A New Kind of Marketplace

Demetrios Brinkmann [00:00:10]: I'm with Donné and Pedro. Pedro's working at OLX. Donné's working at Prosus, all doing AI stuff. And we're going to talk a little bit about what their visions are and how they're working with agents. So Pedro, you've got all kinds of cool ideas. I want to hone in on the idea of how you take the vision of what is possible and then go into actually like executing on it. And then later on, let's talk a little bit about where your vision has been thinking about just like these millions of things that we could be doing with the new paradigm of agents.

Pedro Chaves [00:00:58]: Sounds exciting. Thank you for having me. Okay, so just to broaden up the scope a little bit, so to be fairer to what OwlX is, we actually operate, let's say, in two main verticals and then the general classifieds. So we have motors, so essentially cars. We have real estate, so two very specific verticals, and then we have general classifieds. Maybe we can take a step back, I think, which is as part of these disrupt efforts, both of us within, within OLX, the idea is really to come up with new experiences for our users, whether sellers or buyers. But I think the first thing that was really required and is still required is to actually understand what's this new experience, what it What's the disruptive experience? How do you go from the current status quo, the current way of doing things, into another thing? And what's this thing? And that's probably the most difficult step across the whole development. I'm not sure if you agree or not, but that's my perception.

Pedro Chaves [00:02:04]: And thankfully, it's mostly a challenge for product people. The whole challenge of trying to come up with a new way of doing old things It's not something I envy because there are always a lot of constraints, right? Every time you want to innovate, you end up with the same, in the same old places, in the same chat interfaces, in the same UI interfaces, in the same floating widgets or something. And it's not easy. But I think that both within motors and real estate, we were able to carve out these new experiences within real estate for the buyers, within the motor space for the sellers. And I think we were able to deliver something which is disruptive within these specific areas for a few reasons. Some technical reasons, some design reasons, some product reasons.

Demetrios Brinkmann [00:02:57]: And motor and real estate are where folks can just say, I'm selling my house or I'm selling my car, and then they find— it's like peer-to-peer selling. In that regard?

Pedro Chaves [00:03:08]: Yes, although as major categories within OLX, they are super focused on professional sellers. So most of our customers are real estate agencies and most of the automotive customers are stands. Car— Car dealerships.

Demetrios Brinkmann [00:03:25]: Dealerships, that's the word.

Pedro Chaves [00:03:27]: That's the word I was looking for.

Demetrios Brinkmann [00:03:28]: I couldn't find it either, so I understand.

Pedro Chaves [00:03:33]: So yeah, so the challenge is you as a buyer looking for a house, what does it look like for you to have a different experience than you normally have, which is basically selecting filters or going on a map and just trying to understand whether you can find something within your budget and your preferences, lifestyle. So how do we translate this into a new experience? And the way that we did it, or the way that our product team did it, was okay, maybe the shift here is actually considering looking for a house as a major lifestyle event or a major life event, but mainly driven by lifestyle. Meaning when you're looking for a house, you're not just looking for a house, you're looking for the amenities around the house, the services, whether it has parks, neighborhoods. Yeah, exactly. The neighborhoods, what type of neighborhoods you are looking for. So the challenge was really to try and translate this into a flow where an agent tries to extract this from you without being too invasive, without being too laconic, without being too funny, with the proper tone, and very quickly, hopefully, getting the persona out of you, trying to map you into an intent, which of course is dynamic. We can go into, into all those details later, but the goal really is to have an interaction with you and being able to model you in a way to build your profile and then provide the— within our offering, within our stock, provide the best matches for you, which are enriched in a way that usually these platforms don't do it. Specifically, making sure that these new houses are within like an acceptable commute time from your workplace, for example, making sure that like the neighborhood safety scores or services scores are sufficiently high to match your preferences.

Pedro Chaves [00:05:28]: And basically using all of that and enriching all of that and presenting to you as, well, recommendations.

Demetrios Brinkmann [00:05:35]: And this is just a chat that I say these are the things I care about?

Pedro Chaves [00:05:38]: It's a mix. So we have not generative UI specifically, but basically components which you pre-populate with content which is generated by the agent. So you can click and choose and not having it be 100% chat. Because we have these intuitions which are maybe wrong, that maybe pure chat is not the way that we want to go. But then again, all the examples that we saw yesterday were pure chat, so I don't know anymore. And so between this mix of chatting and clicking and a nice navigation then on the map page, on the listing page, you have this new, more enriched experience of searching for a house.

Demetrios Brinkmann [00:06:25]: And now the whole idea of coming from idea to execution, that was something that I know you were fascinated with because I think you've had a few war stories.

Donné Stevenson [00:06:38]: Yeah. I mean, so yeah, we come from the same perspective, right? We're like, there's an opportunity to disrupt an experience here. And then what What does disruption look like? You know, if you can change everything, but if they're not, if it's not a valuable experience, then you've changed it for nothing. So figuring out the limits of what people are willing to accept in terms of the change of their flow in order to get a better experience. And for us, that was like, it's an ongoing process, I think, as well. We're like learning, iterating in smaller steps. So, so I think like from the real estate perspective, because that's also like a high trust experience that they have to try and, you know, to try and engage in because they have to really be like, this is a more reliable experience than if I search by myself.

Pedro Chaves [00:07:28]: I'm not missing out on anything.

Donné Stevenson [00:07:29]: Yeah, exactly. Yeah. So yeah, it's just like, how do you start that process of earning their trust or do you go straight to like a full-on chat experience?

