Agentic AI - Master ERC-8004

Transcript

So that's fine. Um and so yeah we will start in like three four minutes I guess something like this. Um just the time that um couple of people come here. So people here um basically um probably will uh will ask some question. Uh in fact when I'm live streaming some live streaming something I always have like zoom where people can interact and then also live stream on X so people can see can see it too. So all right let me just start that uh boom boom boom

up up. So you finally didn't go to to Denver. >> I did not. Um too many things working on I will probably be in ECC though at the end of the the month of March. >> And there is also stablecoin summit there that is more like security rented. I'm closing all the windows in the meantime so I can actually share my screen. Um there is this yeah security first small event before ECC. I will be there. But as always, I'm trying to limit my my conferences as much as possible. Otherwise, I would just do that.

>> Yeah, I can imagine. I I mean, right now, people um well, like sitter into the custom, let's say that's uh avoiding to to go too much into conferences just to stay productive, you know. >> Yeah. I mean, I I love conferences. I love the the vibes of conferences. I just do think if you actually have stuff to do uh it's much more productive to just get it done instead of going to conferences. I remember like 2021 2022 I was going to all the conferences. It was >> it was wild.

>> It was the same for me but that's right now um more chilling. >> Yeah. I I will go to Ital um and then uh India for Defcon basically. >> Oh yeah, that's for sure. >> Yeah. So let me check if all good on X. Yes, all good on X. So I will tell I will tell you when I will change the window uh to to um to see your actually your your face and and your screen. So wait, I will also send you the live stream on Telegram

would be great. Tuck t and we are ready to go in two minutes. Uh >> is it a rush day for you? Busy day again? >> Uh no, not too much. I'm working on a couple of things, but I would say it's pretty chill. >> All right. Always. Yeah, I saw you. You were doing a robo. >> Yeah, that was my project for this weekend. It's actually it's a great learning experience, not going to lie.

Never been too much into robotics. Started getting into it this year. Um, and it's really interesting, especially with this AI agents. >> Yeah. Yeah. Yeah. I can imagine. I can imagine. So, all right. Uh I will start streaming an uh X with um Zoom basically. >> So one, two, three. Let's go. Oops. Up tech. It's like all good.

Uh up. I will just put your um speaker. Well, no, it's it's fine. Okay. So we can we can go. >> Awesome. So >> basically yeah. Hi everyone. So pretty excited to have VTO with like us actually like um VTO is one of the main educator as Kevin on the space especially for people coming from web tool. So really excited to have him there. Um and maybe it should be nice like to to share your inspiring journey first of all I guess uh because I mean it's something that can be really

relevant to share with web two developers that are there uh actually so yeah feel free to share a bit about your journey uh how did you start into web three what did you do uh and what's your role right now >> yeah 100% I hope you have some time because it's it's quite a story uh it started in uh 2019 so Before Web 3, I wasn't in in tech really. I was working in the movie industry. I was doing animation for for games and and movies. And one of my best friends, uh, Jack, big shout out to Jack, he was into the polka dot ecosystem. Uh, which is which is fun because he's literally the only person that I know which was super obsessed about polka dot. And for years, actually, my bad that I didn't listen to

him earlier. He was like, "Oh, you should look into web free. You should look into smart contract. You should look into all the the possibilities that they open." And uh but you need to know Solen. How do you pronounce it? Is Solen? I feel like there is some accent there. >> Um there is like Solen in French. >> But uh people say Sen in English and Solen in Spanish. Yeah. >> Love it. Okay. Solen. Um you have to know that my family is mostly like banking people. They they all work in banks. Uh so the first time I told them about web 3 and blockchain you can imagine what was the reaction right it was like you know it's bubble it's never going to work and it was super early so fine so I delayed my entrance and then around covid right before covid I

actually started looking into that I had some extra time and and I got bored with my previous job and and I developed my first smart contract and I was like flabbergasted mind blown and that's when I also opened my my Twitter account because one big problem that we were having is that all of the tutorials and resources you could find at that time were like uber technical and I wasn't technical at all. So I was like okay let me translate what I understand into like human readable content so that other people can actually do this but also I can remember what I'm doing. So I I started publishing stuff on on Twitter which worked decently well I would say and then I got reached out by the alchemy team. Uh at that point Alchemy was very small. It was like 13 people and together with them there was also

chain link. Also chain link reached out and I went through you know the standard interview. Actually Alchemy was like hey do you want to collaborate and I was like hey actually I want to apply. So I applied for a developer relations position there. uh we can get into what the deils are which is a very interesting concept but anyway uh went through the interview process with both alchemy and chainlink ended up at alchemy for two years where I was working on their developer relations team alchemy became like a gigantic company and then uh one day I was at the conference and I met PC Patrick Collins I call him PC I think everyone in the circle calls him PC at this point and you probably know PC he's a big educator he does this very following courses. Uh,

everyone I think learned from him and and we were talking about security and I heard that he was looking for developer relations people. I was like PPC what if like we're friends what if I join. I was like are you sure you want to join? Like it's a very small company. I don't know where this is going to go. I was like no I love security. I mean security is one of the biggest problems that we have in the space. Billions are stolen every year. So I definitely wanted to to make an impact there. So I I joined Cipherin, ended up heading products at Cipher, uh spinning updraft, which is a big educational platform. I use the experience that I got from Alchem University at alchemy of course. And then uh after two years plus then I met David. It's all about meeting people,

right? It's a very small ecosystem. And David is the head of AI at the Ethereum Foundation. and he started telling me about 2004 and agentic economy and how all these things work and how the the blockchain has all the primitives that are needed by this economy and I was like oh god okay this is another thing that is inevitable I definitely want to make an impact there so we started talking started contributing and ended up joining the team at the foundation >> nice yeah pretty cool journey and and um to to let you know that some people into defac like everyone um starting pretty well with siphon like everyone is involved into it like with siphon obj to

to just yes learn a lot of things through through it so thanks a lot for making this uh this resource like available for everyone very very important so so yeah um and also yeah as you mentioned it's all about meeting people every time and start contributing so uh in in term of starting contributing you who will make what especially like open source contribution >> you mean uh Ethereum foundation or in general.

