Stablecoins, AI Agents, and the Future of On-chain Apps | Jeff Wentworth at ETHGlobal Tokyo 2026

Good day everyone. Nice to see you. Some faces are familiar, but of course there are many new ones. This is a session about stablecoins, AI stablecoins, AI agents, and the future of on-chain applications. This is a workshop from Curvegrid.

I'm Jeff. I am one of the co-founders of Curvegrid. We are a 9-year-old blockchain technology company based here in Tokyo, Japan. You may have seen us before at various ETHGlobal events. This is actually ETHGlobal Tokyo 2023.

Three years ago. This is our entire team today in red overalls. It's a little warmer here. I just left the jackets inside. But there are a lot of us here during the weekend.

Please come say hello, we would be happy if you would build with us. Three years ago, we brought a lot more people to the event, partly because building apps was much harder. People needed much more help and engineering support engineering support . I know I was a mentor at ETHGlobal in New York last summer, as well as many other ETHGlobal hackathons in the past. Lots of complex questions about Solidity, node dependencies, how to set up a specific smart specific smart contract?

This time I had zero technical issues with AI, right? Is there anyone here who doesn't use AI for programming? Congratulations . Congratulations to everyone else who uses AI. Of course we use it, right?

The whole world has changed in the last year, since AI agents became very, very good. So this time there are a little fewer of us. First of all, I want to tell you a little about Curvegrid and our journey. And then I'll talk about our thoughts on this hackathon and what we hope some of you can create. So, first I'll just open our website.

So, curvegrid.com—we do three different things. We have a blockchain development platform called Multibaas, and it's been in production for probably seven years. I'm sure some of you have already built on it or worked with it before. We’ve had clients ranging from large Fortune 100 companies in Japan to global Web3 global Web3 projects build on our technology and development platform to bring their blockchain applications to market faster, easier, and cheaper than if they had built everything from scratch.

So, this is the first thing we offer—MultiBaas. The second thing we do The second thing we do is, in fact, consulting services. We help companies write smart contracts or build parts of their blockchain their blockchain applications, typically using MultiBaas. I should add, I was showing you a small screenshot of what MultiBaas looks like. I'll come back to this in a few minutes, but this is the MultiBaas dashboard that is commonly used by blockchain developers and programmers.

The third and programmers. The third thing we're doing, and this is much newer, is we've taken the experience of our blockchain our blockchain platform, how we and our clients have used it to manage corporate treasury of stablecoins and digital assets, and are now building a specialized platform to help companies and individuals do this more efficiently. This is a treasury management platform , primarily corporate, for digital assets. I'm going to show you its public preview for the first time. It's called Neo.

It will be released in the form of a minimum viable product, or MVP, very soon, probably next month. This is a new product we are working on. So, we have two products: MultiBaas and our new product for working with digital assets. This is a little information about CurveGrid. OK.

Now about the ETHGlobal hackathon in Tokyo. Therefore. First of all, I'll tell you a little bit...We need a topic. We need a theme to make it interesting.

Our theme for the demo app is a matsuri, or Japanese festival. First of all, raise your hands. How many people believe this is a real photo, and how many don't? Is anyone against artificial intelligence? A real photo?

Okay, about half. AI photo? Who is not sure? Okay, about equally. This is actually a real photo.

But I thought it would be quicker to just search, and by the way, I took this picture myself took this picture myself . So, it's faster to find a photo that I had in Google Photos. A Japanese festival is a matsuri. Some of you may have been on one of these. If not been on one of these.

If not , and tomorrow there will be one, maybe after the hackathon is over, sorry, not tomorrow, on Sunday. After the hackathon is over, you might want to go take a look. So go take a look. So , the initial idea we have, and for which we have an example repository, is to use stablecoins to buy vouchers that can be exchanged for food can be exchanged for food , drinks, games, and entertainment at a matsuri or Japanese festival Japanese festival . And we want you, given this example of an app with stablecoin payments and vouchers, to understand how to turn this into an agent workflow?

And that's where we want to help you all achieve that over the weekend. So we have this example application, this repository that you can point your agents to. Codex, Claude, Grok, or, you know, any agent of your choice. There is a full set of smart full set of smart contracts for the stablecoin, Matracity stablecoin, and voucher. There is also a full frontend kit—a React frontend kit—a React application.

When you run it, it will look something like this. Is it true something like this. Is it true ? This frontend uses Multi-Baas, our blockchain development platform, to interact with the blockchain, create and send transactions, and read data from the blockchain. Once you get all this up and running, and we've got it running on the Mizuho Key Awaji testnet, you can use Sepolia or your local development network or testnet of choice.

By the way, about Mizuho Key: it is a Japanese blockchain created for the Japanese market. It is based on Ethereum with a distributed Proof of Staked Authority mechanism, in the creation of which our company Curvegrid played an important role. It is supported by a number of large Japanese companies. So, here's this sample application deployed on their test network. I have already released several stablecoins to my test address here.

I will now follow the standard approval procedure for the voucher. So now I'm going to buy this voucher. So, I approve of buying stablecoins, and once that passes, I'm going to...Did it pass? I'm buying a voucher.

Is this happening? Does this work? That's all. You see, I purchased a voucher. And if I come down here, that's it .

I see that the vouchers have been transferred. In general, stablecoin balances are present there. So, these balances and actions are provided by the Multibaas chain indexer, and transactions are made through Multibaas. This is our kind of end-to-end matsuri, or Japanese festival. Um, get MetaMask out of there .

Uh, sample program. OK. So, this repository So, this repository , this example program gives you a working model. But we will reward what you create, not what stack you build on. This is a little different for us, because we usually require the use of our technology.