Pedro Chaves [00:07:40]: I think there are a couple of ways of going about this. So we had some previous data from similar initiatives like Magic which allowed us to model the user behavior a little bit. So try to understand how exactly people and the users were interacting with these platforms. That's one way. The other way is through research, right? Because you can roll out these prototypes, whether functional or not, with a user base and just extract feedback out of them, whether it's explicit feedback, as in asking them what they actually think about things and the user interface and the flow, or implicit, as in in how they navigate the proposed app, the prototype. And then with controlled rollouts and I guess having in place a sufficiently elastic structure to quickly iterate on whatever comes back from these rollouts, right? So our calendar is very detailed in the sense that we are doing initial rollouts for internal teams, initial rollouts for internal teams, but slightly, more encompassing and bringing over real estate experts, then going on to public beta but invite only, and then eventually the full rollout to, to the whole user base that we have from Immoverto. And throughout all of these moments, you have to be ready to collect feedback and iterate on whatever you are doing. So I guess there are a few ways of validating how we are translating this vision, but there's also a lot of Trust in the process, trust in what— trust in our own vision for the product and that it will eventually lead.

Demetrios Brinkmann [00:09:12]: Yeah, it's going to be a better experience. Yeah, I could see it easily. As you said, buying a house, that's probably one of the bigger decisions that you're going to make in your life if you're the average person. You don't go and buy 10 or 20 of them through your life, maybe 1, 2, 3 max. And so you want to be very sure when you're buying the house, what you're getting yourself into. And you potentially having this, these questions, just asking questions about, are there playgrounds around because you have kids? How's the school system? You don't get that on other platforms.

Pedro Chaves [00:09:53]: I mean, there are some, some offerings, but not, we believe, as sophisticated and as enriched as we are trying to offer. That's—

Demetrios Brinkmann [00:10:03]: are you also doing that with the auto stuff? Because that also is a big— like, you buy cars probably more than you buy houses, I would assume, for the average person. So, but it is still a lot of money that you're putting down. Yep.

Pedro Chaves [00:10:19]: Not right now. We haven't been working on that experience for the seller, for the buyer side on motors. We have been focused, focusing mostly on the seller side, the dealer's experience.

Donné Stevenson [00:10:30]: I mean, this is again, like a choice you make with development, right? To develop to a sort of more like mature or sophisticated place and then go live or to sort of do it very small, very beta pieces, you know, that more people see. I think for us with Adidas, like it's a little bit safer to do that because they're not, it's not a single action they're going to take over the platform where you have one sort of flow with them. And then once they've bought the house, probably for a few years you're not gonna need to interact with the dealers. You know, if there's a small mistake, they will probably be back. So you have a higher risk tolerance because they'll be back, so you can, you know, try again. It's obviously not ideal, um, but I think in that case we, we had a little bit more safety in releasing very quickly.

Demetrios Brinkmann [00:11:19]: Um, so, so what are you doing with the dealers in the agentic way now?

Donné Stevenson [00:11:25]: Yeah. So I think that's, we're trying to figure out what, you know, what does it mean to use agent, like an agentic experience for dealers beyond the chat UI? I think this is the big question. Like, well, obviously you start with the chat UI because that's how we can start to collect data. We start to earn their trust and we start to like figure out what they want, but how do you move past the chat? And this is kind of the big question because there is a, like, an overprodu— like, overflow of, of how many, like, chat interfaces there are. And, like, for the dealers, what we're seeing is that they are— they have their routines and they know how to navigate the platform to get what they want. And so you have to interrupt that flow in order for them to find value in the agent. And with the initial test we did, we found that actually for them to move into a raw chat is very, like a huge hurdle because they're not sure what questions to ask. They don't know what the agent can help with, right? Because it can't do everything at this point, especially when you start, right? It's got limited functionality and they kind of either ask nothing or they ask like massive questions where you're just like, oh, that's a lot for one, you know, for V1.

Donné Stevenson [00:12:38]: So there's definitely a learning process there for how to give them value without alienating them, I think.

Demetrios Brinkmann [00:12:48]: I was thinking about this yesterday with the Tiago agent, the iFood payment fintech agent, and how there's this necessity for agents to really help you. They kind of have to embed in everything. And with the fintech, if it's embedded in your bank and it's embedded in your P&L, it's really able to do a lot more because it knows, oh, here's all your invoices, here's what your cash flow looks like. But for— in order to do that, like, you got to ask for, can we be your bank? Can we do this? Can we know your P&L? Which can be a big ask for a lot of things. But I'm thinking about like with these sellers, the more access that they give you, the more you're going to be able to help.

Donné Stevenson [00:13:40]: Yes. And that's where the trust is the important part, because if if they kind of start to see value or they start to see impact, then you can progressively ask for more trust from them. But if you kind of start on your first release and immediately you're like, can I have access to your— that's probably not going to go very well.

Demetrios Brinkmann [00:14:00]: You should try it though, see if it works. Tell me how it goes.

Donné Stevenson [00:14:03]: I mean, we like, it really is a process. We, you know, the way we released is we had the chat, and then we had like a couple of shortcut buttons that they could use. And one of those, all it did was like on the page create like a filter for them so they could see a specific view of their data. But the way we labeled it, they were concerned that it would take an action on all of those ads, and so they refused to press it because they weren't sure what would happen. And that, that's kind of like, you know, the trust process, and I think also the education process. But yeah, I mean, if if we released a button that actually took an action on their behalf right there and then, you can see we would have immediately like lost them.