>> Yeah. Yeah. Yeah. For the return foundation. >> Yeah. So it's really about like share a what I usually do is share what I learn. I think that's the lowest hanging fruit, right? Like you end up spending time learning things. I was learning about 2004 and it was like yo more people should know about 2004. This is fantastic. and and always when you learn like early things uh they usually only have like the specs on Ethereum magician so this very like technical heavy documents so I started sharing about that on social media I wrote a couple of articles I built a open-source kind of like scaffolding tool that is called create trustless agent and now it's called actually created 8004 agent there is a funny story about that change of

name um so it's mostly really like what do you think is missing there uh at that point once you notice what's missing you have two choices or you complain that it's missing or you build it. Uh, I like building a lot. So, I usually build it and and people appreciate that because people like when other people build. So, yeah, mostly open source talking to people, connecting and sharing on social media whatever I'm learning. >> That's very very cool. Yes, the Italian builder vibes. Let's say >> sort of also big shout out to I have the hood right now. Uh uh and um to to talk a bit about the web three stage actually like we can see a lot of AI um a lot of

communities right now starting to to go more into AI. So how do you feel that web 3 um will stay uh alive uh with uh with AI actually like how how we can um work with both world uh effectively? >> That's a great question. I'm going probably to take a step back here because I think I'm writing an article about this uh I'm writing a couple of articles right now actually. Um but this one is specifically about us kind of like isolating ourselves. we have this concept of web 3 uh which is isolating by itself right because it's like web three against the world uh you used to see this like posts on Twitter be like

web three versus web two and this is doing more damage than good I would say tends to isolate ourselves it's not like web 3 and AI is just like the word the tech world right so uh lots of times is like how can AI for example be used with web free technologies and should all AI I people care about web free technologies or should all web free people care about AI technologies? The answer is always it depends. Uh sometimes technologies are useful, sometimes are not. Sometimes my SQL is is useful. Sometimes GraphQL is useful, right? So it really depends on on what you want to use. So I would start stopping I would honestly like remove the term web free completely and it's

just like okay we have these block blockchain technologies uh let's use them wherever it's needed. Now AI enables a lot of things right like not just with web three in general like we've seen that uh speed of execution building and so on and so forth. Um but when it comes to agent agents and agentic economy there is lots of things that AI by itself cannot do and it's other primitives that cannot be offered by by AI that can only be offered from by the blockchain. And these primitives are like trust, right? Like imagine a world where you have agents communicating to each other. Uh how and we're going to get into that because we're going to talk about 2004. A how do agents find each other? What's the directory that allows these agents to

find each other? The other one is how do agents trust each other? How do I make sure that this agent is a reputable agent and is not like someone or something or an agent that wants to scam? Um so there is a lot of questions that we need to answer when it comes to agentic economy. And then I would say the third component which is in a way opinionable. Uh but we're going to talk about that is payments. Everyone is saying oh agents need payments to to transact. Sorry need blockchain to transact. Um the answer there is it depends. Uh it's not that they only they can use multiple things not only blockchain. I would say the blockchain is the easiest to use when it comes to like permissionless payments for sure.

Um so yes uh will web 3 become bigger because of AI? Most probably yes. Uh will web 3 as we're calling it only survive because of no it's inevitable and we can get into that. There is other points of views on why blockchain is necessary. Yes, totally. Like we can say that uh actually web three um is good in term of um like providing infrastructure for just making things transparent and all the things but when it's come to mass adoption like we we have to to know how to really deal with the web to world and so AI is a is a good thing for that.

Yeah. Um could you tell more maybe about uh the agentic economy actually like in in term of numbers um like basically I saw that base is probably the home of AI um actually so do you have any kind of cool numbers to showcase just to see um at which point is big to go deep dive into it. >> Yes. Let me let me see if I can share my screen and if you can actually see it. Can you see my screen? >> Perfect. Fantastic. Let's see if I can also share something more here. I cannot see my screen shared. So I will just go with it and you tell me what you're seeing. So I'm also going to on 804 scan. Can you see 804 scan?

>> Yeah. >> Fantastic. So before getting into these numbers which which are nice. Um, agentic economy we intend everything that uh includes agents talking and transacting to each other but not only agents also the agentic services and I will use this term a lot throughout this kind of presentation this chat. Aentic services are all the primitives that we are used to call like APIs, MCP servers, A2A servers like anything that enables agents to perform an action or agents to get some data that they wouldn't have natively because of their training material and whatnot. Um so to make this agentic economy work you need the following components right you need

agents because otherwise there is not aic in this economy. uh you need data providers so you need APIs, you need MCP servers and you need trust as we were saying right you need the way and and of course communications interoperability um so four components agents is easy we all know what it means right is the chbt powered agent is the anthropic powered agent is the lang chain powered agent so is any kind of like LLM that can not that doesn't only like answer your questions but it can also perform actions more or less autonomously ly and then you have these uh data providers as we said APIs, MCPs, A2A servers, you have trust big question mark in the world like how do you again how do

agents trust each other and then you have the the payments of course how and and interoperability so how do communi do they communicate um with each other now to solve the trust and interoperability issue and I think that's where we want to go um we created this thing David and Marco from MetaMask David from Ethereum Foundation Jordan from Google and Eric from Coinbase, they came up with this very fascinating idea that is 804. You probably heard about 804 is called the trace trustless agent uh protocol standard. And essentially what this does, it comes up with three registries.

