very powerful, but it behaved in ways I did not expect, and I am spending 24/7 thinking about it. 

PAUL KLEIN IV: Maybe, so we have a framework called Stagehand. It's basically like a great SDK to automate browsers. And the way we think about it is instead of a single prompt to have AI go out and do a task for you, as much as you can you try and break it down into a series of smaller atomic steps. And that creates great guardrails where let's say, you know, you’re buying something, you're doing commerce, go to the website, find the item, add to cart, three separate steps. If one of those goes off the rails and it clicks on a free survey, it's not going to be able to complete the next step. 

So the answer people don't want to hear is that you do need to add a little bit more guardrails and determinism, but that's how you can build production AI that actually works and does what you want it to do, as opposed to just going off the rails and subscribing you to 100 newsletters. Though your email is probably already that busy, so I wouldn't worry about it. 

JEFF WEINSTEIN: That might be true. David, you're putting so much AI power in consumer hands. 

DAVID SINGLETON: Yeah, I think this is really important. We've been thinking a lot about, there's a kind of, like, tension between as a user, you want to feel like you have control, but you also do want to give the agents a lot of agency. 

Otherwise, you're going to have to keep clicking yes, yes, yes all the time. I don't know if anyone's used Cloud Code. It's a really awesome, very powerful tool for actually working with you to build powerful code bases on your own machine. And it will ask you to confirm individual kinds of action, like classes of action, and you can say, don't ask me again for this kind of thing. So it's a really nice way that you can kind of show it what you're comfortable with. 

Ultimately, though, I still think that that tool is asking me too many times to say yes in its current flow. So we're thinking a lot about how the base system layer can actually understand how reversible and irreversible certain actions are, and how your preferences about whole classes of action can then be applied to future work. So yeah, it's the real core of what we need to figure out. 

JEFF WEINSTEIN: Have you heard from brands who say, I don't want to get involved in this? Like, what about a million refunds because of an out-of-control agent? 

How are they balancing the power of—and you were just, I mean, we worked together on many of the things you were shipping. We were looking, not that we were snooping on your charts, but you're scaling so quickly. I had to go change the y-axis on the chart that we monitor how you're performing reliably. Some businesses are going all in. How are they debating this? 

TRACI HIROKAWA YOUNG: Yeah, there is a selection here. I think it's important to remember for everybody in this room, everybody here, now that David defined this great definition of AI agents, everybody here knows what it is. The common consumer does not. And so there is still this level of trust that people need to build, and that is the mass adoption piece. So that's what we're seeing with brands, and that's the trend that we're seeing, is they will go where consumers are. They're willing to take on some level of risk as long as you are able to provide a delightful experience. So that's typically how the process goes, the evaluation process. It's how many consumers do you have, who's actually completing the funnel, and then what does the risk look like for me after, and they want to experience that. 

So even more important than, you know, I think it first was when we first started hearing about AI agents to actually launch with a delightful experience from day one. Otherwise, brands aren't willing to onboard. 

JEFF WEINSTEIN: And we hear similar feedback. We're baking in lots of controls into how Stripe is doing financial services for AI agents. So you have a budget. You can use X dollars per day. Please send me a text, which didn't get as reliably delivered this morning as we had hoped, when you need more funds. 

The same way you would treat a pretty great intern, you know, we're trying to build those capabilities into the API so that you could delegate with control. Maybe we'll go to each of you real quick. What is a favorite use case you sort of hope will exist in a year with AI agents that's beyond a chatbot? I know you really don't want to file your Delaware tax returns, but I'll come up with something else. 

PAUL KLEIN IV: I'll go second so I can think about it for a second. 

JEFF WEINSTEIN: Yeah.

DAVID SINGLETON: Actually, the thing that we talked about already, I really, really do want that. I am a really bad friend in that I always have these intentions of things that I want to do, and I can easily write them down on a list, but I so rarely actually get to completing them. So that is the thing that there's no product that does this today, and if I could use it, I would use it very frequently, so I really want that. 

TRACI HIROKAWA YOUNG: In somewhat of the same vein, I think one really exciting one is around subscriptions as well. We're seeing a lot of agents around single purchases or one-time purchases, reminders of like, hey, integrate my calendar and send my friends gifts for their birthdays. 

However, that is a one-time agent that you're running. We're not looking at subscriptions. So typically, like, what is my actual usage? Like, take my family's activities and tell me when I need toilet paper. Those are some really exciting, I think, use cases that would increase that repeated adoption that people need to build great AI agents. 

PAUL KLEIN IV: Yeah, maybe trying to go super ambitious here, I want to see an AI agent 100% end-to-end own a workflow. It feels like it's owning so many different parts. It's like, oh, it'll do customer support, but it won't handle the refunds part yet. Or it will be your AI BDR, but it won't send the Docusign and close the deal, right? 