Pedro Chaves [00:14:45]: Yeah.

Donné Stevenson [00:14:46]: So, you know, it is a progression of in this case we'll just show you the things you might want to like act on. And then maybe once they trust that, you can sort of start to, do you want us to take an action for you? And then only do you start to like automatically do things so that, that progression of like earning their trust, I think is the only way you sort of start to eventually get insights into much deeper things. That being said, with agents becoming more readily accessible, I think people are gaining a certain amount of inherent trust to just have their stuff integrated because they like the convenience. But I think it's been a process to get there.

Demetrios Brinkmann [00:15:21]: I just realized you don't really have a Ctrl+Z and the undo button for when an agent takes an action. It would be very nice if you could just, oh, undo.

Pedro Chaves [00:15:33]: Like the revert changes in Cursor.

Demetrios Brinkmann [00:15:35]: Yeah, exactly.

Donné Stevenson [00:15:36]: That's actually a really That's probably a design element we should start thinking about in agentic products as a general rule.

Demetrios Brinkmann [00:15:43]: Well, especially because, yeah, if you're— if someone is a little bit skeptical or hesitant to take an action, knowing that if they know, oh, well, I can just roll back really easily with a Ctrl+Z type of fast movement, then I would be more aggressive in the actions I take if I can always just like go back 5 steps.

Donné Stevenson [00:16:04]: It's a really, it's a really interesting idea.

Demetrios Brinkmann [00:16:06]: I'll be here all week. Don't worry. Happy to give you all kinds of great ideas. What you're really talking about here is something that I've been saying for a while since I think like, I don't know, 2024 on the interface is the main friction point and people not understanding the interface and not understanding what it's capable of, what it's not capable of, what they can ask of it, what it will do, what it won't do. Because it is a new interface, really feels like it gives you this wow moment, but it also can make you give you pause.

Donné Stevenson [00:16:47]: Yeah, I think I kind of think of it as a learning curve, right? Because you're giving someone a new experience, often with like not a lot of information, and then they kind of have to sort of trial and error their way through it. And that's where the trust element comes in again, right? Because you're not going to trial and error if you're worried that a mistake will be long-term damaging. So like for us with the Automoto project, um, I think that the chat UI is the natural start because that's what— how we kind of think about agentic problem solving is that it, it does the thinking and then gives you an answer to like a normal question. So our chat agent does that at least initially is it's, I kind of think of it as a, like a data analyst for the dealers where they can get more insights into the data. Cause there's a lot and there's across multiple like pages and like, you know, synthesizing that into a conclusion is a bit of a process. So the project kind of started with the idea that we could, we could give them those insights, but I don't know quite how we landed up having a an interface where the chat was only a component of what they saw. And then we gave them also what we call like shortcut buttons or quick actions. And some of those are AI and not all of them are, some of them are workflows.

Donné Stevenson [00:18:04]: But the, like the interesting thing there, and this is where I'm kind of becoming more of a like zealot for just the UX is everything when you're talking about agents and functionality doesn't matter until your UX is solved. Because we did see they asked questions that were either too big or like very unsure. But if you gave them an action that fed them into the chat with like a preset thing that we could do for them, then they were, they had a better idea, they had a much clearer idea of what to ask. And then they were more likely also to just follow up with natural questions. And also I think that's for me at least like changing how I'm thinking about building agent products because they get really smart and there's still this resistance. And so you're not getting the full use out of them until you figure out how to give people a mechanism to use it that makes sense.

Pedro Chaves [00:19:00]: I think that we are stuck in this transition step between pure chat interfaces and non-chat interfaces and what non-chat interfaces actually implicate. Meaning, does the user understand that if it's clicking a button, It's also AI behind it. It's also doing something intelligent behind it. So maybe generative UI is like a step in between. Like you generate the elements that you need to interact with the user. Maybe. I don't know. I'm pretty dumb when it comes to UX/UI.

Pedro Chaves [00:19:33]: But maybe there's something there because then you don't need to like predefine all of the visual aspects and you don't need to go with pure chat. Maybe there's something in between. Maybe you can have a selection of widgets that can be populated or used by the agent in order to convey the information to the user and then to extract information from the user as well and extract actions from the user.

Demetrios Brinkmann [00:19:54]: So what you're trying to do is have the user take a certain set of actions or gain insight. In this case with the auto, they have all this data. How can you see it differently?

Donné Stevenson [00:20:05]: Yeah. And how do you act on it? I think that's our next hurdle is that, right? We have a way to show them information and how do we give them a way to do something with the information? And that's where there's like a, I like this idea of like an interactive components, right? So if you were to say okay, to the agent or whatever, like here's a predefined things of you can, that you can probably do, but only show the mechanism to do it to them when it matters. And then the, interactive UI components are great because then it's not just, hey, this is possible, do you want me to do it? And then they kind of just say yes. But if there's the capacity to press a button or like some, like where they're taking the action, but obviously the AI is—

Pedro Chaves [00:20:51]: Engaging.

Donné Stevenson [00:20:52]: Yeah, exactly. I think that's a really interesting idea and maybe more of what we need to see from agents than either full autonomy or none.

Demetrios Brinkmann [00:21:03]: Well, yeah, it does feel like there's an easy win. You show a dashboard of data that they didn't necessarily get to visualize in the past, and you also say, you know what would help this number go up? If you buy more from us. And then you give them a few options of how you can— hey, I think you could change this number if you do one of these things. Do you want to test this or all, test them all. That's A, B, C, or D, and D is all of the above.

