AiFi: Agentic Identity

Thank you all for tuning in. I'm Pia, I lead AI product comms here at Coinbase. I'll give you a little spiel as an opener. So agents are already a huge part of our everyday lives and increasingly are starting to interact with each other too. And of course that goes well beyond just a notebook or a long-lost internet forums. As agents start to really purchase and discover services from one another, the question becomes, how does your agent know a quality reliable service in a swarm? How does an agent prove its legit? So that's the core of a genetic identity that will unpack today with this fun gaggle of guests. We have got Kevin and Lincoln from Coinbase, Hayton from Browserbase, Leaton from Tools for Humanity and World ID, Marco from Lovanto AI and the author of ERC 8004, and of course Eric, the founder of X4 or 2 and a Coinbase alum. So before we get started with introductions, I wanted to make sure that our audience gets the most out of this session. So we have a little twist. I'm going to be playing a sound. Anytime anyone mentions anything to jargonny that needs explanations, I promise I'll try not to be too too inquisitive with it. Okay, so Lincoln, can you start us off with a 30-second intro about what you're working on and then pass the mic to whoever goes next? Yeah, absolutely. Good to be here. Thank you for having me. My name is Lincoln. I'm a product lead for our agentic trading and payments initiatives. So I really bring it down into two things. One is X4 or two, which is our internet native payment standard to let AI agents transact across the internet and our Coinbase MCP for agents, which is great for giving your AI agent whatever in whatever chat interface you want, access to your Coinbase account for agentic trading. All right, Kevin, take it away. Awesome. Thank you, Pia. Hey everybody. Nice to have everyone here in the space. My name is Kevin LeFueh. I'm on the business side for agentic payments and agentic trading. I've worked with awesome companies in the AI space like browser base, XA, apify, massive and others. And what I'm focused on is solving all the friction around tool usage for agents and enabling agents to reason economically when buying things on the internet. So factoring in like marginal cost decision making into the context window. And I'll just go next. So my name is Layton. Thanks for everybody listening in. I'm from Tools for Humanity. My title there is Head of Ecosystem. And I've worked with Kevin in a browser base and some others on this Payton at browser base who's also on this call on a bunch of different stuff we'll talk about. But primarily what we bring to the table here is proof of human and proof of human for agents. So I will get into talking about that in a little bit. Yeah, I'll go and jump in. My name is Peyton. I lead up identity here at browser base. We basically build out kind of headless browser infrastructure as well as supporting tools for agents to go interact with websites, different services, and things like that. Agents, payments are becoming quite a big thing right now as all these personal assistants are popping off. And so agent identity, which is one, sometimes how can we tie it back to the human that the agent is acting on behalf of or how do we give agent its own unique identity to manage accounts and services and stuff like that. Basically solving that reputation problem and credential management at scale. I'm not sure who to pass it up to next, but. Marco, are you? Yes, my turn. Hi guys. Marco de Rossi, what I'm working on is Meta must be recently launched on a JNT wallet for traders where you can like basically one shot and enter trading agent just with one command in one minute. And one month ago I launched my new company, it's called Levanto. And we are working on models for machines, not like the models for humans that we have been using. So they're like 10x faster and they are humble enough to say, I don't know, instead of like taking confidence like an Alhance usually do. So they're great to be used within code for machines. All right, Eric, bring us home. Hey, America, I worked at Quentes for a while. I helped create x402 and I'm now working on a new company that is in stealth. So feel a little more about it soon. All right, awesome. Well, thank you guys for all being here. Super excited to kick this off. I think let's before we get into the meet and bones to too much about just identity, I'll throw you guys a bone on an easy one. Can we just talk about AIFI? Like, what is that? How do you think about it? Open question for anyone? Yeah, I can start. So for me, AIFI, right, is AI Finance. It's actually the ability to leverage language models and personal systems or agents to do tasks financially who's a human. And so this includes things like buying and selling, different types of equities. For more importantly, it's enabling the language model to develop comprehensive strategies based on your own goals from an investment perspective as a user. And so part of that requires real time context around financial markets. And it incorporates really tool calling, context, and language models in the stack. Leighton, did you have a hot take? You know what? I actually don't. I was going to tell Kevin he should answer it. So he did. All right, nice. Yeah, of course, Brian's been a big proponent of AIFI and we're excited to see how it's building out in a lot of different ways through like, agentic grading, agentic payments. It really runs the whole gambit, which is exciting to see. To me, it's like a temporary label. We'll stop using it soon because like every, like a finance will just become like defined by agents. So it feels like when we were talking about the internet economy a few years ago. So it's just a temporary label to me for describing this, like, economy where like these issues are taken by agents and not humans. But it would be just agents like doing what we ask. I feel like there's a long conversation there around human level approvals for financial transactions. Like what size do humans need to be involved? I don't know. Everybody would you let your agent spend $50 autonomously? Would you let them spend $1,000 daily that's going to make them increasingly interesting? We see this a little bit on the credential side where it sounds great in theory to want to approve your agent's access to your credentials because it sounds secure. But you quickly get into kind of this fatigue situation where it's like you're either approving all because your agent is doing things in the background for you. Or you really need to kind of have a risk model for whether or not transactions become, you know, should trigger an approval flow or not. I think like the AI finance should be interesting as it scales up for sure. Cool. And that's a great segue to our first kind of on topic question, being like what does an agent really need to be verifiable, especially like in an agentic financial way or beyond? Like what does the identity actually look like when you're at the agent level? I can jump in and talk about I think there's probably three components to the stack, right? It's proof of human, which is what we specialize in. And then you sort of have beyond proof of human you have. Can you please expand on what proof of human really entails? Oh, yeah. Oh, sorry. Sorry. I'm too much jargon. So proof of human is basically a way to represent on the internet that you or an agent representing you is a unique human. And the reason that's important is because there's now tons of agents on the internet, and they're sort of overwhelming different services and websites. You see this like recently with instinct and the rezzly stuff and trying to make reservations and getting shut down. But you also see this with just normal websites that are getting way too many agents querying their information which is racing their costs. But then they're also losing money because they don't get any advertising anymore. So what proof of human does is it enables a person or an agent operating on behalf of a person to testify effectively to provide a proof to a website that they're acting on behalf of a unique person. And what that then does is lets websites basically offer their services more permissively to agents that are verifiably tied to a unique human. Because if you know 100 agents on your website represent 100 unique humans, then you can do treat them differently if you have 100 agents representing one human. So that's the proof of human part of the stack that I'll speak to. I think other partners on the call are more equipped to speak to the financial part as well as the identity part. But that's specifically what proof of human brings to agents. Can we double-clip on that first sec? Could you throw out one or two examples of what proof of human might tangibly unlock for an agent or its human user? Well, yeah, yeah. So sort of the canonical simple example in two of examples is like a concert ticket or a restaurant reservation, right? You don't care. If you're selling a concert ticket, you don't care if a human is literally pushing the buy ticket button or if an agent is pushing it. But you do care a lot if whoever is pushing that button already has a ticket. Or if whoever is pushing that button is a unique person or if it's the same person hitting the button 100 times. So that's sort of like, I think the maybe the most like intuitive example. But then you can extrapolate that to all sorts of different things like with, for example, with browser base with payton who's on the call. Like we have a relationship with them where they obviously provide services to agents. And if they