Featured Session: The Internet of Value Meets the Internet of Intelligence

Speaker 1 00:00 I have a few notes before I introduce our next session, we are running our audience Q and A in this room via slido. You can find this by opening the session listing in the South by Southwest Go app or the web schedule and look for the engaged section. We are also live streaming certain sessions each day. So check out our YouTube channel. If you're interested in speaking at South by Southwest next year, your chance to apply is coming up in June via our panel picker process. You can find more information on our website. Lastly, explore the latest immersive art at the XR experience and check out the emerging Tech Expo, both located at the Fairmont what we are all here for today. Artificial intelligence systems become more capable of reasoning and acting autonomously, and blockchain infrastructure enables value to move instantly and globally. We're beginning to see the foundations of a new kind of digital economy, one where agents don't just analyze information, but can also hold transfer and earn value. The implications for global commerce, financial systems and enterprise operations are enormous. To help us understand this moment. We're joined by Lee fan, who leads engineering and AI innovation at Circle and brings deep X experience her roles at Google, Pinterest, blind and Baidu, and is working directly on the infrastructure of power and stable coins and digital finance. In conversation with Mel in the end, founder and CEO of operator collective. She's a longtime technology operator and investor who has spent two decades building and advising category defining companies. Please join me in welcoming Lee Van and Mel and yen for the Internet of value meets the internet of intelligence. I Well, Unknown Speaker 02:12 good afternoon, everyone. Thanks for being here. Speaker 2 02:15 Okay, so as actually, I'm curious, before we start, how many people, if you could raise your hand here with me who will help us provide some context for this? Provide some context for this in the audience, or founders. Great. And then how about works for a startup, Unknown Speaker 02:34 works for a public company, how about works for a private company. Speaker 2 02:42 And then one more question, is a creator? Thank you. Thank you for that. Okay, so blockchain and AI mature in parallel. We're seeing the convergence of the two powerful networks, which is the internet of intelligence and the Internet of value. And so today we're going to focus on that. And so over the past year, Lee, we've seen AI systems evolve from passive tools to autonomous agents, agents that can generate content do tasks on our behalf. And so this is what we roughly refer to as the internet of intelligence. Also at the same time, stable coins and tokenized assets have quietly established new financial reels, one where value moves globally, instantly and programmably. And this is, of course, the Internet of value. So this is a world where AI agents don't just generate content or execute tasks, but where they can actually hold, earn and send money. So what does commerce look like when participants aren't just humans or corporations, but intelligent software systems? So what changes when value exchange becomes machine made and so there's no better person to guide this discussion than the fan CTO and chief AI officer at Circle. She's led engineering at some of the most iconic companies out there, as we heard earlier, Google, Pinterest, etc, and she now sits at the center of building compliant global financial infrastructure for the AI era. And so I have the pleasure of knowing Lee for over a decade now, and I can tell you that not only is she a wonderful person, she and one of the smartest and savviest technologists around, but she's truly a great leader too. One of the signs of a great leader is when the people that you hired and trained as part of your team go on to do great things. And those who used to work for Lee one have come back to work for Lee and continue to break new ground, but also a number of founders or other folks who are now leading AI at these groundbreaking companies. So Lee, you're truly amazing. And also we're fortunate at operator collective, because when I had this crazy idea for this venture fund that was powered by operators, you were one of the first to sign on. So thank you so much. Speaker 3 04:54 My pleasure. It's always an honor and a total pleasure to have. Talk with Madam, because she's always had the right questions for folks audience in the book. Speaker 2 05:05 Thank you. Okay, so you helped scale consumer platforms used by hundreds of millions of people. And now financial infrastructure, which is powered by stable coin and blockchain technology, is a little bit different. And so in some ways, you know less front and center, but more of the foundation. So tell us, like, what made you decide to make the move from primarily consumer facing into FinTech and then, right? Yeah, a lot of Speaker 3 05:35 people ask me questions about this question. This is pretty personal. I don't know how many of Unknown Speaker 05:41 you are immigrant. Speaker 3 05:43 Reis, okay, great. I'm an immigrant myself. I entered the US as a graduate student in 1996 and I remember the first week and announced, I'm trying to call my mom and also send the money once I get to my scholarship and send money back. Both are extremely expensive and hard, right? And remember the international phone calls of $1 per minute. Can you imagine that? And also, of course, transfer money takes a lot, a lot of work and those events, and pay a fee and a way to sweep the Reis and again, try to figure out whether my mom received this or not. So fast forward, 10 years later. VoiceOver, IP, phone calls are so cheap I don't even think about anymore, right? That's like a couple of cents per minute. And then fast forward is like a WeChat WhatsApp, like it's no longer think about how to communicate your content, communication message. However, if you look at how we do money transfer, particularly across border, it remains the same, like 20 plus years, pretty much the same. Yes, there is an app doing this, and a little bit like you can initiate the transaction easier, but the whole process, how expensive and how many hops this path goes through, and how long it takes, pretty much remain the same. So when you think about it, that difference is like, why that? And so when I after mine, I was thinking about what to do next, and Jeremy approached me and he talked about the same, or the vision for circle, and how the new financial