A2A x UCP: Blockchain's role in Agentic Commerce | Ling Zhong - Google

So after listening to a lot about hyperlquid which I love now it's time that we move towards blockchain's role in ejectic commerce. We're all hearing about open cloud. We're hearing about cloud codeex. Do I have any AI builders in the audience today? Any wipe coders?

Wow, a ton. That's amazing. So today we have Lynn with us who's going to talk more about how blockchain's role in in aentic commerce works. And I'm genuinely excited because he has some really really cool things to share. So please give him a huge round of applause.

And Ling, the floor is all yours. Hi everyone. Testing. Okay. Awesome.

And uh let's get started. Hey folks, I'm Ling. I'm part of the community. I've been building on Ethereum since 2018. So, it's very exciting for me to be here speaking on the stage for the first time.

Um, without further ado, today I'm just going to talk about, hey, like we heard about agentic commerce, agent payments. Um, my my key message here is I think blockchain has a lot more to do than just payments in this entire genic commerce. So, uh, let me walk through that. Um, very briefly, let's just, you know, set the stage by having a purely product perspective. what is a shopping experience look like?

And I'm going to talk a little bit about the technology involved to make that happen. And of course, we will discuss some challenges and opportunities ahead on what could be possible this year and going forward. So, let's see if the internet god is kind. If not, this is a toy example that I'm going to voice through. Okay, great.

So, you know, I've been walking around the booth and I love some of the swag here at East Denver and I saw a really, really cool Buffy Corn, but it's sold out. Maybe it's online somewhere. Uh, I want a shiny one. So, I'm just going to ask my agent. It could be, you know, any ln.

It could be a lobster for I care. Uh, what's it doing? It's searching through trying to find the right merchants. And, you know, it's doing all of that fairly quickly. And he found me an order.

Okay. 100 bucks. That's all right. for a shiny baffle corn. I'll do it.

Um, but hey, like I'm here. Uh, do I get anything special for being an attendee of ETH Denver? I actually paste in my Pope link just to prove that I've been here. Um, the agent is working through some stuff, verifying discounts. He's like, "Okay, great.

I'm eligible." And it's applying. Great. Nice. 20% off.

It's 80 bucks for my shiny Buffy Corn. And I am going to buy this. Okay. Okay. Great.

Only if life is that easy, right? Okay. So, let's talk about what happened here. There are so this is the happy path and this is a toy example. I believe that there are a lot of work for us to do here.

Um let's so few things that happened. We first need to find merchants and we actually need to find the right merchant for us right um and then I want to feel a little bit special get personalized offer offering whatever information I'm willing to share and of course pay the merchant I think in all of these steps uh blockchain has something to help here and it looks like this um yeah this is how you know blockchain's involved uh it's it's quite complicated actually it's not that case So we're not going to go through this bit by bit. Um talk for today is just focus a little bit on the first phase. How do you find the right merchant? Notice the uh capitalize right.

And okay, this is a zoomed in version. We have a lobster at the start. And you notice that we have a lot of acronyms. So just to level set, I'm going to walk through what these acronyms could mean. And uh here uh so ERC 804 um the this ERC is called trustless agents.

But the way I really view it is a permissionless yellow page for agents. If you want some work to be done uh if you want to purchase certain items, this is a place where they have things like the uh permissionless registry for agents. Imagine a place that you can just stick your uh business card up there. And also there's a reputation system and is also permissionless. In fact, people can set up their own reputation system.

We will discuss a little bit. Do you care whether something returns very quickly? Do you want something that's always have a discount available? Do you care about the uptime? So those are different dimension reputation that people can play around with.

Moving forward, X42 perhaps this is the one that crypto community is the most familiar with. Um this is the X42 uh extension of the HTTP and it's payment required. You want to access a certain resource but hey you have to pay for this. Um so this is the internet native way a standard to pay and of course uh being internet native uh crypto is actually a first class citizen on this protocol. So going into A2A uh agent to agents um you can think of it as just a way a shape of connector of connecting different types of logic.

In this case just the different types of logic happens to be generative AI logic. So it's not just about chatting with the robots here. Uh it's all about defining tasks and delegating them. and A2A help us do you really useful things like okay I give you a task it has a long life cycle and I actually don't care if you delegate the test further into subsequent things so this is really one of the ways to delegate coordinate and communicate tasks among agents now the newest acronym is UCP universal commerce protocol uh we can buy something as a single thing but if we think about the entire shopping journey, we need to find the merchants. Uh we'll try to vet the merchants and even if they have the right reputation, do they have the products?

So product discovery is there and sometimes you may want to pay for more than one thing. So UCP really structures that whole flow from finding the right shop, the right item, putting them in the shopping carts, how do you manage and communicate that shopping carts across different sessions and also how does it work with fulfillment, shipping, yada yada. So the the whole complexity of that graph that we'll go back to really um here is uh this is a structured way of looking at what is a shopping experience. So with these acronyms sort of level sets uh I want to give a bit of disclaimer. Um so I am working at Google and Google is a contributor to some of these open source software.

