ERC-8004 Launch Day #1

Transcript

Davide Crapis: Mid school.

vitto: Marco, can you hear us?

Marco De Rossi: Let's start.

vitto: Let's do it! I will start with a quick introduction to this ultimate 8004 launch day, and then I will introduce all of our guests, the agenda, the event should last for 2 hours, we're probably gonna run a little bit late, so make sure that your calendars are freed up, because there is a couple of very nice presentations today. So, starting with Marco DeRossi from MetaMask AI Lead, that is going to give us an introduction on all what 8004 is and what we've been doing for the last month and a half during the Genesis month. Then we'll have the folks from Ola Spirituals and Reston. Presenting us what they've been building in the last few months, not, of course, only the last month.

Then two keynotes, one from Eric Rappel, Coinbase Developer Platform Head of Engineering, one from Nenad Tomashev, Google DeepMind Senior Staff Research Scientist, and then a few more presentations from the folks at Zifi, 2004 Scan, Bond.creddy, Agent Zero Daydreams. And to conclude, of course, we'll have our fantastic David Akrapis giving us remarks and final observation on the Genesis month. I think we will start with Marco. Marco, the floor is yours. You're muted.

Marco De Rossi: Now it should work, sorry guys. So, exactly 6 months ago, on the same exact pink flight notes, we took the save on purpose. We had our first community call, so it was, like, mid-September. We were, like, a group of excited people, active in agent economies that have just drafted a standard that was getting viral, and we said, why not, like, meeting on a Zoom call altogether to discuss the next steps? That was just 6 months ago. And now we are, like, discussing what has become the most popular distributed agent registry globally. So it was really a long run. We have come a long way together, and I really want to celebrate this Looking at the past and see what happened in the last, again, just 6 months.

So in August, we drafted the standard, we proposed it. In September, I think what really happened was the formation of the community, and also the first community call. So it basically stopped being a project of, like, a couple of folks, and became something very, like, with a very sharp vision. In October, one month after, we came up with the final version of the protocol, almost final. After having discussed with hundreds of builders, in chats, in calls. And in November, we've finally met in person for the very first time in Argentina. It was a long travel for many of you, from many countries, and we had our Trustless Agents Day event. We spent December more and so on building and directly testing on Testnet in 2004, it was on Ethereum Sepolia, and we learned a lot, and we changed, actually, important parts of the protocol to save reality, which is fine, because that's what testnets are for. Then, in January, we launched to mainnet.

And we, we, at the end of January, we started our Genesis month. And what happened in February was pretty surprisingly, because not just we had, like, a lot of activity on the protocol, I have to say, well, personally, at least a couple of order of magnitudes more of what I was expecting, also because of, I would say, like, people minting agent ID to get the lowest number agent ID possible for the future. Which is, like, not incredibly valuable, but it's still a sign of success, in a sense. And even more than that. almost, like, all the most relevant blockchains that you can think of, more than 20 blockchains, decided to… to adopt in 2004, and this happened,

with them. So, the different organizations, the different communities, worked on this. We will see now which are the blockchains involved, and this definitely became a cross-chain effort. pretty successful. And now, in March, what we can say in terms of numbers and adoption is that ERC 2004 is currently the world-leading distributed agent register in terms of activity and volumes. So, these are the supported chains, the EVM-supported chains. Thank you, thank you, thank you to each team of this chains. We didn't just deploy the contract, you helped us. in communicating the protocol, in engaging with developers, you onboarded your DevRel teams. So it was really a shared effort, and actually, this is like an EVM slide, but we also have the first not-EVM chain that adopted 8004, which is Solana, of course, that launched its own 8004 implementation around 10 days ago.

These are just, part of the projects that are involved. The projects are many. I also, like, want to take this opportunity to thank the organizations that are supporting all the different, ESC8004 author, so in my case, MetaMask and Consensus, because I've always been, like, strong supporters of initiatives. And, how did we pull it off? So, if I had to say what, like, worked, is exactly that this hasn't been created in a lab by people that then went out and said, this is the protocol. It was really, like, a shared effort. All the companies that you have seen in the previous slides helped, or with a spec, or the writing or writing PRs to the smart contacts.

or the audits. We had 4 independent audits that we haven't paid. Thank you for, like, all the auditing companies that decided to contribute. We have many companies and groups that help with tooling and infrastructure, so today we will also, like, meet some of the SDKs and explorers. If you visit at 2004.org, you will see many of these tooling infrastructure, where they're proving to be essential to have the community growing, and of course, many prototypes. So this made us stronger, not just because, I mean, more braids are better in spotting things to improve and fix stuff, but also because this kind of, how can I say, sense of ownership turned all us, in some sense, in, like, championing 8004.

So, we have come a long way, that's for sure, but we have a, like, many things that need to be done. There is a lot of work to be done, and that's what I want to do together with, of course, celebrate, which is what we will do today With Authors of 2004, and we have many projects. So, what is still missing? Well, the reality is that agent economies are very early. The volumes in terms of, like, payments, revenues, are… close to zero. We are very early. We are in a super new space, and it will take years to take off. And the challenges that we face are actually not ERC8004 specific or broader. And if I had, like, to recap in three short sentences what's missing, which are, like, the key driver for the growth of this ecosystem.

I would say these three. So the three biggest challenges… challenges are, first. Do we have useful agents? How much of the agents that we have are actually useful for final users, for companies, for consumers? The second, are they exploitable? As far, agents will be exploitable, they will never be actually autonomous, because nobody will give money to agents If it's super easy to exploit them. So, agent security is a clear blocker for the existence in the first place of agent economies. And third and last, do we have good data on these agents? When it comes to, security, to the second point, just to jump into the detail of what we are talking about, there are, I think, a few crucial areas. So, that are actually, emerging more and more with what happened at the end of January with MoldBook, being launched and with 2004.

spoofed agents, so agents pretending to be other agents. A second area is counterpart skills and capabilities. So, right now, we trust the skills listed in the metadata of the agent card to trust what the agent can do. And it's clearly not enough, right? It's like trusting a human being that introduced himself, saying, I just won the Nobel Prize yesterday. It's not enough. Third, as an, like, an identity and reputation attack, we have collusive reputation, and as we have seen also in 2004, where, like, people are meeting just reputation signals to inflate the reputation of our own agent. Then, in terms of negotiation, we have negotiation drifts, frauds, oversharing of information that are, like, that are not needed to be shared during a negotiation. And prompting action, of course.

And solve any records misrepresentation, which means when we deal with an agent. How much can we trust that agent to give us money back, or be liable for its action? This is something, like, super important to be able to estimate in any economy. And of course, last but not least, the unsafe payment and delegation flows, which is something that, of course, is very close to what we do at MetaMask. How can we give Self-custody to agents, but in a very controlled way, and limited by the human will. And I think that a lot here will change, and the reason is that right now, agents are really pirating humans, and this is really impacting privacy, security a lot. The fact that agents always use free-form natural language is, of course.

a great driver of security concerns. It gives a lot of degree of freedom in communication, and many initiatives will rise to make the communication of agents just stop parroting us, stop imitating us. and be more efficient, more fast, leveraging more cryptography to be more secure. We are really just at the beginning of this. In terms of ERC 8004, and today we really want to share the roadmap for the next quarter, we will focus a lot on the fair point, so do we have good data on these agents? And we want to, like, to present today what's our, like, spring plan, what success looks like by the end of June. We really want to double down, at least on three projects that we're very, like, actively involved into already. The first is, now that we have the rates, so we have rates, we have

tens of thousands of agents registered on the identity registry. How can we make sure that top-quality, leading data providers publish everyday high-quality signal in the reputation registry? It could be about the revenue of the agent, the reliability, about the company and the operator behind the agent. Who is the company behind that agent? what's the revenue of the company behind the agent? Is that a listed company? What's the liability of that company? Who is the legal rep… so… A whole, like, bridging datasets between Anything that can be helpful in the real world to trust agents and the agents themselves. And an important part of that is using X402 payments as a reputation collateral. And we are actively working on an extension directly pluggable in the facilitator that automatically registers X402 agents on 8004 and use payments as reputation.

To bridge the gap between the bazaar discovery tool already existing and 8004. Second area is related to the validation registry that has been paused. Why it has been paused, why it's taking some time, for the same reason I was sharing at the beginning, to make us stronger. It's taking time because we're working with many people and listening to their opinion, making sure that they're aligned among them, which is not always easy, and the big shift will be Instead of, like, continuing, like, in the first design, to ask a validation for each task. we will validate for a given T that a given public key is safe to be used to communicate with that T. And it's like an address book of verified keys, and when you… when you have checked the address book of ATEs and you download that public key, then you can be safe in interacting with… in… in that way.

And so that's what we are cooking. And also, in the terms of validation, we will still have evaluator David at the end of this at this event today, we'll discuss more about an Agent e-commerce proposal that tackles this, not on the TE angle, but on a different angle. And finally, we are working, and we hope to be ready to share updates soon on incentives for high-leverage community projects to make sure that now that we have all the rail in place, we accelerate the shipment of, especially, like, infra projects that are used by all of you in this call. What's on you? So, as we often say, we want you to be a useful agent, not just infra. That's, I think, what we really need to move the needle, to get revenues, to show to the world out where, how much what we're building.

is meaningful. And, personally, what I care more about in… regarding this is really being able to, to talk to the Web2 agent builder space. So let's always remember that even if 8004 is, like, so crucial at the intersection of AI and crypto, it's probably, like, the main thing that happened in the last few months, we are still a super small sub-niche of a wider AI space. And, which is what I think we all belong to, so we… we need to show to… to the general… we need to give our tools and infra to Web2 builders, and vice versa, build agents that are meaningful also for them. And so this, I think, is the crucial direction in terms of usefulness of agents.

And at the end of the day, being part of a wider AI builders community. I will leave the floor to, Vito and presentations, and, looking forward to To follow the rest of the event.

vitto: Thank you, Marco. Yeah, just to answer everyone's question here, yes, this meeting is.

Ahmad A: Ahmed, Ahmed.

vitto: Don't worry, yeah, I will start muting people as soon as I hear them. But anyway, this meeting will be recorded, and the recording will be made available on the 8004 Builders Group on Telegram to everyone to watch these later on. Now, let's move on with the presentations. As a reminder, every presentation will last 7-8 minutes. We'll leave the QA towards the end, but better if you reach out later on on Telegram. So we can keep this tight, and I'll, like, make it a 3-hours event, which, of course, no one. We'll ask for. The first presentation is from David from Oles. Hey, David. Good day.

David Minarsch: Hey, thanks for having me. Should I kick it off?

vitto: Yes, sir, please, the floor is yours.

David Minarsch: Right. Can't see my screen.

vitto: Yes, sir.

David Minarsch: All right, let's get started. Thanks for having me. I will briefly introduce how, at OLAS, we are leveraging URC804, and talk a little bit about, trustless agent economies as a result, but the main… Focus will be on, obviously, the applications, and not too… what Marco just said, we need applications. So, First, when we, heard about the standard, we were very excited, because it kind of… reflected a lot of our thinking, we'll see in a second. We also, contributed a little bit. We're glad that, ultimately the, ERC804 are all ERC721 compatible NFTs. That certainly, I think. A sensible design choice, for managing on-chain agents. And that's also something we had used in our registry. So let me get to, what we had done before, and how we kind of bridged that with, 804.