system will behave and how money will be programmable, and just like transfer as an email, as whatever the message is, and it's really instant hit on me, and I recognize it's also an infrastructure. There is a lot, a lot of technical challenge. I need to talk particularly because this is a very much regulated place, the trust, the scalability, is very important. So both for a technical leader, I saw the challenge as a person, as a person, I really see the need for a revolution in financial systems. Unknown Speaker 07:58 So this is a perfect fit. So what year was that when you joined circle Tucson, 21 so Speaker 2 08:05 2021 how many people were there when I joined single? Less than 200 No. And so at the time when AI hadn't been released yet? No, yeah. And so, one, did you anticipate that at some point AI would get to the point to be able to, you know, to Speaker 3 08:25 work at autonomously. No, I wish I had that foresight. No, I have no idea, but I will say this is orthogonal. Those are both powerful systems and happen to really combine together even more power, but even without AI people AI, the idea of the stable coin as an infrastructure, the layer for the new financial system, it works even, frankly, even without a gentle AI. It's a powerful infrastructure, but now with the agent, with Gen AI, even more powerful and unlock a lot of applications and the future. Speaker 2 09:04 And so how quickly after Gen AI came out, were you as you made the connection that, oh, actually, this could be incredibly powerful. Speaker 3 09:15 Ultimately, there will be a converges, I will say, we immediately recognize because the power of programmable money is because you need that speed. You need that like a human at the end of day, even if they do automate a lot of things, but the agenda part makes a lot of decision steps even faster and more efficient. So it's pretty natural. It's not really a click, click, it's, I would say it's naturally both powerful infrastructures or moving what you call Speaker 2 09:44 Okay, so we have people with very backgrounds in the audience, so let's actually do start with some of the basics. So when you talk about programmable money, so tell a little bit about what circle does, and also tell us a little bit about programmable money. And then we'll dig into agentic commerce. Speaker 3 10:00 Yes, so service mission is willing to build and build financial system to empower internet prosperity for everyone. And what we do is the at the core USDC, one of our stablecoin is really a digital form of currency that you can transact and you can translate on the blockchain technology, and it's make sure that you again, you can transmit the money and the transfer of money and in the speed of current like messages or emails. So it's a fridge, frictionless. It's real time, it's cheap. Speaker 2 10:37 And so for a circle, USDC, US dollar coin is paying one to one to the US dollar is that what the differentiator is between like you do Speaker 3 10:49 versus the other players do? We will say we are the most regulated. We are what the US dollar coin in this system, and particularly last year, the genius act went past, and we are the, I will say, most rapid. We are compliant and regulating the payment coin. Speaker 2 11:10 I mean, that's really the reputation, right, that you guys have always taken the line working with the regulators and working within the balance. I should also mention that there's we will take questions either 10 minutes at the end, but also feel free to put them in and I'll try to intersperse them as might make sense. So feel free to see a couple already that look good, but I'll try to pull up. Okay, so let's move for a second to agentic commerce, or the agentic economy. So for any of us that feels very abstract or speculative, and I should start with so I had the pleasure of interviewing you early last year, and people were saying that 2025, was year of the AI agent. And your response at the time was, Speaker 3 11:58 yeah, at that time, I would say, a little bit skeptics, because I do see and I believe the trend. I just don't know when it's going to happen. And I think that that's probably the because I don't know what's happening in the big AI labs and how fast the progress is, but in the past three months, everyone see all the dramatic progress from the base model, the capability and the agent is a real thing. The power, the reasoning engine, and the capability they can do, planning, executing is amazing. So I think it is. Yes, I still think I'm correct, because I'm trying to find some other year, but I do forget. You think the big breakthrough was? That's the difference between, I mean, things move so fast between 2025, and 2026, I think it's the reasoning engine to the point, and also the, actually the tool layer, like a particular the open claw really show a good example. When you connect the powerful reasoning engine with a tool that connect with both digital and a physical world, it can unlock a lot of things that people haven't seen before. Because if you think about maybe two years ago or one year ago, a lot of people interact with AI as just like a chat box, right? You got a lot of good knowledge, a lot of good answer from it. But in terms of a day to day work, how can you really empower your workflow? You still need a lot of engineer support. Because I'm doing I'm doing this in my company, I noticed I have actually AI engineer, dedicated AI engineer working with other functions to empower them, because you still need that however, fast forward to say, three months ago, when you see all the agent platform and all the framework that you can utilize. Now, everyone's empowered. Now, of course, you still need a little bit training and the time everyone's empowered to change your workflow, individually and team wise or otherwise, and to make sure that you not just have a reasoning engine which is super powerful for the position, the ability to seek deep and ability to seek multiple staff and foresee the result and also retract and learn the mistake that is really happening in the past half of the orcs, the and were there Speaker 2 14:27 also innovation or steps forward that were happening on the programmable money side, that it needed to happen? Or was it really just a convergence, the tip of programmable money was there because AI transformation and AI innovation happens in weeks, if not days, sometimes, and financial Fintech is a little bit slower, right? And so did one have to catch up with the other and or get to the point where down as convergence as possible? Speaker 3 14:55 Yeah, so that's a very