Uh one is identity. So every agent gets an identity. You can see it as like a 721. It gives actually let me let me change here. This is if you want to learn more about 104 there is a 1004.org where we have this amazing learn section where all the questions you might have are answered. So you have this identity registry where all the agents are registered and they get their identity which is then connected to a reputation registry where all the the feedback related to that agent are stored. Now, the way I usually explain feedback is like Yelp stars. Uh, but it's a little bit more deep than that. I would say it's more similar to how if you're familiar with SEO, Google ranks pages on the search engine results page. So, it

can be latency, it can be success rate, it can be price, it can be uh, of course the the historical performances. It can be multiple different things that then get aggregator aggregated. And then you have one last registry which is the validation registry which kind of allows I'm not going to get too ind depth into this because it's not out yet and things might change a little bit so I don't want to to give you wrong information but essentially allows uh validator networks to validate the operations of these agents. stupid example is think when you're on chipt and you're writing to judg 5.3 um you can only trust the UI that you're actually using that model but behind the scenes you don't know if they're using I don't know 3.0 or 4.3 or whatever,

right? Uh which is not a problem if you're just like talking to an LLM about what you want to eat for dinner is more of a problem if you're dealing with LLMs because you are performing high risk financial interactions. So the validation registry allows you to prove how the agent executed and then compares the output and tells you okay the execution is right now the execution is not good. Now it's a little bit more complex than that but this kind of gives you an idea. Now when it comes to number um it depends what I'm track what we're tracking here. This is 804 scan which is not from the Ethereum Foundation. It's from Alta layer. Shout out to them because they're doing an amazing job. We can already see a few cool numbers. So 804 was released less than a month ago

and within the last month we got almost 70k registered agents. Now the number you're seeing here is actually agents that are fully compliant. So agents that are properly registered potentially no spam. Um potentially because there is still a little bit of spam that we're trying we're fixing. We're working with different teams to make sure that we have better feedback systems and better infrastructure to understand what is a proper agent and what is not feedback. So this exchange of valu evaluations more than 32,000 and active users. We're talking about around 40k again filter number here. So this kind of gives you an overview where we're going. Uh we have thousands of agents being created even more after openlow became a thing

because everyone was creating agents. Um a lot of these agents are not even registered on 2004 yet because 2004 is so new and they they don't even know about 804, right? Um they started transacting. They've started posting on social media. They have their own social media now. They started selling stuff. They started selling information, selling jobs, selling labor. Um it's not mainstream yet. is super new, you know, like on Twitter, we're like on a bubble where we think everyone is adopting something, whereas it's just like us and another 2,000 people. Uh, but it's inevitable, right? Tomorrow, we know that agents are going to to be like 90% of the internet users are going to be like 99% there is this cool interview

with Elon Musk where he's like 99% of intelligence will be synthetic will be agents. uh and these agents will buy things, they will do procurement, they will exchange stuff. So yeah, this is the overview of the agentic economy right now. Hopefully that answers your question. >> Yeah. Yeah, totally. Um and so that's why like it's very important to have a standard because as the agentic economy is starting to booming uh actually if there is no standard there's like some how we can say that like some things operating um um not together. And so that's why you you wearing that, right?

>> 100%. Standard and bureaucracy is low. >> We I mean of course we know it in Europe, but it's like that in everywhere, right? AI is super fast. Bureaucracy is very slow. You need to find a way to control what's doable, what is not, what can be done, what is not. And these standards help. >> Yeah, totally. I can make that makes sense. um and how we can um register actually an agent just to to showcase the process. I know it's like friendly but just to showcase the process.

>> Yeah. So let me see if I can actually do it through here. So there is multiple ways. Let's maybe this will allow me to do it. Oh no, I cannot. Let me stop sharing a second because otherwise my wallet won't open. Of course, guys is a is a lifetime since I use Zoom. How do I stop sharing here? Stop share. Okay, fantastic. So, there is multiple ways. Uh again, if you go on 804.org and you go under the build page, I don't know Sen if you want to share this link. I think it's kind of important for the chats. Um, essentially on this link that I just shared here in the chat with you, there is mostly all

the ways you have to register an agent. Let's see if I can actually make it work here. >> I will send it on X after. >> Fantastic. Can I switch my No, it's not letting me. Okay, we're going to wing it. >> You have any question on the chat? >> Okay, we're going to wing it. Uh I will share maybe another one trust here. Maybe it's easier. Okay. Awesome. Yes, it's easier because I don't need to connect and my wallet for some reason is not opening. So you have different ways. Uh the the easiest

I would say is the no code solution, right? So let's say that you already have an agent uh and your agent has nothing to do with a 104. You're already implementing classing MCP. Sorry, taking a step back again. And I want to make sure that everyone understands it. I have an article that is you're reading 804 wrong because I think we made a mistake in naming the standard legit because we named it trustless agents which is great for marketing. It's fantastic. Like everyone was like, "Oh my god, trustless agents." No, the problem is that it's not only for agents. Agents are literally like a small component of what 804 is for, right? Is for agents and is for agentic services. So if there is no if there's only aentic services it's not going to work because there is no one using uh

agentic services if there's only agents it's not going to work because agents have nothing to do because they are just agents and they don't interact with anything. So whether you have an API MCP data layer or an agent you want to get it registered on 804 because you want it to be discovered you want it to be to build a reputation you want to use other agents and be part of this economy. So let's say you have an API then or agent then you can go on websites like trust 804 or 804 scan if your wallet opens and then you can just go here and type the name of your of your agent your agent address which is the wallet of your agent or where you want the NFT to be stored the description the version and