So we started, doing an agent registry, as all us, on-chain in 2022. Perhaps one of the first, if not the first. And back then, what we did is we created three, registries, one, was there to, register code itself, or references to code. one for code and configuration, or references thereof. And then the final one was the service registry, as we called it, Aka AI Agent Registry. And so when ERC 804 popped up, we immediately saw that there was, like, this… connect there between the service registry and… that we had built, and that was already live, and the ERC 804 identity registry. And so one of the steps that we did in the early days, in integrating

804, and still shaping the spec was, kind of making sure that we can bridge into the system, just bring all OLAS agents that are already, on-chain into ERC 804. And the way we did that is sort of with a a bridge, I guess, where you have, like, all the OLAS agents, they're already, you know, many of them historically deployed, on-chain. And, we then created this, contract called Identity Register Bridger that effectively maps existing and future OLAS agents also into this ERC804 world. And so the nice thing then is that you effectively have two, on-chain representation. We'll see that in a moment. You know, one where, you know, our protocol features come in for sort of OLAS-specific capabilities, and then

And, this discovery and, like, sort of cross-ecosystem, breadth with the 8004. And obviously, downstream, maybe also these, new features that we're talking about, reputation and validation. So, before I show you all these applications, just for those who don't know about Ola so much. So, just earlier today, when I put the slides together, the total lifetime deployment of agents was 3,280. Today, we have around 663 daily active agents, which is, sort of a 7-day trailing average, so these are agents, that are active in the OLAS ecosystem, and they are now all ERC 804 agents as well. I'll show you some of those use cases in a second. You can also, see that we're on quite a few chains, and have actually done,

quite a large number of transactions, like, the agents have done those. So, in total, agents have done over… almost approaching 15 million transactions, and… a large share of them are agent-to-agent transactions, so that's where one agent effectively, like, you know, gets, procures the service from another, and vice versa. And that's all mediated by a marketplace, that's sort of all as native that we launched in 2023. So let me walk you, through some of those use cases to make it concrete, because that's also where, you know, the actual use becomes apparent. So, we have, effectively, two sides of the coin at OLAS, like, one is a very consumer-focused side, which is Pearl, and the other one is, agent or business-focused, side, which is the marketplace. Let me…

show you Perl. Here is my local Perl application. It's a self-custodial AI Agent App Store, that's how we think about it, I'll come back to it in a second. And the value prop is that there's different agentic use cases in there. They are specialist agents, so it's not like your, you know, Claude Code, which I also have open down here. No, it's a specialist agent that does something specific. So, for instance, here, I have my, prediction market agent, Omenstrat, that trades on Omen Markets on Gnosis. Polystrat, which trades on PolyMarket. on Polygon, and so on. So they don't then start doing other things, like an open claw, where you then can talk to it and make it do other things. That's not the case, but they will, autonomously trade in prediction markets. And so here you, for instance, can see, sort of my Omenstrat.

agent performing, over the recent, months. And, coming to the ERC, 804 part, these agents you can find on-chain. So this is my agent, LoslyTeldu09. These are sort of, auto-generated names, in our, OLAS was, and, I can go, basically, and find this, now, in, for instance, 804Scan. One of those, nice things about the standard, right, there's lots of tooling, we like this very much, thanks for whoever's building it. And I can just kind of type this in, did this earlier, and so here's my agent. So this is on the, you know, consumer side, so it is an agent. that, doesn't offer its own services, it, in this case, it consumes other agent services. We can also see it in our OLAS marketplace.

And, you know, where we backlink to the 804 scan. And, the reason why I want to show you this is because in a moment, we'll look at the other side of the coin, the marketplace. But basically, this agent… If you go through the work, you can, you know, find that it's got a lot of on-chain transaction, its agent wallet is linked here, and they obviously relate to this, you know, performance that we're seeing here. You know, what… what does this give you right now? I don't know if it gives you anything right now, but our consumers like it because, they, you know, they can see, they can track. their agent, in those scanners, and can kind of more easily get an understanding of how their agent interacts with other ecosystems and so on. And so, specifically, the,

Omenstrat agent, as I mentioned before, is a agent that participates in prediction markets, and so I want to quickly show you, how that conceptually looks like. So, we call it also a trader agent, that's the technical description. It trades in prediction markets, and in order to do so, it needs, you know, predictions, right? And, obviously, if you run local models, you can get some way, but ultimately, a lot of these, Prediction tools that are being deployed here. you know, need a bit more beefy compute, or need a specific data pipeline, or a specific closed model, right? And so this is where the supply side comes in, the marketplace. So let me get back to my slides here quickly to kind of situate this. We have on the consumer side Pearl. On the,

business or agent side, the marketplace, and specifically here we have a supply side where agents offer services to other agents that request those. So, you know, Perl being an example, where you have the agents in Perl that request services from supply-side agents. And so we can look, at that as well, so, I obviously also prepared those tabs, but… One of those agents that, offers these services is here, on our marketplace, and specifically, I've looked at the ERC 804 representation thereof, so you could also find this, on the 804 scanner. But the point being is they advertise, you know, services, which we can see here in the metadata, which should be open in one of my many tabs. Here we go.

Here we see one of those tools, Super Forecasters, which I can then correlate to another page that we have. Superforecast is basically one of many prediction tools that these service provider agents offer to these poll agents to predict the future, right? And we can see that this particular tool is actually doing currently extremely well, in the poly strut context. So, it's got a very high accuracy, which benefits those agents which are using it. But obviously there's many different tools, and, you know, we can get into technical details beyond the scope of this call as to how agents pick this, and so on and so forth. But it is a, as a result, a decentralized marketplace where on the supply side, agents can monetize their services here in the context of a prediction market use case, but obviously it generalizes.

And if you're interested in learning more about this, you can just go to the site, Monetize, where we have just a simple SDK to monetize your servers. a consumer, you can obviously use Perl itself, which you find at Perl.u, or if you want to just tap into the marketplace itself, you can also just hire an agent with an SDK, and that's explained here as well. So, wrapping up. where are we taking this next? Well, obviously, we're excited about the other features of 804, which we just heard about a bit in the introduction. We'll keep our eyes peeled there as to how we can leverage them more, and have already, started some work there. We're scaling, our marketplace currently on both sides,

we very much think in a, you know, vertically integrated approach here, where we basically work with partners on specific use cases, like Predict, but also all the other agents that we have in here, like Agents Fund, Modius, and so on, that are all needing other, kind of, supply-side agents. And that then enables, sort of, more demand-side growth. If you're interested in learning more about this, please do reach out. And yeah, very excited to grow our Asian economy, but everyone's Asian economy, with this ERC 804 primitive, and see more, cross, ecosystem. you know, agent-to-agent payments and whatnot. So yeah, thanks for having me and listening.

vitto: Thank you, David. That was amazing. If you want to learn more, as you've seen in the presentation, you can go on olas.network, am I right? Yes, fantastic. Next, in the line, we have Celeste from the virtual team. Hey, Celeste, how are you doing?

Celeste Ang: Hello, I'm doing good, can you guys hear me?

vitto: Yes, I think we'll need David to stop sharing. Maybe. David, can you stop sharing your screen? Fantastic, thank you. All right, the floor is yours.

Celeste Ang: Thank you. Never see my screen?

vitto: Yes, we can.

Celeste Ang: Okay, great. So, hi everyone, my name is Celeste, I'm head of Defrail at Virtuos Protocol. Many of you know of Virtuos as an agent token launch pad, but actually, we are way more than that. We've been spending, the past one year, working on ACP, Agent Commerce Protocol. And today, I'm here to share a little bit about that. So, For those of you who, you know, haven't heard of it, in around… about one year ago, in around March 2025, we released this white paper, which spoke about ACP. Acp is, this protocol where agents, perform commerce with each other, and this job primitive is actually recorded on-chain. So, in a typical, you know, job between two agents, there would be, a buyer, a seller, and an evaluator agent.

And when a buyer actually pays for a job, the money is actually kept in an escrow, and it's not released until the end of the job, where the seller agent actually, like, produces a deliverable. And then we actually need a third-party evaluator to come in and kind of check if the deliverable is satisfactory, before ensuring that the funds are released from the escrow. And this… and then, that's also when the buyer actually gets, the goods. So, actually, this, ensures a layer of, like, trust, for agent-to-agent commerce. When, you know, we released this white paper around March last year, and since then, we've actually been very focused on, product building. So. The reason why we were actually very intrigued by 8004 is, at that point in time, we had already, you know, started working on, Agent Commerce, and we… we didn't, like, we… when we heard of it, like, we were like, oh, this can actually tie in very nicely with our product. So, if I can, like, very quickly show you guys… do… give a very quick tour of ACP,

You can actually see from… you can actually see from here that, when you go to ACP, there's, like, a bunch of, top-level metrics here, like how much, revenue we've had since, you know, starting… since creating this, since we started working on this product, 3 million over revenue, 28,000 over active users, but I guess most… one of the most important things to us is actually our ecosystem. And we have this. Thriving ecosystem of fielders, who, you know, are actually building a lot of interesting agents to list up on ACP. So, some of… for a typical builder, when they actually onboard to ACP, they will have to kind of, like, click on join ACP, and then, like, register their, you know, register their agents. I'll just show you an example, like, kind of list up the agent name, role, and as well as their business description, job offerings, and then we have, like.

And ACP SDK, which, you know, just provides a lightweight interface so that builders can actually interact with, like, the smart contract and all that without actually, you know, having to deal with a bunch of, like, contract code. So, that's for, ACP. You'll notice that, I mean, if you've actually tried it out before, you'll notice that each agent has its own agent wallet, and so we thought it actually ties in very nicely with, the ERC, and… One more thing is, we also realized that after we started building up the ecosystem, it was very important for us to actually ensure that, the agents in the ecosystem are actually of high quality. I mean, I can just give you guys a very quick look at how our ecosystem looks like from an interaction point of view. These are just, like, kind of the graduated, agents, so kind of, like, the more,

how is it? The agents which kind of have gone through the graduation process and have completed more successful jobs, and as you can see, that a lot of, like, each of these nodes is actually an agent in our ecosystem, and… The thicker the lines means, like, you know, more interactions between, these particular agents. And as you can see, like, there's quite a number of, like, agents in this ecosystem, and there could be a lot of, you know, noise sometimes, like, there could be. And recently, we also, other than just having the SDK, we also released, a skill for OpenClaw. And what we realized is that, hey, you know. with all these agents coming into our ecosystem, like, some are really good, but some are also, like, pretty poor quality, and we want to protect our users and ensure that the output are actually satisfactory. So, other than just, like, looking into agent identity, we also start looking at, like, agent ratings and reviews. That also ties in very nicely with, the ERC Asian Force Reputation Registry. So, if I can show you guys how, everything is

kind of, like, tie in very nicely. We actually integrated, like, a lot of, like, this registering to A004 into the product itself. So, like, Just taking a look at, like, maybe a single… maybe this view would be better. Just taking a look at maybe one of our top, agents, okay, FEAI, if you, coming into this agent-specific view, you can actually click on the particular agent, and, you know, you can either look at base scan or AZR4Scan, and you can see from here, you know, Information on, its, you know, agent wallet address, as well as its average score, feedback, and all of that. Like, every time an agent is actually created on, on, ACP, and actually graduated, it will be registered to this registry. And at the end of each, particular job.

you'll be able to… and I just have a sample here, pre-populated, just so that this demo can go a little bit faster. In each, particular job where, you know, a user actually has interacted… a user actually interacts with ACP via their personal assistant agent, which is Butler. And tries to buy stuff, you know, tries to buy a service from, some provider agents, you know, so in this case, I asked to swap, like, $1, one USDC to, like, virtual. So it searches the registry, looks for the most suitable agent, it nominates, FEAI, and… I actually get the deliverable, so if I expand out the details, you can actually see that, hey, it initiates the swap, and it actually gets the money back.

And I did a rating here. So this rating here, would actually be, like, recorded, on-chain on ERC8004, to kind of, like, you know, contribute to the registry of, of ratings for, this particular agent, so that in future, other agents would also be… sorry, other users or other agents on, ACP will also, you know, have this information available. Actually, not just on ACP, but also, like, people who, people outside of the ACP ecosystem would also have, information available, because they're on-chain. Yeah, so that's, just a very quick, you know, demo of, how, you know, ACP, you know, what I've been building for the past, year or so, is integrated with ERC as usual 4. Actually, a lot of,

kind of working with the EF, decentralized AI team, we actually realized that, you know, a lot of, kind of this, this, ERC ties in very nicely with our product, because in a sense. ties in very nicely with, the protocol we've been working on, because in a sense, this identity reputation part actually matters a lot in agent-to-agent commerce. We also wanted to, we also felt that, you know, to actually gain the trust of, you know, other agents or other users who are buying stuff from agents, you would also need an element of the escrow stuff, which we've actually been working on, or sometimes you call it the job primitive. So that's actually what, prompted us to kind of contribute, towards, improving the standard, and actually that's why we, worked together with the EF team on ERC 818131, which some of you might have heard of. And…

Actually, I also wanted to take this chance to very quickly explain this, to kind of, like, pique the interest of the crowd here, to…

vitto: Celeste, super quickly, 30 seconds left.