Donné Stevenson [00:21:35]: Things like that are starting to happen, like in Cursor, when you put it in planning mode, then it will ask you a bunch of follow-up questions and then it's just like a multiple choice form, which is, yeah, I mean, it's a really nice way to keep gathering new information, but again, moving away from just raw chat because it would be a bit of, it is a bit of frustrating when you're in these flows and have to type the whole time. This is what I meant and everything.

Pedro Chaves [00:22:01]: Inefficient.

Demetrios Brinkmann [00:22:01]: Yeah, yeah. And then on the other side of it, it is probably more frustrating when it— the agent goes off and thinks for a while or does for a while, and then it comes back and it's not what you wanted at all.

Donné Stevenson [00:22:14]: Yeah, the assumption was wrong and it's gone down.

Demetrios Brinkmann [00:22:16]: Yeah, so I would rather have the upfront questions and front-load that, and then I can go and get some coffee while it's working.

Pedro Chaves [00:22:24]: Yeah, exactly.

Demetrios Brinkmann [00:22:25]: All right, Doné, you've got to leave us for an imaginary meeting that you're going to.

Donné Stevenson [00:22:31]: Yes, yes.

Demetrios Brinkmann [00:22:32]: And Pedro and I, we get to talk a little bit about the future and what you've been thinking about as potential realities in the marketplace.

Pedro Chaves [00:22:39]: Sci-fi meets marketplaces.

Demetrios Brinkmann [00:22:41]: Nice.

Donné Stevenson [00:22:41]: Nice. All right, cool.

Demetrios Brinkmann [00:22:43]: Thank you. You're trying to rethink the way that you bring agents into the equation.

Pedro Chaves [00:22:52]: Yep, that's correct.

Demetrios Brinkmann [00:22:54]: What are some ways that you've been thinking about?

Pedro Chaves [00:22:56]: Within these general classifiers, it's where we find, I think, these not just within general classifieds, but maybe it's more obvious that these opportunities are there because that's where you find people which are offloading their, their, their stuff, which are tired of their chair and they want to refurbish their, their room. They are looking for new mics, but they don't really want a new one. So maybe in there somewhere between the professional sellers who actually make a living out of it and there's a decent chunk of people actually living off of, or at least we think they make enough money to live off of just spinning things on OLX. And the other types of sellers which are just, let's say, occasional sellers, maybe there's an angle there or maybe there's a couple of angles there.

Demetrios Brinkmann [00:23:43]: Yeah, and what have you been thinking about? Because some of the things that I can imagine are just easier uploads and so instantly I throw a photo of my stuff and you have a description and you have, and you kind of have that in place already so that I don't have to think as much when I want to upload something. You suggest prices, prices.

Pedro Chaves [00:24:03]: Yeah, in a way. Yep.

Demetrios Brinkmann [00:24:04]: You kind of like tell me that's a little high for—

Pedro Chaves [00:24:08]: there are some price prediction levels.

Demetrios Brinkmann [00:24:10]: Yeah. And what else are you thinking though? Like, to— because the whole idea is to ease the friction, right?

Pedro Chaves [00:24:18]: Yep, exactly. And as a classifieds business, the main point of friction is actually the transaction. So you as a seller, me as a buyer, We kind of connect through the ad page, but then it's up to us, right? And so I have to send you a message and say, hey, is this— most common question, is this still available?

Demetrios Brinkmann [00:24:37]: Yeah.

Pedro Chaves [00:24:39]: Is it negotiable? Like, are you sure? Can't you like skim something off of the top? Can you give me a discount here? Just have a kid, can you help me out?

Demetrios Brinkmann [00:24:46]: Yeah.

Pedro Chaves [00:24:47]: So this is the actual attrition and probably where agents could start helping, like, because imagine that not just from the perspective of a seller, but also from the perspective of the buyer. Where you really want that new toy for your kid, but you have some budget constraint. You sort of have a time urgency as well, right? You have a time window where it makes sense to buy that. So you either spend all day searching on OLX, refreshing your searches, your saved searches, and making sure that the buyers still have this and that and repeat this all day, or you offload this to an agent and you just instruct your agent to go out there, search according to your parameters, And eventually, if they find something, actually negotiate for you because you know your budget, right? Maybe you can budge on your budget. Maybe you can go a little bit on top, but you can formalize this. You can tell your agent, hey, this is my budget, but I'm open for negotiation. And maybe the agent, if they get in touch with the seller and they can't really comply with your constraints, maybe they can come back to you and say, hey, I found this, but it's not in the area you wanted or it's not in the budget.

Demetrios Brinkmann [00:25:50]: Actually, I'll tell you a true story of when I was in Portugal and I used OLX to buy my daughter a roller, like one of those little scooters. And I do remember that we spent a considerable amount of time with the back and forth of the sellers deciding when we were going to meet, where we were going to meet. Why would you do that? Yeah, exactly.

Pedro Chaves [00:26:11]: It's, it's very— that's scheduling. It's a solved problem.

Demetrios Brinkmann [00:26:14]: Yeah, yeah, exactly. It's like, oh, this is a solved problem now. And even like the next step is something that you were just saying, and I was like, we got to turn on the cameras for this, with the lockers where the agents can then play with the locker idea, and then you don't have to even worry about when to meet someone.

Pedro Chaves [00:26:34]: Yeah, that was full improvisation on my part based on the hiring of humans that you mentioned before.