so on and so forth. I need to fill this in. And then it will ask you a bunch of information like the end points. Not going to go through this uh the skills and capabilities uh and then another couple of questions like domain of the the thing review and then it will register it. Very straightforward. You just need your MCP endpoint or A2A endpoint. There is other ways though. If you want to go a little bit more in depth, you can see it here. Um these are all the websites that I'm talking about is supported on mostly all chains. then you can go with the code solutions and for this there is a very very thriving I would say um ecosystem of builders building right now on 2004 one of the most used is agent zero of course

because it's built by Marco de Rossi which is one of the co-authors of course of the of the standard and this agent zero allows you there is also nice playground that they just um released which works really really well essentially has everything you need to register search find evaluate create other agents. It's very well done. Uh it's very easy to use. You can follow the code here. You can just like give it to code and will be easily able to to replicate it and use it. And through this you can do the same thing that we've seen on the UI. So you can insert the name, you can meet your registration file. You can list all the skills that you need and you register it and as soon as you register it, you get the NFT related to your agent and then then your

agent is enabled to receive feedback. And then there is the very lowlevel. Well, there is also another way. This is a package that I created uh and probably is in my opinion the first place you should go to if you want to test uh 804 because it's not complete is definitely not for production ready environment but it gives you an idea of how 804 agents are built. So just a line of code you write npx create 104 agent in your terminal and then it asks you a couple of questions. I can actually show you here. Can you see only my browser or can you see the browser not the terminal?

>> Okay, let me share >> that's where you go through the process. >> Yeah, there we go. Can you see this? >> Yeah. >> Fantastic. So, what you would have to do is just write create MPX create 2004 agent. >> You can zoom it. >> Oh, yeah. Absolutely. better. >> Yeah, actually let me then do this. Can you see this?

>> Uh I saw MX. >> Yeah. Okay, let me go there. Okay, you can see is zoomed in enough. >> Um not really like >> Okay, I can go bigger. >> Maybe just for the stream. Yeah. Yeah. Perfect. >> Perfect. Oh, okay. All right. There's a big screen. Uh, MPX create a 1004 agent. This is everything you have to do. If you have MPM installed on your on your device, that that should work. And then it will ask you to install. Well, actually ask me to install it here. Um, because there is a new version. So, you just install it and and this is the process, right? First and foremost, it will ask you where do you want to create the the agent? In this case, we're going to do my agent. It's perfect. It's going

to create a new directory for me. How do I want to call this agent? Uh, we're going to call it Solen. Awesome. And then the description. Uh, I teach Y3 people. And this is going to be your description, your global description for your agent. Then an agent image URL. We don't have it. We're not going to use it. You're just going to use a example. And then it will ask you on which chain you want to to use it and which chain you want to register it. Now the important thing here is that it doesn't really matter which chain you you use because at the end of the day all these scanners all the aggregators are already multi-chain. So the only reason why you should choose a chain rather than another one is a personal

preference b is for the ecosystem and products built on top of that. If there's something that you you like more than something else then you should use that. In this case, we're going to use Ethereum Tapolia, which is faster. Agent wallet address, which we've seen in the UI before as well. Uh, is going to be your agent wallet. If you leave it empty on this package, it's just going to create a wallet for you and then it will put the private key inside. Now, don't kill me like is a MV in plain text. I know this is not good, but it's great for for examples and to showcase the functionality here. And then it's going to ask you ask you if you want some of these things like if you want an A2A server, if you want an MCP server created, if you want X42 payments, the

synergy with the X42 is clear, right? Because you want agents to pay each other. Now, in this case, I chose Ethereum Seolia. Unfortunately, yet there is no facilitator on Ethereum Seolia. We're working with Pay AI to to make it supported. But the only chains where is supported is Polygon and Bass. Now I think next week we should already have it on Ethereum mainet and sepoleia. Anyway, I will select these two which is totally fine. Enable A2A streaming responses. If you're not familiar with A2A is agent to agent essentially is a way for agents to communicate to each other and it supports this thing that is streaming. So you can just stream uh whatever the A2A server is telling you.

In this case we don't need it. You mostly never need it. And then the supported trust models. So what we were talking about before when it comes to validation is it only like a reputation based validation is through like staking so people can stake to validate what you're doing or the at testation which is not live yet but the standard is there. Uh we're just going to select reputation for now as you can see here what it's doing is essentially if it doesn't freeze for whatever reason because I have so many things open it will generate a new wallet for you and it will create your C804 agent. Oh, here we go. So, then you have a bunch of next steps here like navigate to your agent, uh, back up your wallet. You can see the

private key is int file, configure the TMV, find your wallet, and so on and so forth. You can just follow this step by step and to register your agent. Now, let's see what it created. So, I'm going to open the folder here. Let's see. Let's see if it's opening it. Awesome. >> So we can see that the um 804 um standard is like very complement uh with A2A uh MCP and X402 protocol itself. Yeah.

>> Oh 100%. So the standard is very lightweight like is actually complimentary with pretty much all the standards out there. And if you look at the registration file actually let me just open it. So this is the agent that we just created. Um the registration file. Well, this is not the best way to look at a registration file. Let me just open it for spec. If you look at the C804 spec here, there is an Ethereum improvement proposal. You scroll down a little bit. This is what the registration file is, which essentially is your the identity of your agent. Now what you see here is of course what we seen the type which is always the same the name the description the image all the things that we just set up on our CLI there is this array

that is services now the array is completely free like you can put whatever you want these are just examples there could be a dot dot dot at the end to tell you hey you can put whatever you want but as you can see here you you literally can have any type of service you have ENS you have decentralized digital identities you have the email you can have just an endpoint HTTP you have a website. Uh so literally all for all the for all the services and standards that are available today and potentially all the standards that are going to be available tomorrow. This is not going to change. Uh it's completely flexible. So yeah is is literally just identity. Then how they communicate to each other is is really up to the service array here.