good question, because there's a long time holding. You that because the technology, particular, AI role, happens so fast. But the financial world, for a good reason, it has to be very, I will say, soulful, and think about all the data rails and risk controls, regulatory consideration. But I will say this is a singularity moment that we all recognize that, as a intelligence, happen so fast, the financial world will have to get on the AI speed and not the other way around. And I think in this intersection circle and a lot of other FinTech companies and stable company actually is in the right moment that really just help, because the money that programmable money on blockchain infrastructure, is perfect in position to support that conversion. Speaker 2 15:49 So you mentioned open law, and no discussion itself by complete without a discussion of open law. And so you recently did an experiment, which was the open law and USDC hackathon, which took place on multi book. And so I know so mult book has been recently in the news because they've been acquired. They've been acquired by meta. And so for those of you who are familiar with notebook, it's a social network built specifically for AI agents, autonomous systems that can post comment, debate and collaborate each other, and the humans are just supposed to observe, but not be part of the discussion. I think you have a comment, comment on that. Speaker 3 16:31 Have you been on the book? Yes, it's actually spent one weekend, installed, actually viewed agent, envelope book myself, and trying to speak out what's going on. I'm an engineer by training, so every time there's a new scene, I really like to understand what exactly. Because when people talk about agents have a sense, agents have an emotion. They're talking I'm like, How can that be? So I use it so that I understand what's happening. So one little correction is, yes, human cannot get in directly. But the fact is, between every agent is a human or developer behind it to to really give the agent a person, persona, or give the agent a tool or prompt to say, how can you behave? Right? I of course, have a very dumb agent. Just listen all the crypto news and whatever, but you can see how the agent will interact, because someone behind it telling the agent say, this is how you do, right? You may want to observe how people argue about that topic, you can react and those things. So again, as an engineer, I want to understand how this from the digital business, when you see an app or when you see an app or when you see an agent, how all those things are connected so but things are moving. This is I tried this two months ago, and at that time, it's really early, and I see a lot of issues, and I can see why people are so scared of all the security, like a lack of security. But I actually recently, I noticed they improved a lot. So this is the beauty of this field, because everyone's working on that and really improve every day. So one more question on one book, and then we'll go on to the other pieces of it. So the fact Speaker 2 18:15 that there are, obviously, there are humans behind the agent, is that something you just knew because you are they and know how things work, because obviously there's a human behind it. Or could you actually tell that when you were in it, that Speaker 3 18:30 that there was humans actively prompting the agent, or otherwise? Well, so I think when to do it right. So this is when I said, We can talk later, like, if you one thing I really learned, and this is a reminder, because I've been a leader for a long time, for a long time, I don't have really getting to the code, but this is time I started getting told in myself. And when you start to do real setup yourself, when you really want to set up agents yourself, you will see the steps, and then you will understand. So you have to try to understand that no matter how many tweets or box you read, you don't really understand it, because those are stuff people will not talk about. Those are the, not the wow factor that people will like to discuss. But if you go that step by step and set this up and really configure how it's going to behave. Yeah. So, so this is the, actually, the weekend I try. This is actually before our own Speaker 2 19:29 hackathon, yeah. So then so open clauses to explain. Open cloud is the open source framework that's designed to provide agents with flaws, the tools and autonomous capabilities to allow them to interact with the visible and digital world. And so you created this open quality USDC, half a month, like a weekend after workbook. Speaker 3 19:49 Or I see that it's about the weekend I see that there's a weekend of workbooks. Talk about everywhere. And our team actually, it's like, I don't think quite end of it. We have smart engineer key and say, Hey, do you see. Something we should also try out and experiment. So we took from the idea to launch. It took about 48 hours, but I will say it's it's actually pretty I will say people move fast. We spent every more than half of the time really discuss the security and the potential risk, as an impact company working in a regulated space, even if it's just an open community hackathon, we have to think about the potential impact. That's where we decided we want the hackathon to the AI agent, to use USDC, but we don't want them to use the real, real money, because just just in case people make mistakes and not lose real money, we don't want that happen. So we actually just have a rules that you should really work on the test net, because this is something you should experiment and learn, not something like you do real transactions. I think we learned a lot of the response is overwhelming and it's, we only run one week, I believe, and we get more than 200 submissions. And we, first of all, I want to say this is the first hack of some level, agent to compete. So agent to go build whatever apps and to compete, and agent to vote. So we get 200 submissions, we think we get 1800 plus votes, and we get a lot of comments, maybe 10 ish comments, and eventually we issue this 30k award for winners across three tracks. And so the Speaker 2 21:36 purpose of this Reis to knowing that agents were the technology is getting in with agents, but allowing them to interact, earn, spend, build, up using programmable money. What does that look like? That was really the purpose of this, but not using real money yet. Speaker 3 21:50 Yeah, not real use money, and not using real money yet. But the idea, the API, the mechanics, is the same. So it quickly proved our hypothesis, once you let AI understand API for programmable