Demetrios Brinkmann [00:26:39]: Yeah, for context on that one, I saw a website where the Malt can now hire humans to do things in real life. Yeah. And so you can hire a human to sign a contract for you or to do whatever it is that they can't do inside of the computers, which I found fascinating. But anyway, then you were like—

Pedro Chaves [00:26:58]: and within that, within that context, I realized that maybe, well, maybe smart lockers are another, like, a physical arm of these agents, right? Because, because they could deal with the whole logistic chain and just make sure that whatever smart locker supplier that it's next to you or closer to you now has your roller for your kid.

Demetrios Brinkmann [00:27:17]: Instantly my mind thinks like, oh, my agent has a budget for the roller, but it finds a roller that isn't in the budget, and so then it goes and it buys like 5 different items and flips them, makes enough money to be able to buy the roller, and then— or loses all my money. Yeah, about that.

Pedro Chaves [00:27:39]: Or gets you banned off of Rolex. But yeah, that would make sense. I mean, the whole logic of arbitrage is something that we have explored, like flipping objects with a, with a certain almost guaranteed margin, like identifying good deals, making sure that they are in sufficient condition for us to like up price a little bit, just like tweaking the description or tweaking the pictures or something like that, and then reselling for a profit. That's actually a project that we did And again, there's a chunk of users for whom this would make sense.

Demetrios Brinkmann [00:28:13]: Yeah.

Pedro Chaves [00:28:13]: Within a lot.

Demetrios Brinkmann [00:28:14]: I love watching YouTube videos on people that go garage sales and like yard sales, and then they find, you know, the golden Pokémon card and it's like, this is actually $2 million, wow, and we found it in a garage sale. And that is something that it feels like, oh cool, those are the people that are the professional sellers. Those, or maybe they're not professional, but if they know enough, about what to look for in the— in real life, then they can sell it online. They can make some money. The only problem I have with those videos is that they kind of like say, oh look, this is on eBay, it's selling for $60 right now. So they kind of— in the videos, they'll basically say this is $60, but they don't sell it yet. And so it's not really $60 until you actually sell it, because I see people like they'll buy a whole lot of mugs, coffee mugs. I don't know who's buying coffee mugs on eBay, but they say like, oh, there's $5, this is an extra $10, whatever.

Pedro Chaves [00:29:17]: I'm like, and maybe they have their own mental models and they are really confident of—

Demetrios Brinkmann [00:29:22]: yeah, like this sells like hotcakes. Exactly, exactly.

Pedro Chaves [00:29:26]: But we are still getting humans in the loop here, right? Because, okay, we are in the wake of this mega hype from Clubhouse, OpenClaw, and then Motebook, right? Motebook now has close to 2 million agents operating on it, whatever that means. Full of the degeneracy, full of grifters, full of scams, whatever, makes sense. But at the same time, again, 2 million independent agents, maybe not fully independent, but 2 million agents actually ended up there, right? Autonomously. So what if we are the first stepping into that place, but from a marketplace perspective? What if we are the first ones to, and using Nish's words, Nero's words, to actually build the harnesses for an agents-only marketplace? Because then we skip all this attrition involving humans for the better or the worse, but at least we would be able to see what would happen if we have autonomous sellers, autonomous buyers with access to some kind of stock. So let's think about classifieds still. Thinking about secondhand stuff. But what would this look like, right? Because technically we can do it. It's fast.

Pedro Chaves [00:30:41]: I mean, Moldbook was built, I don't know, in a couple of days, right?

Demetrios Brinkmann [00:30:44]: Yeah.

Pedro Chaves [00:30:45]: Yeah, so there's really no actual barrier for us to occupy that space. No actual barrier for us to build that structure for independent agents to act in a marketplace, like an agent-to-agent marketplace. Yeah. We have the payment integration. Stripe will be chatting about it today. Google will be chatting about it today here at Prozis. So like even the payment infrastructure is there. At OLX we don't have that payment layer, so it's all, um, yeah, you get paid in cash.

Pedro Chaves [00:31:15]: Exactly, exactly. But what if we act as escrow? What if we take a buyer's take rate, seller's take rate, whatever, from these autonomous interactions? I mean, it's incredibly attractive. And I think what— I mean, apart from all the security issues, apart from all the hype that will just be hype, I think what changed in this past couple of weeks was that maybe the general population is now actually considering having agents working for them and with them on a day-to-day basis, doing more stuff. Doing more stuff, exactly. Outsourcing actions to them because the same way that everyone started getting used to using a chat assistant through ChatGPT, like a clear qualitative change in the way people interacted with apps. Maybe we are upon that step now with a new change where we can actually trust an agent to, well, maybe not trust, but we can actually have an agent doing stuff. And I think we should ride that wave.

Demetrios Brinkmann [00:32:17]: Talk to me more about this idea of harness and building the harnesses.

Pedro Chaves [00:32:20]: Everyone or mostly everyone is focusing about building the agent itself, but there's a lot of incumbents there, right? We have, I mean, all the big tech is building their own assistants. I'm using Gemini. I'm not sure what to use. Someone here uses Anthropic.

Demetrios Brinkmann [00:32:37]: Yeah.

Pedro Chaves [00:32:37]: ChatGPT, whatever. Probably the capabilities from a user perspective are already being built or already exist within these agents or are easily extended with MCP or tools or whatever. What we really need is the infrastructure with which they interact, like the actual platform for these agents to, in this case, negotiate their items, to make their postings, to upload their pictures, their descriptions of their items. This is what the harness is. This is what the exoskeleton for this structure is. And probably what we should be focusing on right now. Before it gets irrelevant and then you have agents talking to agents directly.