So going back to this let me see if I can zoom in the UI. Okay. Awesome. So once we have created our agent, this is what it's going to create, right? It's going to create the A2A server. I'm not going to go through the code, but it's a standard A2A server that allows an agent or a HTTP request to request something and then it calls your LLM to do whatever. You have the the server, sorry, this is the client that performs the request. You have the server that allows you to perform to process the request. Then you have a very simple OpenAI LLM. uh if you were to create an actual like service you would edit this everything is commented out so you can decide and you can understand how to to edit this code. You have an MCP server here. uh you have tools related to the

MCP server and then you have this script which is probably the most important which is using as I was showing before agent zero SDK and what this does is essentially asking you for all the information that it will need to then generate this registration file. So for example here you can see I mean there is the chain there is a private key there is the IPFS address on where you want to a registration file uh you have the set 8way which essentially is going to create this field inside services you have the set MCP which is going to create this field inside services you have I think that's pretty much it of course you have the configuration

name description image name description image here. So all of this is going to create the JSON file that you need to register then on IPFS which is going to then be used as a token URI for the 721 that is going to be the the identity NFT. So instead of having an image and the NFD you have the registration file and that's really it. So if you're using this tool then what you would do is just like no I would have to yeah trust the folder of course uh you would just run mpm oops mpm run register you can see on the package json there is register which is going to call exactly the file that I

just showed you and what this is going to do is register on IPFS wait that the IPFS C is uh is minted um set the wallet, wait minted and then register it here and that's it. You have your wallet registered. >> That's amazing. Um and so actually like um to build like a reputation for the agents um all the process work in term of collecting feedback like rating the agent itself. >> Sorry say that again. writing right the agents >> like to build their the reputation like um how did you collect feedback?

>> That's a great question. Um so the way you do that and for this I'm going to open back 2004 because it has a good number of examples there. So once you have register your agent then because it's open right your agent is going to be automatically listed on these platforms uh on all of them. So 804 scan is going to be on trust 804 and so on and so forth. So as you can see here these are all the agents that have been registered as of now. The last one is 8 minutes ago. Um now once it's here uh it means that it's also on the on the identity registry. So anyone can find it and what you will see is that you will start getting pinged by other agents that are looking for you and

potentially they want to use you. Now let's say that an agent uses you. Then what you can do is issue a feedback request to that agent. That's fully automated. And then if the agent leaves a feedback which is done through a function that is also on agent zero, you can directly call the contract. >> Could you? >> Yes. >> Uh just perfect. >> Awesome. >> Um you can see here like for example this agent on monad has 7,000 feedback. Uh this one agent zero has 10,00 feedback. So if we open this and again here you can see the chain name description you can see the services here uh this is a better UI and here you can see the feedback. So actually

probably better to show them here. Here you will see there is a score uh which goes from one to from zero to 100 and has two decimal places for like latency and stuff. So for example this one did a decent job not so good. Uh there's also a comment here. Uh this one is the agent actually saying uh this is a like fake review. Uh you might burn be burning token because of this um fake reviews like this is again like so that's the thing we have a new watch tower which is super interesting that is going to flag fake reviews automatically. Um because this is the biggest problem problem what we seen like two days after we released the standard there were already people

selling feedback we're like hey if you want you can send me like two USDC and I will give you a feedback to to your agent to make it rank. So this automatically is flagging some of these fake reviews so then feedback aggregators can actually um do something with it. So to answer your question, yeah, you have your agent here or in the directory. Agents that already implementing 804 can look through all the agents, find the services, use your agent, and then what usually happens is that most of the feedback, the real feedback are only left when your agent does a poor job, which is just like the real world, right? When do you leave a review on on like a business? But most of the times is if you're very pissed off with that business and you're like, "Oh, they didn't do their job properly."

which is pretty much the same here. Um what we expect is that lots of this feedback will come from agents. A good chunk of this feedback the more trustworthy I would say will come from these watchtowers which are like public good services. They're just scanning the directories and pinging all of these agents to check if they're up actually online, what's the latency and stuff and leave feedback related to that. and feedback givers, which are like trusted parties that are reviewing the the agents. Now, once we have a good first layer of agents that have feedback, then it's easier because you can evaluate the value of a feedback based on who gave the feedback, right? Is this feedback given by a reputable agent? Yes. So,

this feedback is legit. If not, then probably spam. It's going to be a process. Will take some time. >> Yeah. Yeah. I understand there is like real work to do around that like the feedback to make sure everything is through like real one. That's funny to see that finally agent will become the end user not really uh just the human at all. Um and so if we can come right now into more um ID build building ID around that like how do you have like a top three ID you you would love to see emerge from this identic economy?