But that's not the case now. So, there are different paths that we would be happy if you would take, right? You can fork this repository, change contracts or the interface. You can just borrow this template, right? You can take this template and create something different as inspiration, or start from scratch and do whatever you want.

Therefore, we have no mandatory dependency or requirement that you build on Multibaas or use our technology to qualify for our affiliate prizes this time. So, I'm going to step back for a second for those of you who are maybe less tech- savvy, or just to set the stage, because I think it's important to visualize this, right? Typically, you have, at least in this repository example repository example , a user interacting with the interface through their wallet, which sends a transaction to the smart a transaction to the smart contract, right? This is the simplest type of template imaginable. Ahem.

Then, with stablecoins, right? You actually buy this NFT voucher and then the seller converts it later. So I want you to imagine, we're so focused in hackathons on the question, "How do I do this technical thing?" But I think it's much more interesting to think, " interesting to think, " Okay, what exactly are we trying to achieve?" trying to achieve?"

» Isn't that right? And a few ideas on what exactly can be changed, right? Instead of an NFT right? Instead of an NFT voucher, you could make some kind of trading card system. It would be possible to create a multi- to create a multi- master or festival system.

Obviously, there are many different options for what can be done here. And I think an interesting next step that we'd like to see from teams this weekend is replacing the person pushing the buttons with an agent. Yes? This whole user interface that we're used to in the past, where there are buttons for actions— where there are buttons for actions— like "approve expense," "buy voucher," or " voucher," or " activate voucher." This is good, but many of us who have been experimenting with agent technologies for the past few months or a little over a year realize that there are other, very interesting interface paradigms interface paradigms .

Yes? What might this look like? This could be a chat interface. Maybe you'll just talk to an agent. This could be something completely different be something completely different than the usual buttons in a web interface.

In fact, it might not even be "I want to buy a voucher." It could be something like, "I have friends at the festival, we want to go out for dinner." "What should we do?" I'll come back to that in a second. Another thing we really want to see really want to see is...

by the way, the sunglasses emoji here is a reference to "Top Gun," if any of you have heard of the " of the " danger zone." But it's real, right? We've all been in situations where AI has done scary things, right? The first thing we recommend is not to store private keys in agents or give them access to them. This is exactly what we will pay attention to when evaluating and awarding prizes.

Second—don't let prompts become rules. Truth? The idea is that the prompts given to agents are non-deterministic, and new popular technologies have emerged , such as Jev; I just spoke to a hacker who is planning to integrate Jev this weekend. But don't assume that what you write in the markup file you provide to the agent will necessarily be the truth. And the last point—don't forget about human responsibility.

So when we think about our new product, right? This treasury management product that we are building for network finance requires the mandatory participation of people who can take on some of this responsibility. You know, for example, right? The user creates an intent, the AI creates an intent, the AI agent processes it, but on the other side is the policy mechanism, another stage of human approval, a secure signer, and then the blockchain itself. So, as you can see in the gray squares on the right, we have clearly separated the private key and ownership of it, and transaction confirmation, from giving the agent access to that.

However, this is a hackathon. Ideally, you should not give agents keys. You shouldn't give an agent unlimited control over anything. Of course, there will be situations or, let's say, challenges during this weekend when you provide private keys or hints to an agent at a hackathon, which is actually a policy. That's fine, but it's not exactly what we mean.

The main idea, if we think about how agents should manage money, is to try to separate these things. So, here are the three prize categories we offer. The first we offer. The first is the tokenization of real assets (RWA). That is, do something related to programmable assets.

The second is a dashboard for digital assets. I use the term " assets. I use the term " dashboard" very loosely, as we are all used to looking at charts, tables, graphs, etc. But what does a dashboard look like in the age of agents? Maybe it's not even a traditional dashboard in the usual sense.

Maybe this is something I can ask about and have a dialogue with . And the third category is the AI ​​project in general. How deeply can you integrate AI into digital assets, stablecoins, and programmable money? So, using the multi-base functions of our platform is optional this time. We would be happy if you used them, but it is not a mandatory requirement to receive prizes from our partners.

So our partners. So , again, this is the example I briefly mentioned earlier. For example, maybe your entire application is connected to Telegram, WhatsApp, a to Telegram, WhatsApp, a chatbot, or, excuse me, a text chat window on your website. And instead of an interface where you have to press buttons, you just say, " just say, " Look, I have 3,000 yen. Find a lunch for four people and buy vouchers."

Or maybe vouchers." Or maybe don't even buy vouchers. I have 3,000 yen. Just buy lunch. Or where should I have lunch today?

Oh, there's Matsuya nearby. Let me, here's my suggestion: why don't I buy these vouchers so you can get these meals, well, this delicious food, for example. So, it's starting to look something like this, right something like this, right ? When the LLM or agent runtime is actually at the top of the stack, rather than integrated somewhere at the bottom of it. And again, this is something we all probably face in our daily work: how, where, and when we integrate agents into our workflow, and how do we separate what is how do we separate what is "vibe coded" from what has been carefully designed.

But I think that's probably a pretty good way to differentiate between the two things. Isn't that right? Everything related to key storage, authorization, authentication, approvals, and policies, as well as anything related to recovery, audit logs, or evidence of what happened. Everything else, I think, is perfectly suitable for "vibe suitable for "vibe coding." So, that's all.

Our little slogan: Ethereum provides value, AI creates intent, the missing layer decides whether money should move, and it is this missing layer that we hope you will build. This is our repository of examples. If you go to github.com/curvegrid, it should be one of the first repositories listed. We also have a QR code on our booth that you can scan to go to the link, as well as our prizes, which are on the sponsor page on ETHGlobal.

ETHGlobal control panel. And that's actually all. Thank you. I think I have a few more minutes. Have a question?

None? Good. Perfectly. Thank you. Have a nice weekend everyone.