money, for stablecoin, they can actually will be very creative. Have a lot of applications. One of the winner is Paul dollar, and they enable people to pay inference per transaction. When you think about it nowadays, if you want to run a little app and calling the LLM, you will have to most of time. You will have to get on account with anzaron legal opening app Gemini, and you will pay for the API key and pay monthly fee, or whatever the plan you are on. And this is the date interviewed application on blockchain, you can pay per transaction, and sometimes transaction only takes one cent, and this is not possible without a stable coin infrastructure. Speaker 2 22:53 Fascinating. And so when you were running that experiment, how much could you see the humans in the background? Speaker 3 22:58 We don't really see because we just want to see how the agent interact with it is each other, with each other, but we, again, as engineer, you understand the mechanism and but more and more, as I said, this was almost two months ago, more and more the agents are the more powerful. As long as you tell agent, those are the critical decisions you need to check LLM. They will ask the LLM themselves, and they will decide the next step. And they will know which steps to check in with the human owner. They will decide which time you can autonomously move forward. So I think again, ultimately, you can let it go yourself, but that's your decision. You can let the agent run fully autonomously. Okay, so this is a big question. Thanks for the question Unknown Speaker 23:49 from the audience, what is the future of stable Speaker 3 23:55 coins? Big Question, as I just said, because when I joined the company almost five years ago, I believe the state Bitcoin is a critical infrastructure to power the new financial system. Because, again, when money is programmable, just think about how you are used to the digital messages or emails, because all the things are really instant without thinking. And how can we do that with, of course, all the security in The Guardian Reis, that you can transact money with this kind of a speed and this kind of economic value, right? So I gave you a very simple example. I don't know how many of you went to see Taylor Swift concert. Not too many Latin people I will I didn't, because I heard all the like struggles, and it's like a long process, and you have to understand the way to getting the line digitally. And then when a lot of my fellow friends just tell me that they will have to wait at that specific moment and register. I can tell you, I also have a lot of engineer friends. They will totally write a script and they will wait that moment. They don't have to wait for, like, I say, midnight or something, because that's what they do. But think about it, most of the non engineer people cannot do that when you want to wait for a good deal, right? For example, I always want to fly across coast. I want to get a cheap business class, but what I did in the past is like, I go on United, checking everything, and just like two weeks ago, and compare price and figure out what is the best deal for me. But I can totally now have an agent to do that, and I don't need to struggle for few days to write my program. I can just prompt the agency do that with the limitation, maybe less than $1,000 or something, and you this time, give them enough constraint, the agent will do that for you without your waiting. But again, the key is you need to give the constraint. You need to get the proper, precise prompt so that they will do it on your behalf. So I never understand a lot of people may feel scared where they do it. Often they also, I would say, in general, if you do the right kind of a constraint, it would totally be amazing tool. So for Speaker 2 26:15 that example, today, it's a lot easier for a remedy to do that, as opposed to having to be an engineer to build that for you. The Can you do that with Speaker 3 26:27 regular money or really just appropriate with, yeah, so I will say it in theory, you could, but it's pretty awkward. And so think about it. You do want to give the agent to your credit card or bank account, and yes, moving into a virtual account. So it's doable, but stable coin. If you give a USDC wallet, for example, that is working on certain chains, it's much more natural because you can tell the wallet to give a given limit, and particularly if the other side receive USDC, it settles instantly. Even if you give a bank account a lot of times, it depends on the other side, it actually takes hours or days to settle. So there's, in terms of money transaction, there's like a settlement layer that in the traditional banking system, when it designer for human that kind of a in a second speed is not by design. So I will say in certain cases, I will not say only stablecoin, but in certain cases, yes, you may have a way to solve this, but stablecoin is naturally designed. It's a more internet native manipulator. Speaker 2 27:34 So thank you, Reis for this question, which is related, what are some of the use case examples where the system works beyond international commerce. Can it be used for everyday transactions? Speaker 3 27:44 Yeah, so I think it's eventually. I believe so. But again, in any of those money system involved, it takes time. So we are in a very, very early stage of this revolution, I think again, combination of eight, whether the CDC is very clear, and I think we talked about it in ecosystem. It's many, many players recognize it's coming. So we have all the protocols like x 402, 82 and Visa, Master stripe, Google, all into this. They recognize that in the agented commerce world, stable coin is the core infrastructure. And there's lot, lot of protocols now building up. And we are natural. We are a mutual player in this. We are not competing with any of the web two company or pavements that companies. We just want to provide this infrastructure so that the whole system can really upgrade. Unknown Speaker 28:42 Great. So other currencies are becoming common, and so thank you, Peter for this question. How do you see? Speaker 2 28:54 How do you see the, oh, my, another big question. You see the whole financial digital currency system functioning and they're operating globally now. Speaker 3 29:03 So this, again, because this is indesignable, intimately the applies to this naturally global. Speaker 2 29:13 Great and then, so if it's another good question, and thank you, Gustavo, if money becomes programmable. Who will control the goals of the money? Speaker 3 29:24 Yeah, so this is a very good question. It really touches