>> Yeah 100%. So there is one that I'm telling literally everyone that asks me this question because it's one of those ideas that that's like the killer use case for 804 and if no one builds it at a certain point I already said I'm going to build it myself because essentially right now when you want to build a agent you need multiple services. Let's say that your agent I don't know can do 3D modeling can um get weather states like see how's the weather and anything else right? But what you need is to implement these services natively. So your agent would need to have a tool to do 3D modeling, a tool to get the weather state, a tool to get the market state and all these things which it doesn't

make sense because a there might be then another tool that is better in like two weeks and what you would have to do is change the code of your agent which is wild is not scalable at all. with a 104 because you have a central directory of all these services you just need one service one tool which is a routing tool and essentially how that would work is you ask your agent hey get me the price of oranges and your agent calls the routing service and saying price of oranges the routing service scans the entire 2004 directory it checks all the feedback and sees what's the best service to give you orange prices for that time at that price at that latency and just gives you the orange prices and

then in another moment you ask do me a 3D model of a dog. Okay, awesome. Your agent asks the same routing service, give me a service that does 3D modeling. The routing service scans the entire 84. It checks for all the agents or agentic services that do 3D models and gets you the best service based on the feedback for that price in that time. So, routing services bomb. Like literally guys, this is the most killer application and no m like it's not because it's built on blockchain but it's some is a primitive that we need in the agentic economy. Is not feasible that you have to implement all these services natively. Um other things come more from a public good perspective. So more a big problem is

feedback. You you probably like if you have any question one of the feed the question that we usually get is how do you ensure the feedback quality is high? How do you ensure that is civil resistant? Do you sure that is no like fake feedback? Well, is there's no way to actually ensure that right? If you think about Google, if you think about Yelp, if you think about G2, a they have fake feedback like most of the times and even them like the situation there is that this is a central service G2, Google and so on and so forth. They have a very easy time um kind of filtering out feedback, right? they they are like central platform so they can say oh this is a

fake feedback this is not a fake feedback and I may review this feedback because it's saying something wrong about this business and so on and so forth uh and even in that case they have a very hard time evaluating what is true and what is not here is even harder because it's fully decentralized everyone can leave feedback no one is checking the feedback so we need services that are reliable to make sure that they can a analyze the feedback um b make sure that the feedback is reliable and real is not spam and Then also aggregate this feedback in a way that makes sense like waiting all different feedback provenence is important and so on and so forth. So watch towers routers um I would say the third one is not really an idea but is is like how 804 should be used for the

majority of time again I have this article here that I invite literally everyone to read that is you're reading 804 wrong. Um this one kind of explains the fact that everyone was reading a 104 as an agent only standard which is not is agents and aogentic services. Uh it's it's literally like bring any type of data layer on a 2004. There is so much business opportunity there. There's the opportunity to make a ton of money to be honest on this because you can bring like any service that gives you market rating, market prices, it gives you weather even or allows you to purchase things off of Amazon like a Amazon CLI or literally anything that doesn't allow

you to to do like browser automation. You put an X42 pricing fee on top of it and people will love it. So I think these two things plus one two things that are very concrete. The third one is literally is such a a fertile field right now that literally any idea is good to test and potentially very reminisce cases are very interesting to deep dive into it. Um I mean like all things related to writing um or things related to how to hire the best one for a specific task. Yeah, there's a lot of things that you can you can come up

with. Uh I I I can't wait to see what what what what's going to be built during the the next Aaton because I also saw that um some Aaton is just like um agent there is no human. So it's funny to see and I would love to see what's what's uh going to build by agent itself. Yeah. All right. All right. And um I have a question more oriented on the security side because you also have um like a background around that. Um I know that you you you built something with lemon uh with gardens called signing with agents. Could you tell more about it? Um why finally you do you

introduce uh this uh this thing? >> Yeah. So talking about security, security is also another big topic there that we should talk about and it's mostly around education. They are educating people about the the dangers of agents and the dangers of having files like memory MD that open CL has which are literally just a psychological profile of your persona and invites like social engineering like crazy. um or or the fact that we're sharing secrets to our agents and agents. There is this amazing research from zero leaks uh that I shared this morning to a friend. I can share it to you. Maybe you want to amplify where essentially they're saying, "Hey, unless you're using oppus 4.6 six or tragic 5.3, you have like 94%

risk of getting prompt injected and revealing all of your secrets, which is wild if you think about that because it means that 94% of times you try to prompt inject it will be successful and your agent could potentially like leak all of your secrets. Um, and when I say secrets, I mean private keys, I mean API keys, I mean literally anything that your agent knows. Um, when it comes to ensuring that a user or entity navigating on your platform is an agent, that's an incredibly hard problem. Um, if you think about that, is already a problem to understand if a human is a human or a human is a bot, right? Uh, look on Twitter for example, where 90%

of the answers are AI generated and Nikita is every day like damning and cursing the fact that they cannot stop the bots. Uh but even LinkedIn you see any social media is full of bots nowadays. Um so starting from the preassumption that it's already very difficult to distinguish humans from bots. Think how difficult it is distinguishing bots from humans. Um like in one case you can kind of use things like capture and uh cloudflare and they're decent services that help or you can block your API unless you have JavaScript. So you need browser automation and pure scripts wouldn't work. Um we see it doesn't work 100% of the times trying to understand if an

agent is actually an agent and not a human mask as an agent is almost impossible. uh because you can even if you have a crazy good way you can just be a human that tells your agent what to do and they will do it for you right so mbook think multbook that is that's social media only for agents and people are like oh my god the humans are writing a moto book uh influencing what agents say we need to do something about it okay great you can do whatever you want but then as a human I can still tell my agent exactly what to write and it will do it for me so even if you have discerned agents in humans properly is still going to be a problem. Imagine when possible. Yeah. Yeah. Let me which one do you want to see Julia? The I mean

I'm going to share all of them. Uh we have this one that is 804 is you're reading 804 wrong and this one is about clot um security. But so what Siwa does what we've been trying to do with the with lemon is to move our first steps towards a world where we can actually discern the two and this gave you know birth to to this sign in with agent which is kind of a gimmick name from sign in with ethereum. What it does is combines two actually three ERC's 172 81 178 and 804. So 804 is the one that we talked about where you have these

free registries, identity and so on and so forth. Uh 8178 is a sign in with Ethereum sort of where you can sign API calls with your wallet and then prove that it's your wallet that is actually calling that API endpoint and then 179 172 is the one related to signing things is like sign payloads. So, how that works is that you can go and check it out. It's is actually very easy to to use. It's here. >> Could you send the the link?