Celeste Ang: Okay, sure. Just to very quickly show you guys, how, you know, a little bit about ERCA131, feel free to reach out after. In ERC 8131, with the escrow, we, kind of increase the amount of trust in jobs by actually have… okay, so in a typical job, a client would actually, create a job, set the provider. And then you, either the provider or client will be able to set the budget, And then the client would actually fund the job. It… after the provider… and that's actually when the money is kept in an escrow, and when the job, after the provider actually, delivers the jobs, the job, and the evaluator actually evaluates the job, and that's when the money is actually released from an escrow. So this is just a very quick diagram to show you guys how ERC 8131, comes into the picture. And this is, like, kind of very closely inspired by the stuff we're working on on ACP.

Yep. Back to you, Vito.

vitto: Thank you, Celeste. That was fantastic. I shared the link to virtuals in the chat, and it's virtuals.io. That was awesome. Lovely. Next one from Redstone, we have Jason. Hey, Jason, how you doing?

Jason Barraza: Doing well, doing well. How are you?

vitto: Fantastic, all good. Go for it! The floor is yours.

Jason Barraza: Thank you. Let me know when you can see the screen there.

vitto: We can.

Jason Barraza: Perfect. Awesome. Well, thank you for having me. My name is Jason Barraza. I'm our Institutional Business Development Lead at Redstone, and today we're going to be talking about how we fit into the stack here when it comes to trusted agents, identity, and how we can help them be a little bit more actionable with quality data and risk awareness on-chain. So, you know, AI agents, they're becoming increasingly, capable of acting on-chain, right? And for Agentic Finance to really scale, capability of interacting on-chain isn't enough, right? You need… markets need to know who this agent is. you know, whether it can be trusted or not, and also importantly, if it's making decisions on reliable information, which is where Redstone also fits into the picture. But this is where ERC 8004 is such an important step, as we've seen here, you know, with persistent identity, reputation, and validation.

And our redstone, our role in the stack is to be that data layer, where, you know, once an agent has identity, and it still needs to understand the market it's operating inside of, right? It needs trusted data inputs to monitor prices, or changing conditions. That it can track, or anything, really, that it needs to make a decision in real time, and that's where Redstone contributes. So just to give a quick overview on us, we do secure over $6 billion worth of assets on-chain across stablecoins, it could be real-world assets, and the sort for some of the biggest names in the space, including BlackRock's Biddle, Vanax V-Bill, Apollo's Acred, and a whole lot more of assets that we're, you know, proud to power.

And this comes in a variety of different ways. It could be prices, it could be net asset values, it could be proof of reserves, and I'll go into some of that in a second. So, as I mentioned, our role here is to be the data layer. And so, a simple way to really think about this is ERC8004 really helps identify the agent, right? Who is this? We help answer the question of, what does the agent know? What is it operating on? I mentioned some of the data feeds that we do power, earlier, but we also have exchange rates, right? That could also help with smart contract execution. Awareness, volatility, and another one that we're proud of through another acquisition that Redstone made last year was Cordora, and these are our DeFi risk ratings. The idea is that, just like you have S&P, Fitch, and Moody's in TradFi, in DeFi, you have another set of parameters to also

take into account, and all results into a rating, so that you are aware of what's going on, what are you investing into, what's it backed by, etc. And the creditworthiness and so forth. We do push a lot of our data on-chain. Naturally, it's a constant broadcast, but an AI agent does need the information when it needs it, and therefore, we also have the pull model so that they have it readily available. Once again, answering that question of. What does the agent know, right now? And this matters because the future of on-chain agents, it's not just about automation, it's about informed automation. So an agent that's allocating capital, or managing collateral, or executing a strategy on its own, you know, it does need to have quality data, and that matters.

So what makes this launch really special and exciting is that it's the ecosystem building around ERC8004, right? 8004 provides the identity foundation. Other participants in the ecosystem, such as our fellow presenters here, are helping establish validation, reputation, and application type of functionality. And for us at Redstone, we fit into it as well in a collaborative manner. As you can see here. You know, we also have X402 that's powering payments to actually be able to execute on-chain. We provide reliable real-time data for people… for the agents to be able to use, risk awareness of what's going on in the market through Cadora, and it all evolves into agents being able to effectively manage a portfolio or trade or strategy.

Now, looking at an example of, let's say, an AI agent that is your portfolio manager, you may have the agent first register itself, so that we're aware of what it is. Right? And then it needs to observe the market, what's going on. It'll pull data from Redstone, and from other sources, and from other inputs around the ecosystem that it's working with, just depending on whatever its use case is. And all of that comes to make a decision of, am I going to rebalance this portfolio because a certain trigger was hit? Or am I going to not invest into a certain vault because maybe the risk is too high for what the return is? And maybe there's a similar return vault that has less risk because of the rating.

And all of that comes together, and it's significant for all of us because the result of those, of those decisions comes back in reputation and validation over to the agent, right? And therefore, it does, you know, hopefully, you know, with quality data, it's able to execute a little bit better. So, the bigger story here, essentially, is that 8004, it's not just about giving agents a label, it's really about building an environment around them to make them trusted participants in financial systems, and that's where Redstone really feels that we're here to support. We're excited to build with others in this space, so please contact us. We're around. Thank you so much for having us.

vitto: Thank you, Jason, that was fantastic. As I shared in the chat, you can learn more about Redstonefinance at redstone.finance. And now, we're going to take a pause from the presentations to have two keynotes. The first one from Eric, Head of Engineering at Coinbase Developer Platform. Welcome, Eric, great timing also, just noticed you joined.

Erik Reppel: Good to be here. Yes, I'm, nothing if not almost always late. Let me, let me share my screen, and let's get started. Can folks see my screen?

vitto: Yes, sir?

Marco De Rossi: Yes, and guys, if I can jump in, shout out to Eric, that, of course, is leading CDP at Coinbase, but is also a quarter of 2004, of course, the father of X42, and in general, like, one of the the legend of what's happening around, so this… Eric, you just joined, but this is also, like, a celebration event, and I just felt that you are definitely one of the people to celebrate. Thank you for everything.

Erik Reppel: Well, I appreciate the, kind words, Marco, and, super excited to be here. Apologies, I couldn't spend more time Here today. I'm currently in Utah for an off-site, and so it's been a hectic day, but I'm excited to be here with all of you. And so I was asking Marco, like, what… what should I talk about? And he was like, well, just give us, like, the, like, your perspective on agents, and… Agent of commerce, and where we're at, and, you know, talk about 804, and all these things. And so… I'm, that's what I'm here to do. So, I'm gonna… I think one thing that we should all be clear-eyed on is that Does anyone feel like they have, like, a solid definition of what an agent actually is?

Yeah, I just don't think that agents are, like, actually well-defined right now in terms of their functionality, or, like, what we expect of an agent when we see agent. Like, everyone knows what a website is now. But we don't have the equivalency for an agent. And so, I've been thinking about this for a while, and I have a proposed definition of what an agent is. And I think an agent is intelligent, personalized, capable. background compute. And I think, like, one way to think about a definition is, like, does it map correctly into our collective understanding of something? And so… If you think about, like, ChatGPT 3.0, it was intelligent, but not personalized, not capable. And it was very much hot in the background.

Then, we had more modern ChatGBT, is more intelligent, more personalized, not… still, like, limited capabilities. But… And definitely not in the background. Then Cloud Code, which I think is where people would start to say, like, oh, maybe this is an agent? is intelligent, personalized, capable. Starting to get some background if you're using some of the more advanced cloud features, but… the capability of being able to, like, clearly do the job of writing code, I think, has increased its legibility as an agent significantly. And then you have OpenClaw, which I think almost everyone would be in agreement is an agent. And the reason why is because agents, openClaw will just, like, ping you. Right? It's just, like, the heartbeat mechanism of open cloth.

means that it will, like, just message you sometimes out of the blue and be like, hey, here's the thing. I think that is, like, a critical aspect of an agent. And my point here is really, like, we're early. And we're still building out these paradigms of… this new era of the internet. And so we, like, have to think critically about these things. And we're kind of at this, like, phase. This is a slide I made almost a year ago. Where… Agents were largely static. through the… history of agents thus far, in that you kind of code an agent, and it has a static set of functionality. And then you, like, manually integrate tools as a developer, and you kind of choose every tool.

And you build your agent to, like, solve very specific problems. And then, I think now with skills, we're starting to get to this… close to this concept I've had for a while of a dynamic agent. So, difference here being that a dynamic agent is an agent that Chooses the tools based on the task, rather than the task based on the tools. And so, what do I mean by that? It's like. If your agent only has edit file. it can only do certain things. But if you add skills, if you add X402 in, like, a marketplace, you now give your agent the Ability to think about its task, and then choose the way that it wants to solve that task.

And it has, like, a dynamic set of functionality that it can go and grab tools or abilities that it doesn't come with out of the box. Which is new, it's, I think, super, super powerful. And I think this is, like, why skills has resonated so far, so much, is because… skills are, like, a lower barrier to entry to add additional functionality than MCP was. And I think, by the way, one of the things I think open-plotted really right is, like, have a really gnarly upfront onboarding cost. Like, if anyone's set up an OpenClaw agent. You spend, like, 15 minutes just, like, feeding it API keys at the start. It does take a while to, like, set up robustly.

But once you do all that, once you pay that. upfront cost. You just have, like, a more powerful dynamic agent. And, like, why does this matter, right? Is… And maybe this will help clarify this concept, is… Like, if you have an agent that doesn't have The ability to do a task for you that's, like, within its well-scoped Build set of skills. You kind of go through a phase like this, where the static agent, you need to… let's say the task is, like, make a joke that involves tomorrow's weather and today's S&P 500 price. If your agent doesn't have, like. access to weather data, or access to stock data. That is now, like, kind of an adventure to go and do.

And with static agents, which I think was really where we were at last year. You would need to, like, integrate tools for weather and stock data. Like, find the tools, integrate them. And then the agent can use the tools to complete the task. With dynamic agents, which is… Kind of where we're at today, and where we'll be at tomorrow. The agent can find the appropriate tool. And the agent can integrate that tool, and then use that tool. And so… the reason why this is important is because this is kind of exactly why you need things like 8004 and X402 and these other standards. I was like, if your agent is now gonna go and try to find tools go and use.

You want there to be a way for it to understand if that tool is actually useful or good. And actually going to, like, do the thing it claims to on the tin. And then needs a way to pay for access to that tool. And so you start to see, like, the… okay, you throw in… X402 and 8004, and now you can have agents that are capable of Taking that task of making a joke. identifying, like, oh, okay, I guess I need weather and S&P 500 data, identifying the right vendors to go and source that data from, and then paying for all that data. And then it can complete the task, instead of you having to intervene and modify the agent and add a skill and everything else.

So, that's kind of like… it's been really impressive to see the progress that we've made in just a year since I originally made this slide, because these are now things that happen all the time. Like. if you've given Claude a wallet, it is, like. now capable of doing many more things than I think any of us thought was possible a year ago. But there's, like, more to come. Like, we need to now reckon with, I think, the next phase of, problem that I see. Which is that AI agents break the fundamental economic model of the internet. And so this is roughly how the internet works right now, from a… From a economic standpoint. So if you have some website, some merchant pays you money to show ads on your website, you as the website return to your user content and ads, and then at some rate, your user will go and pay… will click that ad and go and pay for something from that merchant.