Uh but rest assured uh it's a open ecosystem. You don't have to use these things. Uh and and I think this is just a way to inspire one of the ways things can be done. And as we talk more about the pros and cons we can think about hey how can things become better. Now we have the building blocks.

Uh let's take a look at this graph a little bit. So we're really doing the first and second phase. We're just discovering, right? Uh you set up your lobster, you ask your lobster, buy me, you know, a bunch of corn from East Denver. They're like, okay, I have no idea where to look for it.

Uh but there is a permissionless registry of agents. So what is it doing? it is uh trying to get all of that data and then it is checking through the reputation of each one and then okay uh it found some qualified agents and they say okay let's actually have a whole conversation about shopping let's talk about are you able to pay this way uh what information do I need to give you to get a discount yada yada yada so uh really just all these players interact this way and let's talk about how that actually look like in code um these acronyms are actually not that scary. They are just JSON files. Um so let's look at the first layer.

This is the X uh this is the ERC 804 agent registry and this is one example uh that's uh abbreviated for agent registration. I highlighted a few important parts. Uh first it has a very well-formed schema and it has an endpoint. It's saying that hey uh this specific entity can be interacted using A2A. It's actually not exclusive to A2A.

You can talk to it through NCP. In fact, you can define any type of interfaces that you want. That's up to you to decide whenever you're registering the agent. And of course, it's like uh everyone can have their own agent registry. And in this case, we are saying this is the registry I want to register my agent under.

Now, okay. So, as the lobster AI, uh, I'm looking at a registry, but I'm seeing thousands of these. How do I weed through? Um, I want to show how one of the feedback may look like after a certain purchase. So, looking at the second block here, it says the agent feedback.

Again, we point at a specific agent registry for a specific agent, and I say, "Hey, uh, I'm writing a review on this date on a certain interaction." Um, at the bottom here's my proof. So, I'm actually a verified purchaser. So, my review actually means a little bit more. And the things I want to rate on is okay.

Uh, the stars experience is pretty good and my success rate interacting with you. There's no downtime. So, you can imagine thousands and maybe even more of this type of feedback uh being registered and they're all accessible permissionlessly. And that's very useful for anyone here to figure out which agent is actually going to be the correct one for your specific shopping journey. Now uh looking at the first block again there is something that I highlighted.

It's called the agent card. Uh this is one of the reference. Let's assume that my agent wants to talk using A2A. So okay inside the agent registration it points to the agent card. And the third block here that's actually what a one example of agent card looks like.

Uh so it says that hey this is an agent that interacts with the merchant catalog. Uh the name of it it's a Buffacorn bazaar agent. Okay, great. Seems like someone who can sell all types of buttercorn. And the thing I really want to point out is at the bottom.

Um, it of course can have many skills and many tools and things like that. Uh, but what it one of the extension it is using is using UCP and how that matters if you look at the fourth code block is that inside this code block where it's really defining the handshakes that we want in this entire shopping journey. What does checkout look like? Are we able to offer discount? If there is a discount, look further to another JSON which is well defined schema here.

And of course at the bottom we also talk about uh does this agent offer certain ways to pay it. In this case it does offer the X42 extension. So after many many steps we get out of corn success. Yay. Um, but let's let's think a little bit, right?

That's that's a very very hefty flow that we have to walk through. Does this really work when we have millions or even billions of agents sift through, right? Like uh maybe for a toy demo, we have 10 of them. We throw them in the context. Maybe the agents able to figure it out, but we probably need to delegate some of this logic elsewhere in a different way.

We know where the agents are. We know the ratings are but the algorithm there is no fixed structure right now that anyone's proposing on how do you rate it. So it might be a good thing might be a bad thing and I think this is an area for people to think about and innovate here. Um the second one is how to incentivize feedback. Uh I would like to invite people when when was the last time you went to a good restaurant and you left the feedback.

probably the right way to do it is not that often and especially you had to go through all of the crypto UX beforehand to do this. So uh we should think about different mechanisms to incentivize feedback but also we need to make sure once these incentives are in place we don't have abuse where we also had to combat malicious feedback as well and then I think the third point is really really interesting is if you look at the flow it's really structured in the same way that how humans shop is that really the right assumption how different will agent shop uh compared to humans right so I think this is why uh across the entire stack here. I I believe that the blockchain is more than just a payment rail for agentic commerce. It has a lot to do with identity reputations. It has a lot to do with just giving feedback and how do you handhold what parts do you want verified?

And I think the challenge here is um with any inclusion of blockchain technology, there's always a little bit of tax going on. How do we balance the value that we get from adding this trustless characteristics but also we make sure that the shopping experience is not burdensome for both humans and perhaps increasingly important not burdensome for agents. So um I encourage everyone to read the specs, read the code that's the source of truth and yeah so uh we would love to learn your thoughts. Please start a discussion on GitHub, create issues. If you have different ideas on how the flow should be, this is a prime time to innovate there.

Um, and yeah, my DMs are open and welcome for the reach out. Thank you very much.