the call. So it's not just a programmable that programmable side, the money side of we do still need a regulation, and we still need a lot of data bills. This is where the hardcore like, how circle position ourselves. We always compliance first, and we have the design that the compliance and the director design from the beginning. And this is even when we push out a lot of AI work within the company. We always make sure that the compliance control is enabled, not a blocker. Now. It is easier said than done, but I think when you have the right part of the right people who think about the right way, it's doable. So there won't be control, to be honest. But who's in the control? I think when, when you when the money goes down to really customer the USDC is a backed by us, dollar, one, to one, right? So in when we issue, first of all, as we circle, is a fully regulated company, and we are the how we how companies can issue, and a process and a strong money is, there is very clear regulation rules, and we are follow, and we are checking, make sure that all the accounting, and we do KYB for people who transact with us. There's a lot of rules we still we actually have to follow now, but this is a fast moving company, and everything's developing, so there's rules will be updated to really to match what's happening right now. So we are working with the regulators and other companies to make sure that the speed of the intelligence and the speed of a financial system can really match. Speaker 2 31:12 So when you describe it, it all sounds so much more efficient than traditional money. Do you think that it will become more commonplace and replace traditional bank accounts. Speaker 3 31:21 So I think it's a long process, but I do believe eventually people want more efficiency. Want a more I will say one frictionless transaction. Want to do more instead of working with many years old system. So I think with so many things already updated in the world. I do think that the management will move forward. But of course, because of a real money date, we believe we need to do, you know, more right way, in a careful way, but it's doable. Speaker 2 31:53 So, so we talked about the open cloud USDC hackathon, which was theoretical. Let's see what happens if we let the agents kind of go wild with an experiment with USDC that's not applied to real money. So where are we going the real world with allowing that to happen? How far did this journey? How early are we? Speaker 3 32:13 Yeah, it's we are extremely early. I still remember when I came to Silicon Valley 9099 and I saw all the internet. And at that time, like Google is a very basic tool, so it's always reminded me as that time of the internet, it's so early, a lot of demos, a lot of prototype, but I at the same time, you already see the real signals. And as I said, early, there's a lot of protocol building for 282, lot of company engaged in the protocol building, and we're going to support all the protocols. And in In addition, circle is launching arc. We launched the test net last, end of last October last year, and it is within a couple of months, we already find out we have one 90 million transactions, and every transaction settles about half second. And just think about all those, like, if it's a real financial transaction, settle half a second, and that's kind of throughput. And by the way, this is just text net, maybe at this year, there's a lot of things you can see, the earning size, and we believe this is where the future is. And there's a lot of active developers in this space too. Speaker 2 33:34 So we're still early and agentic commerce. We're still the regulations are still in development, and risk right on the financial side. So you are, you wear two hats at circle one, you're at a highly regulated, very compliant company. And then you're the CTO, so you have the crown jewels of the code and your technical, technology secrets. And then you also wear the cheap AI transformation hat. So I think I have two questions for you, which is, this is building on Ed's question here, What are the benefits and risks right now and going forward of the intersection between agentic AI and digital currencies, our financial institutions in particular, what actions can people be taking now? No Speaker 3 34:22 question. So I can say the action is just getting going by doing it, right? So I one thing I learned is I should read the lesson news, but just getting doing and of course, I learned so much just if you continue reading news and whatever it is, way too many. But as an engineer, as a builder, just getting to it. And also, the other thing is just talk to the people, the real people, what their pain point is, if you're thinking about doing something, and just ask them, what's because I. Think what happens is when have, when we have so many powerful AI tools, so the engineer overhead was is much less, right? So you, if you have a great idea, the building time and building overhead is much less. But what's really important is your saw building something that truly useful, instead of people talking about it. And one thing I learned is, when you're talking to people and just keep asking, like, Tell me more. What's the detail? What exactly your pain point is, why you cannot do this? Why can't you not do that? And once you understand that, and then you will feel there's still so many things that you can build up. So in terms of this, I think a lot of people, the people of this, I think a lot of people without thinking about the risk, particularly the Finland financial, as I said, We purposely designed a lot of our launches on test net first, including the head time, because we have to think about the impact. If people don't really understand it easily, what kind of mistake they would make, right? So, so we as a company, we want to make sure, at least for general population, we will be able to help them to get a benefit with the minimum risk. And one thing I will say, if you're in this space, you probably should think about, particularly if your app involved money, I will say, Yes, stable coin is a super efficient infrastructure, but it also makes things easier and faster, so you better understand how your agent is transact. And as I said earlier, give them a limit. I want my agents to go find the ticket and buy it for me. Why not sleep? But I will give a limit, right? I will say this is the wallet. Only have $500 go for it, even it happens. I know this is not going to be the end of the world. So you have to understand where your limit is, and have to understand the where it actually is Agent by making mistakes. So