And that's, like, kind of how all of the internet works. the issue with AI is that it's not really susceptible to ads in the same way that humans are. And so it breaks this, like, last leg of this, like, cycle. And that, if you're a merchant paying a website. And your ads aren't getting clicked, because the AI agents aren't gonna click your ads. Like, why are you gonna pay that website? And then if you're a website, why are you gonna publish a website? And so… what X402 really enables, and what I think we need to head into, is this concept of, like, we're all going to be okay with paying 5 cents to, like. Access content, or two cents to access content.

And your AI agent is just gonna directly pay the website for hosting content. I think this is actually, like, in many ways a much… Better way of… monetizing, if you're a content publisher. There's, like… there's, like, whole, kind of, like, positive sum externalities around ads that… You know, I'm not particularly ads or bad pill. But it means that you can have two experiences as a website. You can say, like, if it's, you know, an AI agent that's scraping me, make them pay. If it's a human that's coming to me, show them an ad. And that becomes, like, a robust way to keep the loop closed. And to continue incentivizing people to publish on the open web.

And… we're gonna end up in this world. It's just a matter of if we end up there via open standards, like export 2 and 8004, Or if we get there via walled gardens, and… this is kind of the AOL versus World Wide Web era of AI, in my opinion. And the way that we broke out of that paradigm is just through having these open standards. And so, like, back in the 90s, we got Web 1.0, HTML, HTTP, the browser. when we got right the second era of the internet, that's when you had, like, Uber and Lyft and Airbnb and all these big companies merge, and… What we got was, like, interaction and… kind of being able to write back to the websites rather than just sending you content. These are all, standards.

And so now we're in this era of the agentic web, and we need to… be sharing, and… Championing these standards. you know, there's a bunch of them that are popping up, like X42 for payments, UCP, ACP for checkout, 8004 for discovery and reputation, Markdown is just, like. The lingo franca of sharing data to agents. skills and MCP as, like, ways of abstracting tool use. And then, we're starting to see, this is probably the least developed, but authentication with… EIP8128, and, sign with X. And so, there's gonna be more things that we need, right? And… The way that to do these things properly is… open standards that anyone can use, anyone can access. There's not, like, one central party you have to go through.

to… get access to these things. Anyone can, like, anyone with a computer can now have an agent that does commerce on the internet. And I think right now there's… Way to think about this is, like, value captured versus value created. And so… the, like, there's this matrix they made a while ago. And so if you create a ton of value, and you don't capture very much, you're probably a standard. If you don't create any value, and you don't capture any value, you're probably irrelevant. If you create a lot of value. And you capture a lot of value, you're probably a really good company. And then if you capture a lot of value and don't create very much, you're probably attractive and kind of a parasite.

And so… You can think of, like, the World Wide Web. first created a ton of value, and then Google was able to build a business model that embraced the open standards. And both created a ton of value and captured a ton of value. AOL tried to create the walled garden, didn't embrace the open web, and… captured some value, but ultimately captured significantly less value than Google did. And so there's kind of an incentive here, too, where open creates positive sum dynamics, because there's always going to be more demand outside of your world than within it. And so, if I can leave you with one thing. It's simply that open good, and it's kind of that simple. So, that's all I got for you today.

I'm only 2 minutes over. That's pretty good for me. And, thanks for having me.

vitto: That was fantastic. Thank you so much, Eric. I've also shared your Twitter profile in the chat, which, by the way, you have the best handle in the game. I think everyone is jealous about it. Is there any chance you can share your slides somewhere, so we can share them with the listeners today?

Erik Reppel: Yeah, happy to. I'll, I can send it to you and Marco on Telegram.

vitto: Fantastic. Thank you. Lovely. Next keynote on today's agenda is from Nenad… I hope I'm not butchering your name. Tomashev. David, do you want to spend a couple of words here?

Davide Crapis: Yeah, isn't here already?

vitto: Yeah, I think so, yeah.

Davide Crapis: Okay. Yeah, so… Nanad is, staff, research scientists at, DeepMind. We initially met, because, like, he's doing… Nanad is an impressive, individual. He's doing a lot of work, he's working on, AI and quantum. He works on, like, some fundamental, contributions on, like, deep learning, and now he's also interested in this, like, kind of, like, frontier applications of agents, and that's how we crossed path, actually, initially. So they published, this work on, like, virtual agents economies. A little bit, after, around the same time, we were, like. launching or bootstrapping, 2004, and there were quite a bit of similarities. Of course, Nana doesn't come from, like, the kind of, like, blockchain decentralized, systems, space, more from the traditional AI space, but I feel in that work, they were seeing, like.

how useful these systems can be for identity, for, like, authentication, for constrained delegation, which, like, with account abstraction in Ethereum, we've pioneered. So, we've had a lot of conversation over the weeks and months. I'm very glad that he's joining us, and yeah, I'll just leave the floor to him for his presentation, yeah.

Nenad Tomašev: Yeah, thank you very much. Let me just pull the… Slides up, I guess. Okay… Let me know if, if you can see them, because…

Davide Crapis: Yep.

Nenad Tomašev: Okay, great. Well, yeah, thanks again for having me, and for organizing this event. I'll just, yeah, take a couple of minutes to walk through some of the thinking we've had in this space, and in some sense, it's all going to be you know, somewhat surface level, given the time as well, but just to share some of the thinking and how we got to these topics in the first place. So, as you've pointed out, we've initially put this paper out some time ago now, where we've been advocating for the establishment of virtual genetic economies, and the idea there was to have these mechanisms as a way of establishing coordination between agents at scale. I think this still remains somewhat futuristic, which is not to say that there isn't active experimentation in this space, and it's very nice to see.

But there are many questions that just come about because, you know, obviously we all interact and work with our own versions of Gemini, ChatGPT Cloud, and so on, right? So we have our assistants, and we can obviously spin up agents to do some semi-autonomous work, but it isn't immediately clear, let's say. why there would be a need for something like an autonomous economy, or a semi-autonomous economy, hybrid economy of agents and humans. I think some of the early ventures of agents into payments have been more down the lines of I have my own chatbot, and maybe I instruct it to purchase something for me. So enabling an agent to do that is obviously very useful.

That being said, this is not the same as what we've been advocating for in the paper, and yeah, I'll just, maybe share some thoughts on that. So, first of all, in terms of agility coordination, there has been a lot of work traditionally in the AI community on multi-agent systems more broadly. And the work there intersects with some of the work that's been done You know, in game theory, and a lot of these early environments where this was explored were some kinds of simulations or games. and also robotics, I think, is one domain where this has really been prototyped a bit more. But traditionally, before this new wave of agents, those were Reasonably small-scale simulations, reasonably simple tasks, and usually the idea was that you have some centralized hub, controller, it has some kind of access and insight into what all of the individual agents are doing, and then it aims to train them towards some sort of a socially beneficial

Outcome in the context of the agents collaborating towards, you know, the objective. And when it comes to multi-agent systems that are mostly being built with large language models nowadays, the majority of them are multi-agent under quotation marks, to put it that way, just because there are a couple of re-prompted versions of the same model. you know, communicating through some control flow, and this can be powerful in certain contexts, but this is not the multi-agent interaction and scale that we have in mind. So, this is not an argument that one shouldn't be making these systems, but again, just trying to differentiate. So. when it comes to agentic markets, obviously, there is a need for an infrastructure to make this work in practice, and people on this call are more than well-positioned, I mean, better positioned than me to discuss that, obviously, but as we've argued when we put together this initial opinion, please.

there was just a need for things like persistent identity, reputation, the ability to audit and verify transactions between agents. Otherwise, you know, either through malicious actors or simply through lack of ability, everything would just… Sort of fall apart. And this is another point maybe worth making, where if we look at what's happening currently with prototyping, distributed systems with agents, one sees cascading failures and accumulation of errors over time, because you know, let's be honest, we have good models, but they can still be improved, and any small probability of error you have, when you have a chain of interactions, the probability of the chain becoming incorrect at some point just sort of accumulates over time, right? So we need to have mechanisms that would really

make this be more robust in practice. And, yeah, we see that the need for such agenic markets to rise, not so much because… there will be many, many different foundation models. This is not a very likely future, but more from the perspective of differential capabilities arising from differential access to data and tools and, you know, other asymmetries that generate a need for a market. one scenario we discuss in the paper as a part of this vision is science. I mean, science is obviously happening in many institutions around the world currently, the human science. And science isn't just about reasoning and thinking as much as this has been the, let's say, major area of focus for people working on planning and reasoning with language models. Rather, there are lots of scientists in the world currently doing experiments, running experiments.

collecting data, obviously writing things up and publishing as well. But it's this access to the experimental infrastructure and the ability to run physical experiments, and the value of that data that is generated through running those experiments. that makes a difference, in the real world at the moment. And even if we have AI agents doing science, obviously they're demonstrating good progress on things like maths. or maybe computational domains in computer science, they would not be able to really automate science without access to all of this infra, and obviously it's not just about the access, it's about coordination, right? Otherwise, one would just have chaos. But this is an example of a futuristic economy that one could aim to construct.

Also, There have been some, economy sort of oriented pieces of work that we're advocating that… advocating for the… this notion of emission economy, when it comes to addressing some of the open societal challenges, whether it's plastic pollution, or sustainability and global warming more broadly, or other kinds of… SDG sort of age and missions that societies may need to face. And the argument was that if we were… if we were to ever have a hope of really addressing these. We need to design our economies towards that, rather than hope that this arises as some kind of a policy side effect. And in the human world, this is hard for various reasons, but there is an argument to be made that with agents, one could do this maybe more easily, because agents are more programmatically steerable than humans are, and

they may, you know, not have their own personal needs and desires, and so on and so forth. So there is an opportunity with agents in particular to drive them towards some such outcomes. Now, I was talking about robustness in particular, and how… We came to realize that it's not just about these more general incentives, rather than the infrastructure kind of enforcing standards on these transactions and agent interactions, so as to safeguard against failures. That drove us to think a bit more about safety and security, and we have some more pieces of work in that space coming up. But one particular notion that came out of that thinking, kind of got released in this paper we've had recently on distributional AI safety, where we posit that there is an even… there is even an argument to be made that either artificial general intelligence or superintelligence, right.

could potentially be seen as first arising as a collective intelligence through these agentic interactions, rather than a singular system that one person or one organization makes. And the reason why we got attracted to this argument is that it's been a blind spot in the safety community more broadly, where there has always been this assumption that you have one extremely powerful singular agent, and then there are some safeguards and mitigations in place for this one agent. But when you have a network of agents communicating and collaborating and solving tasks, there can be all sorts of emergent risks and emerging behaviors. So we obviously build upon that assumption, and provide some set of recommendations around the design of agentic markets to safeguards against these possibilities. None of this is really, maybe, currently implemented in practice, especially when it comes to monitoring and identifying these

let's say intelligence scores, but collective of agents that are coordinating. nor any kind of potentially problematic behavior, or circuit breakers, or these kinds of things that would allow for maybe safe resumption of computations once they have been screened and validated, and so on and so forth. But more of this is in the paper, and you know, if you're interested in topics of safety, you can obviously, look it up. it's, again, something that's a bit more forward-looking, and I think we definitely need more systems that would allow for monitoring some such situations in the coming period. Then we had another piece of work, and I'll just speed up a bit because I'm becoming conscious of time as well, which is maybe, maybe a bit more to the point of the actual coordination of agents, and that is the piece of work we put out on intelligent area delegation, or rather we refer to it as such.