Like, I want to hand off a full business function to an AI agent and have it reliably work. That's what I'm really excited for. And it feels so, so close. And the models are getting better, tools getting better, infra is getting better. Within the next year, you're going to be able to hand an AI agent some business workflow and have it be owned end-to-end. And that's just going to be a really exciting moment. 

JEFF WEINSTEIN: David will know that this question is coming, because we’d sometimes do interviews together in the past at Stripe. Let's imagine that's a 5 out of 5 ambitious answer—and this is from Airbnb's Brian Chesky—what would a 6 out of 5 ambitious answer to that same question be? Imagine you have infinity robots with infinity compute and infinity time. Now what would you have it do? 

PAUL KLEIN IV: I guess I gotta go level up my answer then.

JEFF WEINSTEIN: You'll file in two states. 

PAUL KLEIN IV: Yeah, I mean, I think the 6 out of 5 answer of ambition here is that AI is going to be as smart, if not smarter than us, and do all the things we do. We're going to be the slower operator online compared to AI. And that means that the web is going to be built for people still. Like, if you don't have AI, we're the consumer of last resort, right? So I imagine a world where we're going to have to build websites not for AI, but because people still exist. And that's going to be really interesting because the web will be built for people once again, because AI is just going to be able to browse it as well as we can. 

JEFF WEINSTEIN: Six out of five, David?

DAVID SINGLETON: Six out of five on the exact same product dimension I talked about, I think it's pretty easy. So, rather than me saying that I want to do this thing, it knows me, it knows who I care about, and it should automatically say, I've already gone and researched this, here's the perfect gift for Jeff, this is the note you might want to send him, and I can say, yeah, I really do think that's a wonderful gift, but I'm going to tweak the note, make it a little bit more personal, off it goes, you've got your stuff. So it proactively tells me what to do, and I like it and trust it. 

TRACI HIROKAWA YOUNG: I think to go back to the commerce space again, what the 6 out of 5, it's funny, this sounds like a lame 3 out of 5 answer, but for commerce, there really hasn't been too much disruption to date. If brands are able to adopt that throughout their entire workflows, everything right now in the commerce plumbing space is incredibly manual and it's incredibly painful. And so any adoption in AI there that then eventually feeds into consumer activity and reaching more consumers generally is going to be a huge win for brands. 

JEFF WEINSTEIN: So we have, it looks actually a completely sold out room. Let's imagine that, you know, by the end of the week everyone here is going to take one practical next step to getting themselves or their business involved in agentic commerce. What should they practically do? 

TRACI HIROKAWA YOUNG: I can go ahead and start with the brands. One of the things with brands is just that they all want more channels, they don't know how to connect to them. Not to plug Rye too much, but just finding solutions like that that allow you to find more of these selling channels, that allow you to find the AI agent selling channels, that's a very reasonable next step. Right now, I think brands hold their selling channels really close to their heart, but to be opening that up is meeting customers where they are, meeting consumers where they are, ultimately will lead to growth and further AI adoption there. 

PAUL KLEIN IV: Yeah, if you want to go do this, you can go do it today. You can go build a POC. It's never been easier. Use Cursor, right? I think that's what they were demoing earlier. We have a cool docs guide on Browserbase about using Stripe to issue a credit card, then using that credit card to go buy something. I think the demo example is a donation to the American Red Cross or something. So you can go build an agent that buys something today. And hopefully, that's just the little POC that turns into a really big, beautiful company one day. So, excited to see what you build. 

DAVID SINGLETON: Yeah, and riffing on that, I think you can go try the little prototypical consumer interfaces for this stuff that exist in the industry right now. I already mentioned ChatGPT's operator mode. Go try it, see what it can do. I think it offers some glimpses of what's going to be possible in the future, and you'll probably then have a lot more questions that you want to answer for yourself.

JEFF WEINSTEIN: For me, you really just need to use this stuff yourself in any possible way and kind of be very sensitive to all the frictions we've already sort of baked into our daily lives that I think one day will just become, wow, why were we doing that? 

And so almost just like write a list, like why was this so hard to get this done? Or if I had another 24 hours today, what would I do? I think that is a very good glimpse into where agentic commerce is going. I want to thank Traci, Paul, David, our panelists, for this lively discussion. A big takeaway for me is there's such a variety of businesses that this is going to impact. 

It's not just startups. It's larger enterprises, it's retail, it's services, traditional, new, that are going to use AI to find new channels and new ways of serving their customers in discovery. Very exciting. We're really excited to hear from you all. We'll also be right outside the room to chitchat, take any questions. Otherwise, thank you.