Demetrios Brinkmann [00:33:21]: That was what I was gonna say, like, why would there need to be a platform in between? Why would it not just be agents talking to agents directly and sending photos or putting out a bat signal saying, "I've got this to sell if anybody wants it." Technically speaking, currently, for that agent-to-agent communication to exist, you still need some sort of configuration in between, right?

Pedro Chaves [00:33:43]: So either start by exposing the agent first as a sort of a server and then have the other ones explicitly connecting to it. As of now, we don't— you don't really have that sort of mapping capability or radar capability of checking out what's out there. And in order to check out what's out there, especially when we have multiple, multiple technical solutions, multiple existing assistants, I don't see how you can go around them having some sort of intermediate layer that at least tells you what other agents are out there. And if it tells you what other agents are out there, which protocols they are using, which preferences they have or whatever, maybe there's room then to still carve out something specifically for marketplaces. Because then you will have this layer where, okay, maybe you don't exactly provide all of the capabilities of a traditional marketplace, but you provide a discovery of other agents. And as of now, I don't think there's a good solution for it. I might, I mean, everything moves so fast, maybe I'm wrong, but I haven't seen it.

Demetrios Brinkmann [00:34:44]: And when you're talking about marketplace, you're talking primarily about selling things, right? Because I've heard and seen—

Pedro Chaves [00:34:51]: or services.

Demetrios Brinkmann [00:34:53]: Yeah, I've seen a lot of folks try this idea of, oh well, in the future agents are going to need things done from other agents, which I'm not so sure I fully grok or believe is going to happen. But that is trying, that is like in flight right now, and I haven't seen any of them take off as of yet. Maybe it's because we needed this moment in time, or maybe not. But you're really talking about like, how do we enable stuff to be sold, like physical things to be sold? I guess the services piece, I was thinking like, oh, services from agents, like an agent needs a job done. And so people go and, yeah, pay for that.

Pedro Chaves [00:35:43]: One of the categories that we have within OLX, it's services. And by services, I mean plumbing, electricians, painters, whatever. So you can actually buy those services, right? You can actually find the people who are willing to do the service that you need for, for a fee, obviously.

Demetrios Brinkmann [00:35:58]: Yeah, but it's not like a newsletter agent.

Donné Stevenson [00:36:01]: I—

Demetrios Brinkmann [00:36:01]: my agent goes and contracts a newsletter agent Because I see these agents becoming so generalized that they don't need to contract specific agents to do specific things. Yeah, you have the agent that can do whatever it needs to do on the computer. Now, as soon as we get out of the digital world, then you need things done. But right now in the digital world, an agent isn't going to go to another agent and say, oh cool, you know how to write newsletters. Come, I'm going to contract your service.

Pedro Chaves [00:36:34]: Yeah, I'm with you. I'm super skeptical about that model. I don't think that we can make a business out of it because not just because these agents are becoming very generalist, as you say, but also because if they are not good enough and we could see that from Tencent talk yesterday, they can build their own tools. Yeah, they can build their own sub-agents. So why would you sub-hire something which is 150% commoditized, right? Which is building your own agent unless it's something ultra specific. Which if you are within the real— within the digital realm, probably you can replicate.

Demetrios Brinkmann [00:37:09]: I've seen something come up that maybe is along these lines where it was a harness marketplace, and so you could buy harness of different agents. And so maybe that's the closest thing. But again, this is— it's a really hard sell. Yeah, to be like, I'm gonna actually pay money for a harness.

Pedro Chaves [00:37:29]: Yeah, that's basically saying that your agency is not good enough. It doesn't have the tools to build its own tools, for example, which is more or less a given.

Demetrios Brinkmann [00:37:35]: Or you're not willing to put in the, like, it's not that hard to make it work, you know, like even a genius, maybe the most incredible harness ever is.

Pedro Chaves [00:37:43]: But yeah, then when we jump to the real world, as you were saying, then I guess we get either into the sci-fi realm. Well, not just, not really sci-fi, because again, Tencent investing heavily in robotics, the whole Chinese ecosystem investing heavily in robots as that last mile of interaction with the world. Or you get humans doing it.

Demetrios Brinkmann [00:38:03]: Yeah, which humans apparently are happy to do these days, judging by how many people have signed up for this website that you can hire a human for.

Pedro Chaves [00:38:14]: Yep, because someone will need to come in here and actually paint that smudge off the wall, right?

Demetrios Brinkmann [00:38:20]: Yeah, yeah, exactly. Or fix the toilet when it is broken.

Pedro Chaves [00:38:26]: What do you think about that? 'Cause everything that we have discussed in these past few days has been about selling stuff. You probably noticed that.

Demetrios Brinkmann [00:38:35]: On WhatsApp.

Pedro Chaves [00:38:35]: Selling stuff on WhatsApp. So is there room for any social or ethical discussion these days? It's a pretty weird feeling, right? 'Cause it seems like everyone is developing at a really, really, really high pace. But to push stuff for you to buy.

Demetrios Brinkmann [00:38:56]: Yeah, it is.

Pedro Chaves [00:38:57]: It's amazing.

Demetrios Brinkmann [00:38:57]: I thought about that at one point. It's very much capitalism at its best. I think it was when the Tencent talk was happening yesterday.

Pedro Chaves [00:39:04]: Yep, yep, yep, yep.