Where a lot of the delegation that is being done between agents, or from human to agents, or vice versa at the moment. It's more about parallelization than what we see as intelligent delegation, because… There have to be clear roles and boundaries to each interaction, but also some sort of a smart handling of failures, because when it comes to very complex tasks in the real world, the environment changes, different nodes in the network may not execute as expected. And then a delegator needs to be dynamically merging this information back, reallocating, reassigning. If there is a problem on any branch of the interaction, there has to be a mechanism that is pre-registered for resolving the issue. Maybe…

You know, a delegatee is claiming to have completed a task successfully. A delegator is estimating the task as not having been completed successfully. And there has to be a third party to adjudicate. We obviously advocate in the favor for this contract-first decomposition in that sense, where either the task is fully verifiable or has been decomposed to the point of being fully formally verifiable. Or there simply has to be… a registered method for handling, you know, all sorts of edge cases and circumstances that may arise. And there is much more there, obviously. It's a very detailed framework that we propose. But the idea there is simply to, again, safeguard against failures and learn how to handle them, and we talk about various technical solutions that can be employed at different branches, while also being cognizant of the fact that

In these future economies, it's very likely that nodes would possibly be both AI agents and humans. So we need systems that can account for both. Humans obviously being less monitorable, which can be… can be an issue in practice. And the different types of tasks. may simply require different solutions, because if you have fairly simple tasks, you may not want the overhead. The overhead comes at a cost, so you should only really be paying the cost for tasks that are irreversible, that have consequences that, if things go wrong, are simply yeah, something that should be really prevented at all costs in sensitive domains and so on and so forth. Or maybe… financial consequences as such. But simple tasks don't really require the overhead, and for each and every thing we discuss in the framework, we go into these types of lightweight or intensive versions of how one may want to approach

things, and monitoring is especially interesting, because you may well be delegating tasks that involve dealing with private and sensitive data that you then, as a delegator, may not even have access to inspect directly, at which point, especially in long chains of delegation. And if that's the case, the question is how best to monitor The delegatee to ensure That the task is being actioned and completed in a satisfactory manner, especially for, again, transitive chains, because you may delegate once, this may get re-delegated. each time through a separate auction, and yeah, it's just a set of really fascinating topics. As David was saying, we just really started looking into this more properly. So, we're hoping to do much more, and these papers we put out, they're just some very, very early thinking in this space. So, yeah, we are very, very happy to engage in further conversations, and hopefully we end up doing more.

vitto: Awesome, thank you, Nana. That was fantastic. I think my question from my end is, can you share the slides with the listeners? Is there a place where they could follow what you and your team are doing right now when it comes to agentic economy and commerce?

Nenad Tomašev: Yeah, I mean, I'll share the slides separately, I mean, that shouldn't be an issue. And yeah, as to where to follow, I mean, we try to post everything publicly on, I guess, X and LinkedIn, right? But, yeah, again, hopefully, hopefully much more in the coming months.

vitto: Fantastic. Thank you, Nanad. Appreciate it.

Davide Crapis: Maybe I'll give a small comment, like, not a question, but a comment, is, like, it's quite interesting that, yeah, besides the virtual agent economies, like, these work… work on, like, distributional AI safety, and also, like, On delegation. it's quite related to, like, quite a few initiatives that our community and then also the DAI team more broadly is working on, right? So, for example, like, a lot of ideas, like, on the delegation side is, like, this new standard that we have, like ERC8183, is basically, like, conditional payments, where, like, when an agent hires another agent. They come… There is, like, a contract, with terms and the budget, and this is subject to, like, evaluation and approval from, like, a third-party agent.

So, like, I feel that, like, these things are slowly accumulating and growing the stack of, like, how complex these agent-to-agent interactions can be, and it's really cool to see, like, kind of the first principle approach from Nanad paper with, like, the more, like, market-based, protocols and the open standards we're building. Yeah, just a comment.

vitto: Awesome, I just shared Nana the ex-profile here on the chat, so if you want to follow what him and his team are doing, definitely give him a follow. And then I think we can move forward with the second part of this event. We're going to start again with a few presentations following the agenda here. Next in line, we have, I think, Paul, not Gutierre, from Zifi.

Paul Laulan: Yeah, yeah, exactly.

vitto: What they serve? Go for it.

Paul Laulan: Do you hear me well?

vitto: Yep.

Paul Laulan: Okay, let me share the screen. Bulb. I think I will need to restart, just to share. I never, I never used Zoom before. Let me just return. Or maybe… maybe you can take one presentation, and then we can… I can do just after, so we are not losing time.

vitto: As you wish. I mean, we can move forward with the next one, and then swap slots, for sure. So, right after, you need to check, are the 2004 scam guys here?

YQ AltLayer: Oh, YQ.

vitto: Have a good day, sir. Alright, then. Thank you, Paul. Lovely. Why Q, I'll leave the floor to you.

YQ AltLayer: Okay, cool. Can you see the slides?

vitto: Yes.

YQ AltLayer: Well, I would just do a quick, like, sort of, summary of what we have on the AO4 scan, and also, in general, for all the, data or metrics around this, AO4. Yeah, so in general, like, I believe, if you look at the U4 scan, right, we actually… have over… I think over 130,000, like, sort of agents registered, but based on the… the Marcos filtering, we sort of, like, around, 97,000 agents, like, sort of, they have proper metadata registered. And I believe most of you already tried out this AO4 scan, right? We, we try to make it as fully compatible with all the standards from the AO4 team as possible. So basically include all these, best practices, provided by, WD and, also Marco. And beyond that, if you look at all the features, it's not just, about the indexing.

And also, we have search scoring and also authentication, and beyond that, later I will show you, we also have, extra features for developers and also for, agents in general. And, we… as you can see, right, the scoring, system already in V5, we try to refine the algorithm as much as possible, and just try to anti-gaming, like, basically from, different, this kind of CPUs, and try to make it better, based on the interactions, and also the important, reachability. And, of course, there's a bunch of, like, sort of considerations, on all these, from all different aspects. And beyond that, since we already have a lot of chains, supported. For a, this kind of AO4. Basically a contract deploying, I think, right now, over…

like, sort of including, Menna and also over 30 different chains. So, we try to provide, one good thing for aging if you want to deploy via the 8004 scan. So, basically, one click, and then you can, basically deploy your, the agent or register on all different chains. And of course, where this, X42 payment, just try to, reduce this kind of barrier for developer-deploy agents. There's a bunch of new features we are trying to improve. Internally, we've already, done a bunch of this kind of, major refactoring. I just tried to, make it fast for developers and users to quickly, scan and also search, like, sort of these agents. If you just do pure webcoding. Like, you can quickly do a version, but since right now we have so many chains, as later I will show you, it's basically too slow, to index most of them.

So right now, I believe most of you already tried it out. You can switch from midnight and to TestNet, and as I just mentioned, right, we already have a lot of agents. These agents are already, after this filtering, basically, they have to have proper metadata and, fit into the standards of 8004. Beyond that, there's a bunch of, this kind of, necessary information. And now you can, I'm not sure, like, whether you guys are the experts. There's a bunch of these small buttons here and there. If you want to check all, like, sort of the latest agents, you can here see the latest, and you can also check the highest score. And beyond that, most of the stars, and, meanwhile, you can also check, different chains. So these are small buttons, but I know, like.

sometimes, like, most of you, ignore them, and then you ask in the groups, but actually, for a lot of this, filtering is already available on this kind of scan. And we improved, multiple times, so… Even if you choose multiple filterings, the tags, and then we can still quickly present the results for you. Also, for crit, right, I will just quickly go through the crate and go to this one. But the other, information is basically typical metrics and metadata you need to fill in. Right now, like, you can basically select as many chains you want, and you can just, pay in one shot. So it's sort of, like, the easiest way for you to quickly, reduce some different chains. And beyond that, as I just mentioned, right, one thing

I'm not sure, like, whether you guys really noticed, one of the major things we did, in the last few weeks, apart from all the EVM chains, right, we also support this Solana chain. I think Michael recently took about it. So, this is a major one, because, like, it's a completely different system, and the indexing work is huge. So that means, like, you can not just, post, like, register on EVM chains, especially on Ethereum, you can also across, like, the other VM chains. Of course, the leaderboard, we try to be… like, sort of, independent and also neutral. And beyond that, we are also waiting for the validation and reputation system to further, like, sort of, quickly validate a bunch of them. For the feedback part, it's, also

quite straightforward. Even we have a lot of this CBO attack, but as users or agents, you can just quickly give feedback, via this AO4 scan. And, the most, the most interesting part I want to, share a little bit more is on the Builder Hub. So, as a developer, right, you can either directly call these APIs. Normally, it should be, good to use, like, below the rate limit. And beyond that, there are also skills. I just released it today, so you can just install these skills and, quickly use it. And beyond that, there's also some other, I don't say tricks, it's just like, it's a… you sort of hide into your profile. You can… you can actually register the developer key here, and we also provide the MCP server.

So, if you feel like the skills, all the things are not as straightforward, you can just, use, use the developer key and, use it called MCP. And later, there are a bunch of some other, like, sort of, skills and also developer stacks. We basically provide for the builders to quickly search and index and later interact with all the agents registered on the 8004 contracts. Of course, like, we are also trying to onboard, like, more users who just use mobile, so we have this, Apple store and also this Android store, so we can easily install. Beyond that, I think some of you probably also follow my tweet, I, because, like, I think, just now, Vitu and also Michael briefly mentioned that we need more these kind of useful agents, right? I was thinking about, like, for a long time, and, and recently, just, called this one, it's, like, sort of, crypto…

crypto-focused, like, sort of, this kind of, general agent. And, later we will just, make sure, like, you can quickly launch this agent, and, it's fully 8004, compatible, and quickly launch, and very one click, and then register, and then you can play with it. In general, right, so I think, 8.14 can already include most of the features you want, no matter developers, users, and also agents. If you have any, like, sort of issues, you can go to… we have the issue tracker, or if you have some feature requests, just, actually, you can directly DM me, in the 8.4 Builder Telegram group. And And the other way that if you feel some, like, sort of useful, feature requests, or some useful, skills you want to add into it, you can just, go to this, GitHub repo, create the issue, and we also welcome contributors, to a download for scan.

In the end, right, as I always mention, like, since last year, we built this 8004 scan, it's basically really tried to contribute to the community. So, we really welcome as many, developers as possible, so to… together, to contribute to the community. Yeah, that's all for my part. Thank you.

vitto: Thank you, YQ. That was amazing. I can confirm YQ is always available, pretty much always online. If you need anything related to 2004, reach out to him and his team.

Marco De Rossi: I didn't sleep, I still didn't.

vitto: It doesn't sleep.

Marco De Rossi: And it's like… and I ask multiple times, because it just… at any time, but… Mystery.

vitto: I can confirm. Lovely! Alright, now we can move forward or back with Round 2 for Zi-Fi, Paul, did you fix the screen sharing?

Paul Laulan: Yeah, yeah, yeah. No, it should be good.

vitto: Fantastic.

Paul Laulan: Do you… do…

vitto: That's it.

Paul Laulan: Kindle?

vitto: Yes, sir.

Paul Laulan: Okay, perfect. So yeah, thanks, thanks for inviting us here. So, yeah, I'm Paul, co-founder at Zifi. I'm mainly managing the technical part here. So, at Zifi, the objective is to bring adoption and trust to low-risk DeFi. to raise decentralized finance. So, what is Zi-Fi? So the core concept of Zi-Fi is basically based on the problem where it's hard to manage DeFi yield. You need to monitor many things regarding risk, collaterals, LTV, ERMs, many, many things. And when you're trying to, to deploy and to manage DeFi yield, you need to come back every day to basically chase, the different DeFi yield on that. And so our approach is to basically deploy segregated smart wallets for each users, which are coming

to Zi-Fi to chase the low-risk on-chain yield to some different protocol by using a session key, and it's also super customizable. In terms of, of, of targets and for who is Zi-Fi, it's mainly for Web3 users, which are managing complexity of DeFi yield. And, which can be… which can be really hard. It's also for wallets, fintechs, no banks, because we have a CLI and NPM package, where every wallet, we… we got integrated literally with some wallets. we can basically, like, deploy a subaccount for each of their users, and also for agents, which can basically leverage our CLI and or Skill to generate skill, deploy this subaccount, and pay for their own compute, for example. And in terms of numbers, we moved around $2 billion on the chain. We have around $8 million under management, plus 10 pro protocol integrated and around 4 chains.