>> Yeah, here you go. >> Because uh I was checking on my side. >> Yeah. So this is a sign in with Ethereum. Builder's garden is this team of builders that are working with us to gave birth to it. And it essentially does let me see if I can find that amazing schema they created. Oh, here you go. So you have an agent, you have a service, and of course you have the blockchain. So what the agent does, it calls, it works just like signing with Ethereum. It calls the service. It receive a number that they can only be used once the the nons uh and then the nons is sent back and then what it does is signs the the nons with their wallet.

So every agent has a wallet. So it signs the nons and then it sends the the signature back. Then what the service does it verifies against the blockchain. What does it mean is essentially checking that the wallet that is signing that call is an 804 register agent and the reputation is high. So if that agent doesn't have like enough reputation potentially can be not allowed in uh in that platform in that service which is very nice if you think about it because if you think about like payments for agents for example um think if you want like a visa for agent then it want to know if that's a reputable agent or not because you probably don't want your like non-reputable agents to use your service. So it does this check and then it sends back a verification request.

Now at this point received at this point the agent is authenticated. So it has kind of a JWT um JSON web JSON web token that enables the agent to make API calls to the service. But these API calls still need to be ERC at128 signed. And the only reason why we did that is because if you add a JWT like mbook for example, you can just steal that JWT and do calls on behalf of your agent. Whereas if you use 8128 then you still needed to sign that JWT with the wallet the agent uses. So potentially as a human you can still do that but it's much more difficult. Uh and then there is a response. So this is what sign in

with agent works. There is another component that we're adding right now. Now let me see if it's already added. Yes. And is sign in with capture. Now this is funny looking into that because essentially the guys that sign in with agent like had this idea was like okay we have capture for humans what if we have capture for bots and what is a capture for bots is something that a human cannot solve right like write me hiku in 5 seconds so I send you back a capture it's like okay write me a full poem in 5 seconds and if you go above 5 seconds we're not going to let you in.

An agent can do that assume that there is like decent latency a human cannot. Now the human cannot though again comes with like gigantic quotes because potentially you can as a human you can do everything right you have power over your agent but is another layer of security and check to make sure that what's signing in is actually an agent. Now that plus reputation should give you a decent enough kind of data point to understand if you should let something in or not. Um yeah this is signing with agent.

>> Yeah. uh that's like a good things uh into the ecosystem to to have that like in the game changer I would say um because you can control finally which agent um will enter into which things um and so the fact has the reputation layer into on top of that is really really smart uh congrats uh I I've been really impressed by the how you you were fast to deliver that like did you already work on that before uh like as a specific release or I mean um how how much time did you put for for doing that?

>> Yeah, I know the builder's garden guys are just super fast. So essentially how it went was we talked on a Thursday and Limo called me and he was like what should I've seen what should we build? I was like, we need two things. You guessed it. One is we need the routing system. Please, someone built the the routing system. And B, we need a way to discern humans from agents when signing in. I was like, "Okay, I got you." And then I think literally 24 hours after that, it was like, "Okay, we have an MVP." So he sent me the MVP. I reviewed the MVP. There were like a couple of security issues there, like 8128 wasn't implemented yet. They reviewed everything and then on Monday, they sent me another version. I was like, "Okay, this is great." Then Yakobo, Yakopo is the guy behind 8128 sent us a message.

He was like, "Hey guys, I'm releasing 8128." And we're like, "Okay, no way. This is perfect. Let's implement 8128." 24 hours was Tuesday. I checked it worked. Wednesday we released. >> Very very uh efficient. Yeah. And they are still building stuff. Anyway, there's a question uh for you into the chat if you want to to answer it. Um and after yes we will we will close. I have like one more question but um yeah feel free to to reply to the question of Moses.

>> Yeah. Are there resources that agents make use that might require being bought and solar is only API keys from other lamps that require to operate with resources agents may use it require B and solar is only API. So here's the thing like there is all sorts of things going on right now. There is like an agent creating 3D models. So the example that I was making before that's a real example that happens. There is agents building websites and PDF presentations. There is agents there is an agent from Ganesh from Aenlayer who's creating drawings I think I think and he made like a bunch of money recently. Uh

there is lots of agents of course related to DeFi and financial operations. Uh there is like the Zyi guys made this they're awesome by the way. Uh yeah also they're French and the the guys there have created an agent that allows you to adjust your Yel positions real time based on what's the most convenient thing. There is both I would say definitely the majority is APIs and API calls because that's very easy to make right and uh again very fertile field.

Thanks for all your insights. So I have a last question for you is like um I mean one or two um more for people who are curious like web to developers want to really uh start um deep diving into agentic economy. If you have like one resources to go through which one is it? >> That's a great question. I I don't think it's a resource. I think as a group >> it's complicated I mean to stay updated about everything every time. So having one one point where you can get informed about everything

>> I would say specifically about blockchain and 804 the 804 builders group on telegram where I'm I'm going to share the telegram group here. We have around 2,000 members now and every day there is a new PC. Every day there is someone asking amazing questions. Uh I will also share the ENS XAI group. Now I'm not affiliated. I I just I cannot but let me know if you want to to get invited. I think I need to ask the Oh no. Okay, here we go. This one. I'm not affiliated with with them. Um, but I'm just like impressed by the sheer amount of like brain power there is in that group. They

have a weekly session every single week where they discuss CNSX AI and there is two guys also like congratulations to K that is the admin because he got engaged today. But um there is these two guys that are Prem and Rich which I do legit think they spent like the entire days thinking about how to improve the agentic economy and they are year C producing machines. They're really amazing. So these two groups are definitely where you want to be. There is the AI uh website of the Ethereum Foundation which is handled by my team.