it's not like they're yes, they're smart, but they it's not a mistake. So you have to understand, as an engineer, is always a possibility to make mistakes bigger. So you said, you one of the Speaker 2 37:10 things you do is make sure you're going to test that environment first. Can you share some of the mistakes that we're learning? Right? That you saw that happened? You said, oh, good thing. We were in a test environment, and environment, Speaker 3 37:27 and now we can avoid that. Actually, I maybe I have a great engineer. So, so far in terms of how can we, of course, there's lack of scalability. We always push. I don't actually have a good example of for particular that even the risk Speaker 2 37:39 tolerances right in terms of nothing, tolerance is, right, in terms of nothing, really 100% invaluable, right? Speaker 3 37:50 So what do you think? What do you think is good enough, right? To get to the point, I will say this depends on the domain. There are things you I give you example. So right now, everyone else, like a coding agent. The coding tools are so powerful. We adoption of the engineer in terms of having the AI coding is tremendously high. But at this point, last thing is, it's the correct solution. But at this point in our company, we still require so every check in the code, if there's a human need behind to make sure the code is what we wanted to do. As I said, we very a lot of those rules. In general, I want to give engineers and employees a lot of freedom to try out, but when you touch these people, our customers and users money, we want to be super careful. Say we know what's happening, and there's a name who's accountable. I see in the AI world, even with AI intelligence for the accountability of the auditability, the security access control is still needed. So those are the challenges. A lot of people talk about it, but in the enterprise world, in the world, responsible for people's money, for things like high stakes, interest in me. Speaker 2 39:07 So what did you take over? When did you add the role of chief AI officer? Unknown Speaker 39:11 That was, I think it's beginning Speaker 2 39:15 of last year. Yeah. And your role so CTO, you know, that's pretty apparent. What you run into engineering and r&d and as chief AI officer, what does that mean at Circle? Speaker 3 39:26 Yeah, so very good question. Because sometimes people like a widest I don't want to explain, because when we start the AI journey, I think the year before, almost two years ago, it is like, I'm doing it because it's natural. I'm the CTO minor engineer AI. Start engineer team is the far front group that adopt AI, right? So the coding tools and whatever, and also for other functions to adopt AI, it's natural because they need engineers. So at the beginning, two years ago, I'm doing a CTO. I have a. One person when they feel engine doing the AI work for me, and I feel it's natural. So but as times goes by, when we look at the organization, particularly when we decided that AI is so powerful, we should really have everyone in the company, not just the engine. So let's say financing, product, illegal, team, everyone, literally everyone, they can utilize, they can adopt, they can really leverage AI power. And we recognize this is not a technical problem anymore. It's not just to say engineering solve this. It's a mindset shift, it's a cultural shift, it's organization shift. So Jeremy and I talk about it, recognize this is a special role that is independent. In addition to my CTO role to manage the tech off and we added this title to really signal to the organization that how important this transformation is and how important it is for me and for the organization work together, not treat me as a technical leader, but for the AI that people can really value together, to get on the journey together, because it's actually much more than a technical issue. Speaker 2 41:15 So So operator collective, right? We have a large number, over 250 what are called operator LPs, who have built scales that are running, often public companies. And so we've surveyed them, and we are constantly asking them so, so in the beginning, when Gen AI first came out, it seemed that most people, especially the public companies, was like shutting down, right? Don't access it, because you're going to infect all of our system. Our data is going to datas are going to be used to train all the data, train all the models. And then it became clear, oh well, there's a lot of innovation happening, and we just shut it down, then we're going to be behind. So then it was, okay, let's experiment. Let's let people experiment. And so, and I know that Jeremy and you are very AI born, he is always asking, it comes from the top right, but he's always asked, always asking, What are the two or three things that you have done built into the DNA now? But what are some of the bar Reis that you put into place to allow people to experiment, but not just sign up for every single new app that comes out, you know, every day? Speaker 3 42:16 Yeah, so once you have feel better when I started a guy journey, actually, and more than a year, like, almost like two years ago, we decided that AI, the transformation is not just technical pillar. So we actually have a three pillar. Even though I'm an AI officer, we have a three pillars. One is, of course, like engineering and tech support, but one is talent, talent organization, because we have to understand the psychological impact, organization impact of AI transformation, and that the other one is compliant and risk. So we work very close. I work very closely with our chief risk officer, and then we start from day one to think about what are the potential risk we should watch out, not saying everything is a blocker, but to understand what is the potential risk, what is the mitigation we can build alongside a tool when we should slow down, when we should just say, push forward, about with education, with gun Reis. So I have a very capable team just working alongside with our engineer team, constantly thinking about and evolve with AI. Because, you know, when you think about sometimes I look at the meeting notes, like one year ago, the issues we're talking about, and then the issues we're talking about today is totally different. Everyone's involved, and also this group is also using AI to sharpen their thinking, to really understand them, to To be honest, like a survey, all the companies and aware and maybe the potentially, where is the potential weak point? So frankly speaking, like it's really, I would say, fascinating to work with a group and understand