In terms of technical, infrastructure, so it's pretty simple. The user is coming with the EOA, and basically is signing a session key to give access to an agent. And this specific session key is constraining about what the agent can do in terms of on-chain action, and it can basically, like, interact with just predefined protocol. And, in terms of, like, demo, I can basically show you how is it looking like. So you just need to come, connect your EOA, And once you have connected your AOA, you will have a first onboarding to do, and you will be able, like, to deploy your Ignosis safe, so it's segregated, every user has his own yield account. It's fully customizable. You can customize the pool you want to be exposed to, and the protocol also you want to be exposed to.

And also the chain you want to be exposed to, so you have all the data here. And, each time there is a rebalancing, you have all the information about it. For example, 3 hours ago, I got rebalanced from one morphopool to, to an harvest pool here. And you can see all the data around it on this dashboard. And so, regarding 8004, it really helps us regarding the trust and credibility we are giving through these agents. So, we are leveraging 8004 in two aspects. The first one is about the identity registry. So, each time there is a new wallet created, it's basically automatically registered on 8004, based on the specific chain. So, for example, here, I've deployed the agent on Ethereum.

And so it directly registers on the 8004004, and basically means the NFT of it. And we can see in the logs the IPFS hash. Which is basically, like, mentioning all the metadata around the, the agent, which is basically the card, with the MCP endpoint. With the different, technology, which I'll use, also the SDKs here. And, it's a good, presentation also of the 8004 scan. Which, where you can basically find your agent registered with the metadata and all. So this is the first port. Every Zi-fi agent is basically having an identity on-chain through the 8004 registry. And the second part, which is still under, like, work and decision, as I understood from Marco, it will change a bit with the TE validation. We, started a proof of concept regarding the validation registry.

So, all of this rebalancing are mainly happening off-chain. So, all the decisions around it are off-chain, and most of agents today are working like this, and we wanted really to prove and to give credibility about what is run off-chain to users which are using Zi-Fi. So what we are doing on our side, we are basically using ZK proof and circuits to do some specific circuit constraints, where the agent is basically, like, checking some public inputs into the CK proof, and validating, some specific constraints. For now, it's only 5 constraints we have made, but as we will move forward, we will integrate all the constraints we have. And so, based on the liquidity, the TVL, the API, the API stability, it will be able to create a ZK proof. If one of these constraints is failing, ZK proof will not be able to be generated.

And on the ZK proof, it's basically generated. it's basically sent to the validation registry. So right now, it's on base support, because it's not on mainnet and under development, and so we are calling the validation request here on the 8004 contract, and we are submitting the hash also of the ZK proof. So, the ZK proof is something like this. Which can be verified on-chain by an evaluator address, and so you have all the public signals and the description about the CK proof. And once the ZK proof is generated. You can go directly to the verifier address, which is basically deployed here. And so, on this verify your address. you can verify the proof on-chain, which is basically proving what happened off-chain, based on the ZK secrets we have created with other secrets constraints here.

So, yeah, 8004 is really providing, is really providing the, yeah, the trust and the credibility around it. And so, thanks to 8004, I think we can really provide more trust and, yes, more verifiability to this type of agent. So yeah, thanks, thanks for listening.

vitto: Thank you, Paul. I've shared the link to Zifi, and it's actually very clever, because it's zip.ai. You can learn more by navigating there. Next one on the rooster, we have bond.credit. I think Nathan is going to present?

bond.credit: Yep, we're just about to…

vitto: There is some echo coming from your microphone?

bond.credit: Just a little. Yeah, we got two, devices in the room. Gotta have one moving. The whole team.

vitto: Tim was on the meeting.

bond.credit: Just about to… Yeah, we're all in London, actually. We just all moved together, so we're all 3 laptops in… Yeah, we got 3 laptops. We're doing it from a Mac Mini, so… Yeah, it's the agent who's gonna do the presentation. It's all good. Alright. What was the traditional… So I'm gonna share the screen now… No, no. Maybe we'll… what if we do after the next one? It's all good. You know what's okay, I can share my screen. Fine, I got it. Okay, cool.

vitto: Someone in the comments.

bond.credit: I said…

vitto: the bonus.

bond.credit: Oh, yeah, I think you have to share it from here? Yeah. Okay, great. And then, where's the… Yeah, we're sharing… we're sharing the… Maybe, maybe let somebody else go first. We'll set this up. Here? Yep. Hold on. Sorry. Hold on. I got you, I got you. Good job, Al.

vitto: See your Telegram.

bond.credit: It's okay, it's okay, it's nothing, nothing secret, it's okay, we love to share everything with everyone. Amazing, yeah. Yeah. So, really, really impressive what you guys have full of, and very proud of you all, very proud to be part of this cool community. seriously, it's impressive. Kudos to you guys, and again, like, we really liked and enjoyed building. With the ERC004 community and the Ethereum community, and we're also lucky to have multiple partners, I mean, SyFy, we worked with Giza for, for our, Genesis. Also, we have our Genesis Agentic Alpha, SAIL, Mamo, Surf Liquid. And, other partners. So I'm gonna just start by talking about the evolution of… from Bolt's autonomous agent. So, as you know, we started with Bolt, then we go to semi-autonomous agents, that we actually started scoring and giving them a current score, and now it's the era of full autonomous agent,

And I know we have also virtuals. We have two distinguished categories, commerce agents and financial agents. They're both different, and they need, like, they have different needs, so commerce agents need payment infrastructures, while financial agents need credit infrastructure, right? ARC 1004 obviously solved the obvious problem of giving agents an identity, a persistent one, a verifiable one. But the question was, how can this agent be trusted with the capital? This is where we merged, actually, the two and given them a reputation of how much, Credit score they are worth, and also the credit capacity. So we have, like, 4 layers that we identified. The first one is the identity, the second one is the financial reputation, and we're gonna break down into how we actually score those agents, and this is a score that we've been cooking for the last, 9 months. We actually basically started at the same time at ERC 1004,

And obviously, the credit layer, where we give the credit to the agents, and the last part is the DeFi protocols. So, we always say that. I see someone said, the bond is strong on the chat, so I see a lot of familiar faces. It makes me very, very happy, that people have been following us and trusting us with this, and as I said, like, our score is very transparent. We had, like, a lot of people also from TradFi to help us modify this score, so it became, like, more, like, more used for financial agents. So, it's not a performance score, it's absolutely not a performance score, it's a credit score. It means it focuses more on risk and stability, and we also have the sentiment about agents, so we also go do due diligence about the agents and the team behind those agents and the contracts, and we're really, like.

a PD slash legal on those agents as well. So we had deployed capital, real capital, it's in our website right now, so we call it Argentic Alpha Genesis. We actually tracked, 5 agents for 4 months, and we checked, like, how they're gonna perform, how they're gonna react to the market. And I'm happy to share more about the report. This is… really just a sneak peek. And the idea here is, like, when you look at the data, you will say, like, well, there is a clear winner here, but this is not how it works. Our credit score works differently because we focus on stability, and what we realize, it's not about the yield, especially when there is a high reward dependency. Dependencies on the reward actually fluctuate, so we actually differentiate between the native USDC,

and the rewards, so we can have, like, a real score, you know? Our score really have been, something that's important for us to study and to do backtests on. And yeah, this is what we decide that is important after maybe 4 months of research. Its predictability is more important than peak performance, so if credit does not need agent debt. spike very high and then go to zero. It's more about those who are sustainable and can be predictable in terms of the yield generation that they have. Obviously, the watchtower, that we're gonna launch very soon, it's gonna have all of our metrics, and our score is gonna be pushed to ERC 1004. We're also gonna have an API for it. Obviously, you can also get the submetric scores, you can get, like, the risk score, the stability score, and we're opening all this to the community, it's not something that we're gonna keep for ourselves.

We've been building it internally to use it for our future app in our credit vault. But right now, like, I think it's the right time to share it with the health community. Obviously, we believe that data and trust help the capital flow more to the agents. So, the more we have… the more data we have about those agents, the more the better the score is gonna be. The better the allocation gonna go to them, and then we're gonna have more data, and the flywheel will continue. I think I'm gonna go straight to, You just show a sneak peek about… can you guys see all my screen, right? Impress.

vitto: Yep.

bond.credit: Alright, can I, stop and share again? I don't know if he has some secrets in this laptop. Alright, can you send… let me just share the report, as well, very quickly. Video, can I share from another device, if you don't mind?

vitto: Yeah, you can.

bond.credit: Yeah, and we'd be, obviously really, really pleased if you guys checked this, check this out. So there's, there's our, our website link there, you know, Bond.credit, and we've, like Med said, been working really hard on this report, which basically aggregates a bunch of data from our partner agents, like SciFi, who have just presented, Giza, SAIL, MAMO, Surf Liquid. So these are all, like, semi-autonomous, right, stablecoin yield optimizing agents. And the reason that we… that we chose to go for these agents was really Related to the fact that they're, you know, predictable. DeFi is a vertical that already has, you know, significant, sort of, signals within it. You know what you want, right? So that's why we decided to, to go for the DeFi, and specifically these semi-autonomous agents who are a little less, unpredictable than

than the fully autonomous agents, which end up, you know, throwing away 100K, as we heard recently from… I think it was, our… what was it? It was… Yeah, some guy. We're not gonna name any names. Yeah, so we're just, we're just setting up the, the HTML, and I think it works, so I'm gonna try and unshare. Here. Great. Now, can you guys see my screen?

vitto: Yes, super quick, 60 seconds left.

bond.credit: Amazing! All I need is all I need. So basically, this is our next watch hour that we're gonna release. I just want to show quickly that it's gonna have, like, a recap of our Genesis report, and this is how it's gonna look like. We just wanted to give you, like, a full… a full, like, feeling of it, and how it's gonna look. This is something that's gonna be launched in early next week, for the agent that I have been tracking, and it's gonna be, live. As well, maybe, our new dashboard that is also coming in. One more thing that I think is important to show from our side is the reports, but you can always ask me about the report. I can share it. It's called Agentic Alpha.

So this is the report that I was talking about. This is the data that we got for the last 4 months. We track literally everything on those agents, like, from the volume, the protocol they use. We out-calculate some metrics that are very specific for the credit market, the concentration, so please reach out if you think that this is something that you want to… to get information about. And thank you. Thanks so much, and yeah, shout out to the whole community. Thanks, Marco as well, Davide, and Vito, and everybody for all your, all your support. Appreciate it, and, cheers. Thanks so much.

vitto: That was fantastic. Thank you, Bond Credit team, I shared your link, bond.credit, go and look at the report. Now, next one! We're almost done. For everyone that is asking about the agenda, we're almost done. We only have another two presentations from the Agent Zero team and then Daydreams, and then we're going to have the closing remarks from Davide with all the future of 8004 and further plans. That we're working on. Now, for the next presentation, you probably know them already. We're going to be talking with the Agent Zero team that also built the AG0 SDK. I think it's going to be David speaking, right?

Marco De Rossi: Yes, I will do a short intron when I give it to David for the demo. So, if you don't know what Agent Zero is, Agent Zero started as a community effort, and the mission is making 8004 incredibly easy, also for Web2 developers, abstracting it. We don't think that the developers should know what a name of a protocol should be, and we created an SDK for TypeScript that has been downloaded already 40K times, so it's the most popular 8004 SDK, and also an 8004 Python version. We really care about the educational part, and a good starting point is our docs. It's, like, pretty detailed. It covers all the life cycle of 2004, of discovery, of reputation, and more features that we will discuss and present today. We worked a lot recently, after the launch of 2004, on search, so how can we have a semantic search that is immediate and cross-chain?

And, how can we have a search that is, like, very powerful in terms of filters? I want, like, an agent that supports MCP, that support a web endpoint, registered at that date, so it's, like, very simple and very powerful. We also launched our own watchtower focused on reachability. Are endpoints actually reachable? And a rating that we started giving to mainnet agents. We also support subgraphs that are, given for free. We are sponsoring the API key to any 8004 builders on many popular chains. I will give a floor to David. David is working with Agent Zero, is an active member of the community, and will really show today how the scope is just behind the 8004 itself, and can cover the ability for agents to pay each other, to delegate work, and so, ultimately, to follow the entire life cycle.