Um I I would suggest also to follow now selfplug me but just because I'm I'm being used as like amplification channel not because you should really follow me. So uh me on Twitter David the crappies on Twitter absolutely which is the head of AI at the Ethereum Foundation and I'm going to share his account here. Uh so he's sharing a bunch of things there and all the latest news. I think these are the main places where you want to be to be on top of things. >> Amazing. Yeah. And your account too as you mentioned like I I love reading all your article.

You are really specialized on that. That's cool. Um and so yeah, what's next for the DA team like your team uh at the term Foundation for this year? >> Yeah. I I don't know why we called it the AI. It's just AI. You can talk call it AI and that's totally fine. So one sec before that I think there was this question super quick uh by Rosio that is asking what's your advice for learning to build with vibe coding or agent while considering these security risks. Uh that's that's a great question. I don't know if I have an answer because cyber security is multiple things is different things that you need to put together. Uh

I would say good sense first and foremost like start from the preassion that if you share a information online uh especially to AI or whatever that this is vulnerable. So try to limit the amount of information your AI sees or can access without your permission even with your permission just can access full stop. Um there is a couple of things there is Shenon I will share the link if you need more links I will I will send them to you just send me a DM. Um, Shannon is like an automated tool that checks on the security of your project. Uh, follow a couple of accounts like chief I think the name is very not politically correct because like chief of autism but that's

a very new um account. Uh, and she or them or him are doing an amazing job here um like posting about security. uh great account really like super new great account went from zero to 12k followers in no time because wow like the the amount of information they're sharing is great and if you see their pin tweet is about this cloud code but for a king which is super interesting but yeah start with the pre assumption that everything you share with AI is at risk and uh that's going to make things much easier if you just vibe code your project what I usually do is use one LLM to write my project and another LM to review my project so maybe use cloud code to to write the project and then

codeex to review the project which kind of improves things uh is not going to make them secure. So for that you just need to manually go there. Um >> we have a session uh on on Friday about openlow. So I will ask to Austin all best practices around that as well. So if Rosio you want to come >> fantastic. Um talking about the AI team. Um so what we do is three things. One is AI, the other is AI and security, and the other one is AI and privacy. So this is literally the three things that we're working. 804 was like square one. Super excited of what's to come there. When it comes to AI and privacy, we're focusing a lot on fully

private uh inferences, which is a very difficult challenge to be honest. which means when you send a prompt to an AI, that prompt is encrypted and what the AI answers back is also encrypted. So no one sees what you're telling what you're talking to with the AI. Um there's things like fullyomorphic encryption involved there. There's this amazing project that is Zama Z a MA which are working on that very complex tees are also another component. Um, also about privacy, we're working on like private API calls which would anonymize like whoever is paying for for any service.

So that's that's definitely the main focus. And then there is the the section about security. So how do we make like interactions with agents a little bit more secure? About that is mostly research and setup and infrastructure. How do we make the infrastructure more secure? How do we use like wallets with agents without like leaking private keys? I think these three are the main things. We have a one big projects project that I cannot talk about yet because it's going to be massive and super fun that if works well we will announce in the next few months uh will with the European Space Agency. Uh so yes that's that's really really interesting. Um what else? Yeah, I think these three things will probably we have

also like if anyone here is interested if you go on our website that I shared before ai.etheum ethereum.jshed we have a section that is like work with us. We're not directly hiring right now but we are sponsoring lots of PhDs. So if you are into a PhD please let us know. Uh we're happy to sponsor it. We happy to mentor it and and help you out with that. And uh yeah we're talking to lots of universities. I think another part of our job is also making sure that not everyone thinks that crypto is only memecoins and monkey JPEGs which uh which is probably the most rewarding rewarding part but also the most difficult but yeah these are the the things we're working on right now.

>> Thanks for telling us uh everything. Yeah, looks exciting about your upcoming plan. Quite uh yeah I want to see that. All right. Um, so thanks a lot VTO for your time, for your insights, for all the education that you are providing for everyone. Um, very precious. Um, and um, yeah, I really appreciate. Um, thanks a lot. And, uh, yeah, I I will just, uh, stop the recording on X. Wait. Um, tuck t. Just wait one moment.

Um, Um and stop. Wait, I just want to end the Wait, I will stop here. Stop here. Stop here. Is it all good? No. I'm coming. Okay, fine. Right now. All right. Just a video on on X. So, yeah. Thanks again for for your time. Um like if you have anything to to add um feel free to let us know. Uh great station. Yeah, thanks

a lot people. Just thanks thank you uh for for it and yeah I mean um can't wait to see everything that's coming up on your side. Wish you also the best because it's not easy uh to do all these things. Actually you are how many to your team? three, four, >> uh, three people. >> Wow. Oh, yeah. >> Yeah, >> you need you need people. >> We need the help of the community. Absolutely. >> Yeah, that's true. Um, okay. Okay. So, I will I will take a look as well into open source contribution uh to see if any any people uh into the community is able to contribute into specific repo.

So, if you have any anything that you have in mind, feel free to let me know as well on Telegram. Um, and uh and yeah, let's keep in touch around that. >> Awesome. Thank you for having me. It was a pleasure. >> Thank you, too. Bye-bye. >> Bye, everyone. Bye. Bye. Bye.