Demetrios Brinkmann [00:39:05]: It's like, wow, absolutely.

Pedro Chaves [00:39:06]: The whole ecosystem is based off of recommendations.

Demetrios Brinkmann [00:39:09]: Yeah, yeah.

Pedro Chaves [00:39:10]: Like the question about someone in the audience asked, but can we trust recommendations? How do we know if recommendations are actually good for me or they are just ad-based? And I think the answer was really fair. It's like, no, recommendations is at our core. It's embedded within the society. So every purchase that you do, you're actually incentivized to leave a review. So it's like it's next step capitalism where you end up like you as a human are embedded in the recommendation system.

Demetrios Brinkmann [00:39:38]: You're just living in a recommended system.

Pedro Chaves [00:39:40]: Exactly. And that will push the next bubble tea.

Demetrios Brinkmann [00:39:43]: Yeah, exactly. It's not a simulation. It is a recommendation.

Pedro Chaves [00:39:46]: Exactly. Exactly. And you are, and you are part of it.

Demetrios Brinkmann [00:39:48]: Yeah. Yeah, I do feel like it happens already. And so why not figure out ways to reduce the friction? Because like I was saying, I have, from being a human and living in the capitalistic society, a lot of stuff that I've accumulated, especially since I have two daughters. And as they grow, I just accumulate more and more toys. And so this brings us to that next idea that you were talking about, kind of adjacent to what we were saying in the beginning. If I could just take a photo of my living room right now and you could tell me, I've got buyers for those Legos over there, I've got buyers, they're willing to pay €20 for those horsies that you've got. You've got 50 horsies. Do you want to sell one or two of them? I'm happy.

Demetrios Brinkmann [00:40:41]: I don't think my daughters would ever realize they're gone, and I can make an extra €50 and free up the living room. No, that's, that's impossible. I would love that. Like, you can choose how extreme you want to go on the volume.

Pedro Chaves [00:40:55]: Yeah, the agent would estimate how much stuff you need to also—

Demetrios Brinkmann [00:40:58]: like the, what is it, the YOLO mode cursor? It's like, I want to go YOLO mode.

Pedro Chaves [00:41:05]: But, but that idea, what does it require? It requires this, again, this platform, this harness for your agent to be able to expose your stock in this case, let's say, to multiple other agents, which have maybe previously registered with some sort of interest. And then they would negotiate between themselves. You wouldn't know nothing about this. And at the end of the day, you would get just a notification to confirm the sale or something.

Demetrios Brinkmann [00:41:29]: That is different than how I was thinking about it too. I like that because they've— the agents register interest and they're there just saying like, hey, if you got anything—

Pedro Chaves [00:41:39]: As buyers do nowadays, right? Because when you know, when you go in a platform, what you are doing when you are clicking and navigating, you're registering interest. Yeah, you're building an intent, and we in the background are trying to extract that intent, right? But agents probably, or most likely, need it to be explicit, right? So I'm looking for this or this and that within this budget, as we were saying before, from the perspective of the buyer. But then you have a million of these guys. Yeah. And then your seller agent can just either broadcast or, I don't know the details, but, or directly go through some previously vetted buyers. I mean, the possibilities are endless here.

Demetrios Brinkmann [00:42:20]: If agents are acting, how can fraud be exposed on that level?

Pedro Chaves [00:42:28]: You have to beat that layer of trust. That's probably one of the best reasons to justify why we need these Middle layer. This middle layer, it's trust, right? It's an escrow. Yeah, it's trust. I don't know other ways of—

Demetrios Brinkmann [00:42:42]: You would in this case, right now with OLX, you're not taking any payments. But in this case, it would be—

Pedro Chaves [00:42:48]: There's a subset of categories and countries which have explored paying chip, but it's not a business model that it's going forward. Yeah.

Demetrios Brinkmann [00:42:57]: Yeah. And but in this case, it would be more hold your money Until you get your thing.

Pedro Chaves [00:43:03]: Maybe, maybe. Or we could use the same model that we have right now, which is VASes, Value Added Services, where you as a seller end up paying for extra services within the platform. So more ads, promoted ads, and so on, which could clearly work with agents, obviously, right? Your seller agent is limited in our platform. You can only— your agent can only post one or two things. So you can only sell one or two horses. But then we can copycat the model that we have and say, okay, you want to sell off your 25 horses? Just— it's a—

Demetrios Brinkmann [00:43:34]: but when I go back to the friction, one friction point that I had when I was buying the roller on OLX is I had to get cash. I never operate with cash these days, which is, yeah, for better or worse. And so then I had to drive to the person's house when they had their window of openness, right? We had to figure out when that was, and I had to come with cash. So at first, before driving to their house, I had to go to an ATM, pick up cash. All of that can be erased in—

Pedro Chaves [00:44:07]: although that could be erased, um, I don't see why not. If your agent has access to your Apple Wallet or Google Pay or whatever, the transaction— and then the transaction can happen immediately. And again, we have at least two protocols now, one from Google and another one from OpenAI, right, for commerce protocols to try and unlock this, these verified transactions between agents. So again, I think the tech is there. Someone just needs to occupy that space.

Demetrios Brinkmann [00:44:37]: But going back to this locker idea, I'm not sure I fully understood what the value of the locker would be or how the agents could interact with the lockers.