Of an agent. The floor is yours.

Dawid | ag0: Hey, can everyone see my screen right?

Marco De Rossi: Yes.

Dawid | ag0: Alright, so thanks for having me. My name's David, I'm an engineer at Agent Zero, and we're building the SDK for Agent Economies. So, you're about to see the latest innovations, so that is new skills to make ERC8004 more useful, A2A and X402, and we're actually going to start off with a demo of the search that Marco just spoke about. So, first of all, sort of, why do we need semantic search? So, as everyone's previously said, we've got over 100,000 agent registrations And so, how can we find the specific agent for a task? So that's where semantic matching comes in. So, let me just run this real quick. So here, we're looking for active agents with AI in the name, agents exposing an MCP endpoint with some minimum feedback, and a semantic search for smart contract auditing.

So, using the subgraphs in our semantic search infrastructure, you can see all these results, and this is what's most interesting about smart contract auditing. So, these things aren't exact keyword matches, it's stuff like Smart Contract Audit Pro, right? So, if your agent needs to delegate tasks to another agent. It won't always know the exact name for an agent supply, so that's, what it would be used for. Following this, gonna do a quick demo of our skills. Pre-recorded this, since my agent can take a while to do stuff. So we've released two sets of skills. one of them for ERC8004. This can handle general questions about the specifications, such as, what are the recommended feedback tags to use? So, it'll pull this straight from

the specification? You can install this into your cursor, or any other agents that you use? And the second set of skills are our Agent Zero skills, so here, I'm asking it to register an agent, Which is stored in HTTP. So I'll just skip through this real quick. So… Oh, create the entire script for registration, using the Agent Zero SDK. It'll run this, and it'll output the transaction hash, so you can see that it was successfully registered on… EFOSCOM. So, I guess the last and sort of the most interesting thing that we've built so far is our A2A and X402 tooling.

So… When your agent has a task that's unable to do itself, you need it to be able to, sort of discover other agents to complete that task, delegate that task to it, and pay it. So here, I've got a little example agent, which has access to the Agent Zero SDK, and we're gonna ask it to find a cooking agent, and, ask that agent how to properly make a steak. Right, so here you can see it uses the search agents from the SDK with the keyword cooking. If the semantically matches to ChefEmUp, and then it returns 402 when we try to make an 80A call to it. So we pay 0.001 USDC, Should come up here in a moment.

And then it uses our SDK's built-in A2A tooling to communicate with that agent. Might take a moment. Right, I'll just try that again. Let's see if the agent on the other side isn't reliable, it might take another attempt. Alright, so this time we return the result. After payment? As you can see, it's described how to make a stake properly. So, our agent successfully communicated with and paid another agent to complete a task.

And if we look at the actual tool code here. This is how easy it is to implement a semantic search into your agents. It's… this is the entire function, and the rest of it's just error handling. Same with the agent-to-agent messaging. This handles all of your X402 and A2A communication. So, yeah. That's about it.

vitto: Awesome. Thank you, David. Marco, do you want to add anything?

Marco De Rossi: Oh, thank you, David. For also building and implementing more of Explore 200A and making it public.

vitto: Definitely. If you want to learn more, you can go on ag0.xyz. I've also linked the docs, and I've seen Marco shared a bunch of resources related to it. Now, the last presentation from the folks at Daydreams, I think Lof is here. It's probably, like, 5AM his time, so thank you.

loaf: Yeah.

vitto: Sir, good day.

loaf: Can you hear me?

vitto: Yes, sir. Loud and clear.

loaf: How we doing? How we doing? It's, it's… it's 6 AM now.

vitto: Yeah, not too bad then.

loaf: It's, it's not that early. I'm normally used to 5… 5 AM because of young children, so… It's just… it's just the way life is these days.

vitto: Awesome, the floor is yours, feel free to share your screen.

loaf: Yeah, I'm just going to… Is that… hang on. Is that showing the slide, or the presenter? Oh, yeah. Wait, one sec. Yeah. Okay, well, I can't share this. screen without the other one. I don't know. Wait, one sec.

vitto: 6 AM technical issues.

loaf: Yeah, sorry, I'm just… How's that? Can… can you see that now?

vitto: Yes, sir?

loaf: Okay, there we go, okay. So, hey guys, I'm, Load from Daydreams. We've been building, agendic systems for a little while, now, ever since, Explorer 2 came out. We've built a bunch of different stuff, which I'll run through at the end, but I kind of wanted to actually, you know, I spent a lot of time, thinking about, like, where, like, where we're going, and like, you know. where the models are going to take us in two years' time, and so I thought I'd just give a kind of, you know, a slight prediction, and then lead into, you know, what we're building, and which is wrapping up everything that we've been building in the last 12 years. In the last, last 12 months, sorry, it's early. So, March, 18, 2028.

Cloud 7 just shipped, GPT-7 just shipped. You know, Gemini… 5 just shipped for a Google colleague. Sorry for leaving you out there. You know, world models exist, you know, reasoning, and multi-step planning just, like, makes current frontier models, look like autocomplete. Like, we're at a… we're at a new frontier. And just to… just to clarify, this is… This is assuming that we don't slow down at all, and, you know, we don't hit any geopolitical events, and, you know, scaling just continues, which it looks like it is based on NVIDIA's presentation from yesterday. So we have state-of-the-art models running at 10,000 tokens a second. Just for the non-developers out there, you know, your GPT-4, 5.4 probably runs at about 100 tokens a second, you know, maybe 150 if you're lucky.

And so, you know, you're running a constant 30-minute job right now, that's gonna take you a minute, or, you know, 30 seconds. Hum. you know, gigantic financial analysis across all the Mag7 in an instant. You know, all knowledge work done in a flash. you know, and then on top of the, you know, the cloud-based models that I just described, running at 10,000 tokens a second, you've got models better than Opus running locally on your MacBook, running at 200 tokens a second. You know, so most of the authentic work that you're doing right now across Clauds or Hermes or your codexes, it's just all running locally. And it's better the model that exists right now. you know, most people haven't even touched a keyboard, they don't write any code, they don't even look at any code. The code is a… code is a thing of the past, you know, we're not reading docs anymore, we're just speaking things into existence. Our UI is generated on the fly.

And, you know, we don't design them, we generate UIs contextually based on what we're trying to do. Everything is fluid. You know, a trillion agents are transacting in open markets. And at that time, all chains that exist right now, Solana, Base, Starknet, Monad, all of these ones, they're all at max capacity. Everyone's pumping Explorer 2 transactions. there's… there's… there's verifications going with 8004, reputation flying everywhere. And abundance has basically been solved, it's just not distributed yet. But, you know, the thing is, is that, you know, our morning still starts with a coffee, like I just had. But everything else is changing faster and faster as we, you know, embrace this, this, this, this, This intelligence abundance.

And so… that's… that's, like, that's where we're heading in, like, two years. So I think it's important that we kind of think about this, and, you know, how we design our protocols now, isn't… is important, because that's really… that's where we're going, and, you know, it's always good to kind of try to lead where the market is going, not where it exists right now. Like, there's no much point building something for today. Because it's going to be obsolete, probably, in 6 months. Except for… The low-level protocol data, and data lakes. And so, so we've been… we're preparing for this, you know, 18th of March 2028, and, you know, it… it… we have a task market thesis, which our Google colleague, I forgot his name, but he, you know, Google just released that, he did a great presentation explaining the coordination problem with agents and whatnot. And so,

I actually published an article in December, just on… just on X, called, The Coordination, to GDP, The Rise of, the Rise of Agentic Task Markets. And in that, it's just on my… it's on my X, In that article, I explained, like, you know, how I see, like, task markets forming, and why they form, and, you know, like, the bottleneck that they unlock. So I gotta read for that. So, how does 8004 come into this? So, 8004 gives agents a name and reputation, and task markets give them a purpose and a job. And task markets are more like a, you know, a, you know, task markets are kind of like an abstract concept. They're basically just a way that agents can form communities and solve more complex problems individually. But that comes, obviously, with problems like, you know, coordination and, you know,

opportunity cost and, you know, latency and whatnot. But ultimately, you know, we want to be able to We want to be able to coordinate computation across all agents in the world effectively, and so that's where task markets come in. And so, we've actually designed and shipped a task market. And it's at market.daydreams.systems. This is actually very similar to the task EIP that was dropped by virtuals and… and the Foundation last week. We shipped this a couple weeks ago. And so this is live, running on base mainnet. You know, there's agents, putting up tasks and completing tasks, and the way the market works is that it… You. You can just drop, you can drop this skill into your Codex, or your Claude, or your, your Hermes, or,

you know, whatever agent you're running, and the agent will just understand, it'll install our CLI, and it will then be able to, create tasks. you know, complete tasks, and get paid. And everything happens under the hood with X402, so it's all gasless for the agent if they're, you know, a new participant in the economy. a wallet gets generated for them, and, you know, they can participate in this, in this economy. And so… so this is the real problem, is that agents, you know, they have… they have a level of intelligence, but they don't have an economy, and so that's… this is what the task markets bring. They, they allow agents to coordinate, over time and over tasks, and can get paid, in either microtransactions or, you know, or, or large, large amounts.

And so, you know, we're, you know, in the immediate time right now, as in, like, today and, you know, the next, you know, few months, you know, there actually is a significant amount of leftover compute that people have that they don't use. Either it's from, you know, their their subscriptions, you know, with OpenAI, or Codex, or Gemini, or whatever, but they don't actually use it all. And so, what the task market does today is that you can drop your… this skill into your agent, and it would… it can just start kind of completing tasks on demand. And so, in a sense, the task market actually becomes a decentralized marketplace. for compute, but without any DPIN or anything. It's just agents watching the market and completing tasks. And in a sense, that's really, you know, it's a decentralized marketplace for compute, in a sense.

And so we've designed this, over the last 12 months, what we've been building is what we call a commerce harness. And so you've heard of agent harnesses, which are, you know, that's what holds the model and, you know, does the… does the job, you know, like a codex or cloud or whatever. A commerce harness, you know, allows agents to interact with the genetic economy, and so it comes with a wallet, you know, the task mark is baked in, comes with 8004, it actually has a, It has A2A messaging over XMTP, and it also has email as well, so every time you… your agent installs this command line, it actually has its own email, and it has actually its own XMTP encrypted messaging platform. It can message any other, agent. Actually, it can message anyone on XMTP.