that they are actually not blocker and enabled. Speaker 2 44:01 So would you say, what percentage of your company in, regardless of what function do you think it's actually building using AI, trying to drive efficiency within their within their roles? Speaker 3 44:11 Yeah, so I try not to get to but from what I can see is everyone now building can means differently. So for engineer, they can totally use an agent to build a project like from like a zero to one. But for other functions, they might be different. They may use the AI to check on regulatory reading and understanding what's the impact to them. They may be using a workflow to build a simple workflow to really get some get some data from this way, and check in the other way. So the degree is different, but we are doing tremendous trainings within the company, and everyone's participating. So my AI engineers are holding office hours every week, and there's a lot of engagement from different functions to ask. Questions, and this is what learning I didn't know one year ago. Because as an engineer, I always think about AI again, like more of an engineering or technical track of empowerment, but now as an AI leader, when I see how many people really can uplift their productivity and the creativity of wizard AI is truly amazing. Speaker 2 45:24 So early last year when we spoke, we were talking about the demos versus what happens in reality, and then also build versus buy. And we know things move lightning fast right now. So where do you like in terms of even your own journey at that circle, build versus buy, demos versus reality. Speaker 3 45:45 Yeah, so I will say demo versus reality is something that, again, with the organization wise, you have to just pay attention. Because I, I remember almost one year ago when we first want to enable everyone. We just want people to build, like, say, hackathon or Demo Day or show and tell. But now we really feel like people already reach the stage, and also the power of AI reach the stage that we can ask the real throughput, right for engineers, like already half year ahead, like we see all the throughput and productivity for other functions, also we start to have the real conflict goals. And this is something I think everyone can company can really learn by themselves, is when you go to demo, go from demo to production, there's a lot of tooling, and this is where the CTO has it really helpful is there's a lot of tooling you do need to work on within the company, because as the model the other enterprise AI companies like catching up. There's a lot of tools they may not build for your own name, particularly within our company, we have to be very thoughtful as we talk about what are the reasons we need to avoid. Sometimes we have to build our internal tool. But if we look at a few little features that we feel is common sense that other AI company can't really provide. And we will say, Hey, you should just ask them to provide. Yeah. So one of the so Speaker 2 47:10 operator collective, which released our inaugural stated AI transformation report, we're just a couple weeks ago, where we surveyed 123 operators in terms of what are they? Where are they on their AI transformation journey? Whether it has created more efficiency, everyone has said it has made virtually everyone has said they're using it. Their experimentation is just part of the job now, because things change so fast, and everyone said it made them more productive, but it hasn't resulted in demonstrable cost savings yet. And I think part of it, it seems to be, it's like, oftentimes, right? You can push a button and that task is faster, but then when you're working not just an individual, you're working as part of the company. It's got to integrate and push into workflows, right? Are you seeing that as well? Yeah, so Speaker 3 47:55 I think we are working on that. We recognize individual productivity. It doesn't really necessarily immediately transfer into team productivity, or all productivity. But I think at least in a lot of a pocket, we recognize why, because a lot of learning is in individuals that right? And, for example, we the engineers, build a lot of skills and to help them provide a good code and to efficient, but a lot of capable engineers writing individually, and there's a lot of repeated work, so we have dedicated engineer really, to make sure all the learning, the content, the knowledge, can be shared. And that sharing is tricky, because, again, within a company who you can share with, what are the things you learned that you can share with the other side without to compromise whatever access or the privacy or something. So all those things we are solving by now, so we have seen like if you solve those things, the productivity of the team will increase. But ultimately, the brain of AI need a context, and that context can be your individual context, and the context and the context can be a company context or team context. And when you want to increase the productivity of the creativity of the whole organization, you need to give the right context. And I think we are still in the age of every companies trying to figure out how to do it Speaker 2 49:16 most efficiently. Yeah, I've heard you use the great analogy with about data and AI, which is data is like your senses and AI is the brain. Speaker 3 49:26 Explain that a little Yeah, I will say it's AI is the brain. But if the brain cannot make a judgment with the wrong information. So when you think about human being, right, you see something, you hear something, those are the information you get there, you process it with your brain, and you make a decision. And AI is the brain, but the LM is the brain, but it needs to be fed into in the right data so that it can make the right decision. If you feed the misinformation or the wrong data or the partial data, it cannot make the right decision for you. So for particularly in the enterprise world, that our job is to make sure that when, if you will, you need AI to make those critical decisions, you have to check, you have to give the right data. And again, in the enterprise, many of you, if you build a company yourself, you know that the Data Silo or the data access control and although privacy consideration is a real thing, and those are the hidden challenges a lot of people don't talk about. Speaker 2 50:28 Okay, so please, we're down to the last 10 minutes. So if there's any additional questions, I'll try to get through as many as that we can here. So thank you for the questions. Okay, so the huge job, you've seen a lot of technology sifts through internet, through