Pedro Chaves [00:44:49]: I was thinking about your idea of, or your problem, which was you had to figure out like the physical connection with the seller in this case. And what I think these Amazon lockers, these Vinted lockers end up doing is they basically do a very nice analytical work and try to discover what are the best hubs. Right? And then they have their lockers there and then these lockers are connected to the courier companies. So you can send stuff to a precise locker within a certain space, which in principle is relatively close to, like it's in an optimal location, right? So you as a buyer can probably rely on one of these lockers to be sufficiently close to you at any time. And whatever you want would just be delivered to you there instead of you having to negotiate the location with the, with the actual seller, which is also a risk like this physical contact. As much as it's also a good thing about the human nature and interaction, it's also a risk. That's why police stations actually have locations for these transactions to happen. That's why there are a lot of recommendations on how to do this transaction.

Pedro Chaves [00:46:07]: But still, maybe there's some attrition there that could be reduced because ideally at any time you would be sufficiently close to any of these smart locker hubs to place your order whether you're here or in Lisbon or whatever. Well, not for you to place your order, for your agent to place your order.

Demetrios Brinkmann [00:46:24]: Yeah, for me to just take a photo and say get rid of all these damn toys.

Pedro Chaves [00:46:29]: And then you have a notification, hey, your stuff is in locker C, here's the PIN code.

Demetrios Brinkmann [00:46:33]: Yeah, or in this case, if I'm selling it, I need to actually put all of the stuff into a bag and then—

Pedro Chaves [00:46:40]: or have the courier pick it up. I mean, eventually it could be like a drone, like in San Francisco, like those automated delivery bots that go on delivering food in San Francisco.

Demetrios Brinkmann [00:46:51]: Also a drone, yeah, in my area of expertise.

Pedro Chaves [00:46:55]: But yeah, and I think iFood is playing with drones as well in Brazil.

Demetrios Brinkmann [00:46:59]: Because there it does hit on the speed. You want things faster.

Pedro Chaves [00:47:05]: Yep. Yeah.

Demetrios Brinkmann [00:47:05]: And frictionless experiences. So if you're able to get it to that locker as soon as the agent says, yeah, you want it? All right, cool. Boom. And you figure out a way to get it.

Pedro Chaves [00:47:16]: The wheels are in place.

Demetrios Brinkmann [00:47:17]: Hmm. This does— it does feel like we're quite a few steps away from that world, but it's not that many steps.

Pedro Chaves [00:47:26]: It is not that many steps.

Demetrios Brinkmann [00:47:27]: You could envision a world where agents are acting on your behalf in the marketplace, and it's as easy as you taking a photo, or it's as easy as you saying, I want this, get me a plumber.

Pedro Chaves [00:47:43]: Yeah, or give me a horsey if you're on the other side of it. Yeah, and the equation—

Demetrios Brinkmann [00:47:48]: and then a drone comes to your door and you just give it the bag of horsies. And it goes through a locker. And I don't know if the locker idea and the drone idea is the most optimal, but it feels like it would be the easiest first step. Like it would be the path of least resistance.

Pedro Chaves [00:48:13]: Maybe do that. I mean, and there's a marketplace. Maybe that's not our worry, right? Because that's last mile logistics. It's a whole different space. Maybe we just connect to existing solutions. Okay, so in this region, the agent can rely on smart lockers. Cool. In this region, the agent can rely on ground autonomous robots.

Pedro Chaves [00:48:33]: Okay, let's outsource it. Okay, in this region, they can actually call for a drone to come here. But I would say that that's not really, I mean, it's part of the scenario. Part of the whole equation, but not part of the problem that we should be solving as Olax.

Demetrios Brinkmann [00:48:53]: Yeah.

Pedro Chaves [00:48:53]: Or as browsers.

Demetrios Brinkmann [00:48:56]: You're mainly focused on how can we get the agents to interact in them in this scenario, in that middle layer. How do we enable the supply and demand?

Pedro Chaves [00:49:06]: Yeah, exactly.

Demetrios Brinkmann [00:49:07]: To happen, because right now the agents aren't able to have that marketplace.

Pedro Chaves [00:49:12]: Yep, exactly. How do we unlock this item circulation?

Demetrios Brinkmann [00:49:18]: And I had been thinking of it before as I'm going to ask either Claude or Gemini directly for something, kind of like we do right now on Google. I want XYZ, so I search on Google and then boom, I look through things and great, I've got it. I hadn't necessarily thought of it the way that this is where, and I guess you could still interact through Claude or Gemini.

Pedro Chaves [00:49:47]: Why not? As long as it's connected to the best marketplace.

Demetrios Brinkmann [00:49:52]: Yeah, exactly. You are making a bet that Claude or Gemini is going to use you as the marketplace layer. And you would have to be that marketplace layer because you would have the most supply, I guess, or the most seamless experience.

Pedro Chaves [00:50:10]: I mean, I guess there are multiple angles here, either the now traditional, still modern GEO, right? Generative Engine Optimization, like making sure that you pop up in these results, making sure that the interaction between these agents and your platform is seamless. You could go the integration way and have like Shopify, Veev integrate their, their role ecosystem within ChatGPT, you should become as— you should become so good and so large that they want to integrate themselves. Yeah, I mean, I guess there are multiple options here, but what we— what I think we can see is that there's like the demand for this is clearly here, right? People want to buy, are starting to want to buy stuff in there. In their, or through their agent, through their personal assistant.

Demetrios Brinkmann [00:51:03]: Yeah, through the chat, which we were discussing yesterday, might not be like the best interaction layer, but, but it's here and it's what we're dealing with right now. And so can we make it better?