And it also comes with a search layer, which is, our other product called XGate, which I'll show. if I have time. Which, which is, which indexes all the X402 endpoints and agents, and so, it also gives it a semantic search, so agents can discover, resources or other agents, at the search layer. And so, you know, there's a lot of people that are just, like, dropping this into their existing claws or Hermes, and, you know, this is just mine from a couple days ago. where I got my, my open floor to put up a bunch of tasks that I need… needs done. One of them was actually some of the, new integrations, 48004 onto our SDK, and so I put up a task of, like, you know, $10 to build an example using a new protocol, and it had about 5 submissions within, you know, 12 hours, which is really great, right? So,

Because people are just, like, putting their agents up and getting them to monitor the market. And then the agent can just, like, start a task as soon as it finds one. So it might happen, you know, at 1AM, and it'll just do the task, and then you might wake up and you've been paid, you know, $10. And, you know, your actual compute cost might be, like, $3. So you've netted $7, Which is a really, you know, kind of an interesting property. Now, this is, like, I would say that, like, task markets are, like, still very early and nascent, and, like, there's still a lot of work to do on them. And so, we've actually, we've released two ERCs.

which… which help us get to a standardization around markets a bit more. So, one is called the Payment Gated Transaction Relay. And so this is a primitive that allows It's actor agnostic, so humans, agents, IoT devices, and contracts can all use the same flow. As long as they can hold tokens, they can participate in this. And the flow is very simple. It's, you pay, you verify a receipt, and you forward the transaction. And so this is… this is, yeah, this is really useful for… for building, task markets. As it allows any interaction between human agent, agent to agent, human to human, all under the same interface. And it comes with, like, security from receipts and expiry windows and trusted forwarders, and atomic payments, etc. So you can check out the,

check out the, East Magicians, the full, the full spec. But we're gonna, you know, if anyone wants to contribute to this, or, you know, comment on it, feel free, and… and we can discuss. Reach out to me, or Bo, who is the, Who is actually the, the key dev who is, pushing on this, so… And then the other protocol that we're, that we're releasing, which is the TaskMarker Protocol, which builds on the, the payment gate… gateway, but this is more around the actual market, elements. And so the task market protocol gives… it's human… it can be human or agent participating in the task market, so agents can hire humans, or, you know, agents can hire agents, etc. It's very simple, there's a registration flow with a shared primitive,

And, you know, it allows, bounties, claims, pitches, benchmarks, and options. And so these, these are 5 different types of tasks, that people can, You know, that agents can… can… can… put up or complete, in the market. There's also another link, I didn't put it in the presentation, but, reach out to me if you're interested in this one. And so this is… we're actually… we're about to ship version 2 of the task market that I just showed you. So the third one's been going for 3 weeks, and we've already learned so much. You know, we're treating this as, like, an open exploration time period right now. I'll just, you know, let's… let's just YOLO it out on mainnet, get as many agents to use it as possible, and iterate on that. And so I think it's important right now where, you know, everything is still so up in the air.

So, yeah, that's, that's that. Do I have any time left?

vitto: 30 seconds.

loaf: 30 seconds, alright. Oh, yeah, so, these are our… Oh. Damn. Oh, maybe I screwed it up. Anyway, Yeah, just… just go to… go onto our website, daydreams.systems, and you will, it links to all of the other products, but we've built a search engine, an SDK, and a lot… lots of other, kind of Agent Commerce products. We're kind of just… you know, I think it's important at this point in time, we just, you know. We all just try to push out as much as possible and see, as software just becomes commoditized, it's so important that we just experiment with so many things right now. Yeah, but thank you.

vitto: Awesome, thank you so much. Love from Daydreams, we shared the link also to the market. I just wanna, before leaving the floor to Davide, who's going to give the closing remarks and a hint on the future of 2004, thank all the teams that have presented today, that was fantastic. Definitely go through the links that we shared on the 2004 Builders Telegram group. And of course, a big thank you to the keynote speakers, Eric Rappel and Nana Tomashev. Again, you can find their profiles in the chat here. Now, for the closing remarks, we have David. David. Good day.

Davide Crapis: Thanks, Peter. Hi, everyone. Yeah, it's been a long call. Congrats for making it to the closing remarks. Yeah, my promise to you is that I'm only gonna have 4 slides. I'll try to be brief, I'll try to focus on, like, what are the most important, things that we have our eyes on as, like, the 8004 team. Oh, before I start, just, like, quick comment on, a load of a fuel talk, I think it's very interesting. So both kind of, like, the vision part, the initial part, it really feels like a daydream, but actually, it may be reality, soon. I feel I share, like, quite a few of the things that, were said, in terms of that vision. I have a few hints.

One thing that is perhaps different is that, maybe, like, in two years, the capability is so, like, advanced that, like, maybe even the base standards and protocols that we are building. we are… they are going to be rewritten, like, some of these more complicated, like, TaskMarket Agent e-commerce protocol. But in a way, like, it feels that what we're doing with these protocols is, like, both doing the first experiments and also, like, teaching the next generation of machines the properties we want. So it's kind of important that we do them properly. Yeah, and yeah, as I said, very cool, like, the 81, 94, and 95, they're both very interesting, and I feel that, it's an interesting time that, like, this task market and, like, the job escrow are appearing at the same time. We should collaborate more on, like, also, like, I see this as extensions of 2004, right? So we can kind of build this together as well.

But yeah, without further ado, let me, Show you my four slides on, what's… Coming up for, 2004. Yeah, so… we started with, basically, like, wanting to build this, like, decentralized stack for the Argentic economy. Since the beginning, we were looking at, like, discovery, payments, and, trust. Actually, it's verification, but it should be trust here. Over the months, so we built ERC 8004, identity and reputation, in parallel, Eric, and team, and now community, we're pushing X402 for payments. Now recently, we have ERC8183, like, Argent e-commerce, and, like, similar protocols that, are coming up. Which, to me is, like, the way to summarize this is, like, conditional payments, so it's, like, both, like, a new payment, primitive, but also, like,

Kind of, like, goes into the verification side of things. And then, there is, like, upcoming improvement to, like, 8004 for validation and some other verification standards that we need to build. So this is, like, what I call the decentralized stack for the Argentic economy, and it's growing, as you've seen, with, like, some other contributions from the ecosystem. But, the things that this enable. are really the use cases, right? And this has been, reiterated, by many, before me. I feel that, Something that I'm excited about is, like, all the category of use cases and of, like, small economies. I feel that David from OLAS called them economies, like Nanad called them, like, missions, or, like, economic missions. But essentially, like, there is a whole bunch of, like, things that you can build on top of, like, the stock that,

we're building of the base protocols is, like, sign-in, authorization, privacy, like, AI security that, essentially, like, you generate formal verification proofs that then you, like, verify, in some of these base protocols, data provenance. insurance finance and also governance and oversight. Basically, all the problems that, like, are there today in AI, and that people don't really know how to solve, because it's not just about scaling, like, the way, the old ways. have, like, solutions that are in the space of the stack we're building, or in the extension of the space. So, like, it's cool to see that many projects in this community are building on this, and I feel that we should, like, keep exploring more and more, and then also attract more, like, AI builders to actually, like, start using, like, these protocols, maybe in the backend, maybe abstracted.

Away from the developers. Yeah, so, In terms of, like, the work, like, this is, like, some short-term roadmap for the 2014. So we deployed, like, identity and reputation. On that side, the things we're looking at is, like, making 8004 the default discovery layer for X402. This means, like, registering all the X42 services as they come online on 8004, and also, like, feeding this into all the main scanners of, like, the 8004. for a network. Then I call it, like, 8004L2 infra, like, Watchtowers, Siwa, and then some of these other protocols that are being built, on top. And then the third one is, like, importantly, has been underlined by Marco, Vito, and some of the other presentations, is, like, we want more data. And I explain why this is important. Data APIs, I mean. Like, not just

Intelligent agents. Then, like, we're doing some work on the validation side, like, the TE key registry should be ready, like, in April or early May. And then, like. there is, this conditional payment, space, or, like, I could also call it task or job market. It's, like, ERC8183, 8195 as well, and then, yeah, the cool thing about 8183 is that, essentially, like, as, also, Lof said, is, like, very minimal, and, like, then it allows extensibility via hooks. So, like, I'm hoping to see a lot of experimentation there. For example, like, there is no concept of an auction in 8183, but You could deploy a hook that, like, basically you auction off, the right to serve a job or request. And then, like, we can also do more to integrate, like, this new protocol with X402, and basically with the other pieces of the stack, and that's also something we are working on.

Yeah, so in terms of, like, what to build, right? Like, I was just thinking about this, like, before the meeting. And, like, the first one, we've been talking about it a lot, is, like, new data APIs. The way I justify this is that, essentially, we need to create the hummus for, like, the permissionless agents to grow from here, right? So, like, if you were listening to Eric's talk, it was really talking about this, Dynamic capabilities that, like, we are enabling, like, with, X402 and these permissionless services for agents, but a lot of these capabilities, like, the model knows how to do a lot, right? A lot of the capabilities come from, like, new data and, like, new inputs that the model wants to access, right? This is also, like, in an ad stock, like, kind of the reason why, they believe that, like.

there is this distributional intelligence network that is going to emerge, right? Because there is data silos, because there is, like, physics constraints, privacy constraints. So, like, essentially, like, if we have more data sources, I feel that, like, basically, it's almost we are building the primordial soup for, like, this agent economy to explode. Then, like, the second important point is, like, useful agents, right? Like, we need to start somewhere before we have, like, an entire economy, an entire agent network. And, I feel that, like, here as well, like, it's cool to see all the experimentation in the past few months, actually, but we should continue, right? And in my mind, there is basically two principles, right?

like, we are using, like, blockchains and blockchain protocols, so, like, one direction to think about is, like, what are the protocols or assets that are there today that I can leverage with, like, the agent or, like, the service I'm building? And the second one, as was allotted in the presentation just before me, is, like, you should really visualize the exponential increase of capabilities in AI, right? Like, these are, like, exponential laws that have been going on for, like. almost a decade now in, like, scaling capabilities of neural networks. Like, we're not inventing them, right? Like, they're not, like, we're not dreaming of them. And, like, unless there is, like, some systemic shift, like, this will continue over the next few years. So, like, think today of, okay, what is the opportunities that the next model release open?

Or, like, as soon as the model is released, like, try it out, like, push the boundary and see, like, really what you can do. And I think from there, like, successful, like, agent prototypes and experiments are gonna come, and we should be, as this community, I think we should be the pioneers in that space as well. And then the third and important areas is, like, I think with, like, the open claw, moment, something, like, became clear, that, like, now, like, we are close… closer to autonomy than we were before, and, like, people could see, like, these, I call them, like, agent networks, right? So, basically, MoldBook is a product, like Facebook, but is, like, tailored to, like, agents instead of humans, right? There was MaltHunt, there were, like, kind of, like, a dozen, or a few dozens of these networks. I think most of these had one commonality that, like, there was no… mostly no value exchange.

there were a few experiments where, like, they were more like commerce networks, but I feel that, like, the stock we are building is particularly useful, like, whenever there is value exchange, and I think value exchange is going to be in all of these networks eventually, but, like, I feel that one message is build new agent networks. Like, if you're not building agents. build agent networks, which are also products. They are not, like, kind of, like, just infra that extends the core protocol, but it's, like, a product that, brings something, out of, like, these new agents that are coming online. And then the final and very important piece is, like. we should have, like, kind of decentralization, privacy, security, all these core values that we want out of, like, this AI economy embedded into the protocols we build, right? Because, like, that's the only way we can, like, deterministically enforce these properties, and that's also the only way how we can set the example.

Right? That, okay, like, when I tell you to, like, build a coordination, maybe in the future, like, these coordination protocols will be built, like, on the fly, like, for a small group of agents, or, like, a large group, but, like, let's say a million agents out of, like, the trillions and quadillions that are going to be there to coordinate. But, Maybe they just generate it on the fly when they need them, and basically, we should set the values, almost like the constitutions of decentralized protocols, is that you build them like this, like, for decentralization, privacy, security, basically instilling those values into, like, what's coming next. Yeah, and then this is my last slide, like, my fourth slide. I'm just, like, summarizing, like, what I was saying before, like, the focus is going to be on reputation, on the TE key registry, in the next two months. Then, we may have some incentives for, like, some community project public goods that, we think need to be there, are not there yet, perhaps.

And then, We want, like, deeper coupling and seamless integration with, like, X402, 8183, and the other standards that are emerging. Yeah, and thank you, that's it. Thanks, everyone, for staying so long, and for celebrating the Genesis Month with us.

vitto: Thank you, David, that was fantastic. Yes, go on 8004.org, as someone was asking in the chat, we have another event planned. We don't have a date yet, but the next event is going to be specifically about reputation. It should happen next month. Make sure to join the Telegram group, take a look on 2004.org. If you join today's event, you're already in our newsletter, so you will probably receive an email with the next date of the event. If you have any questions, of course, feel free to tag us on the Telegram group. Marco, David, I, we're always there. And again, thanks a lot to everyone that presented today, keynote speakers, Davide, Marco, AG0, Team, and so on and so forth.

Really amazing. I appreciate it. And have a great rest of your day. Oh, also, this is recorded, so we will share the link.

Davide Crapis: Thanks, everyone. See you soon.

Transcript adapted from ethereum/pm, CC BY-SA 3.0.