cloud and SAS, and now through AI. And you're also very successful angel investor on the side. So right now, if you were a founder or a builder, where would you lean in, if you're starting from scratch, what would you say? Speaker 3 50:56 So I will say this, this is a fascinating time to do to start a company, I think both fascinating, but it also kind of, I will say, a little bit tricky, because there's so many opportunities. When you think the traditional I will say that the effort and the time you have to spend on developing things, when developing a product or app is reduced to the minimum. What can you do? I think you can iterate or not. You should spend more time to understand the real issues, the real client. And I think that this is where I told my engineers, yes, your coding. My time may reduce, but you have to understand what exactly building, why this is useful for our partner, why this is useful for our developers and ecosystem partners, and that understanding translates into the real good product to the AI so that it Can it still requires a lot of human genome. Speaker 3 52:02 Why? But identify the right problem. Speaker 4 52:22 How can you stable? Democracy? Speaker 3 52:40 Yeah, we actually already see that, like in the use Unknown Speaker 52:52 now part of Speaker 3 53:02 it, fundamentally now each jurisdiction will make their own position, say how they want to control, how they want to Enable but as you also mentioned, and this is why I'm excited to have the transaction also Unknown Speaker 53:32 reduced. Now, there's a lot of work need to be done different space, but I think the convergence of a stable quantum stable infrastructure and Speaker 3 53:52 intelligence will unlock our potential. So if you are a founder, imagine what can be possible, right? There's a lot of things like you have to imagine and try it out. I cannot. I don't have right, right answer, but I tech opportunity. Thank Speaker 2 54:09 you. Improve on existing global wire payments. Does this exist in conjunction with them, or can it eventually replace them? Speaker 3 54:18 Oh, yeah. So I will say wired payment system, like at the end. This is more. I would say traditional one stable coin enables the transaction as fast as why, but without all the intermediate blocks. So we are so it can be incremental improvement. It can be fundamental improvement. It will take time for certain, let's say jurisdiction, to decide to do what side but we can really help some system to implement, to help speed the whole transaction experience. But it can also fundamentally change. I think the flexibility is. Speaker 2 55:00 Okay, so we're going to close with the very self bias question, which is, Do you worry that AI is taking away from natural intelligence and creative thinking? Speaker 3 55:11 Actually, I, so I will say I don't, but I simply depend on how you define again, what is intelligence sense, right? So I'm a coder by myself. When I first started my career, 30 years ago, I program since the C language or something so but what I realized is, when I get on my career, the ability to learn a new language, learn to solve a problem, is nothing to do with my training of specific language or specificity. It's really how I think about the problem and approach the problem. I think AI particularly remove the need of remember certain syntax, remember certain way to do things. Instead, you should focus on what exactly the problem I need to solve. I see that that particular challenge is still, at least today, still rely on human to make a judgment, because you understand what you need. Everyone's need is different. There's a lot of vertical AI companies and they are focused on what, again, customers need, and focus on what data that AI can utilize so that it will be mostly smart. I just want to emphasize, for those who are doing it, AI is not like magically learn everything. Human have to teach it, and human have to give the right scenario and context so that it won't make the right decision. It's not just magically know everything you know. It won't magically know what in the problem in front of you. I think our human creativity and intelligence will really like to identify the problem that is most equitable and most immediate. But I think one of the things you mentioned Speaker 2 57:04 because I think people may look at you and say, okay, you've been a technologist on your life. You can, let's do all these tech transitions. Of course, you know how to program. You know how to do this. But when AI came out, you didn't Unknown Speaker 57:17 just do what you've always Speaker 2 57:19 been doing. So just tell us a little bit what you did to actually start to stimulate so many parts Speaker 3 57:27 of your brain that maybe I've been using before, right? Because, yeah, I want to choose things. One is what I cause. Again, like some it's hard to do, because when you become a leader, and there's a lot of organizational issues, and I'm like a checking, checking the trade off a document, but I haven't really looked at the whole of what's the code, and I haven't really told myself for years. But when AI comes along, because I said, I really want to understand why everyone's talking about why the code is so smart. So I start to read the papers, and I start to hold myself over the weekend, and I realized I'm in Frankly speaking, I feel you guys are holding slower than I hope, but then, because you have to struggle with all the mistakes, and trust me, when you do it, you will feel it's not as smooth as they said. When you struggle with the mistakes and with glitches, you recognize, okay, this is how the system put together. And when I read the papers and I think out, okay, why people do the improvement here that will save your inference cost, all those things help me to understand, at the end of the day, again, the Cuban judgment is still needed. You will have to identify, maybe it's a cost issue for you, maybe it's a compliance issue for you, you will have to identify what's most important for you, what are issues that AI can solve for you and not solve. So as a company, as a human individual, our job is to, again, lower the friction, build a system that is safe for you, to enable things you don't have to worry, so that you can focus the things you can worry. And one more thing I want to comment is It is fascinating to observe how non engineered team learn AI. So I talk to them when they talk about the issue they want to solve, and I recognize, wow. So no matter how smart AI is, it takes that domain expert to recognize this is the issue I want to solve, because we have, hey, I have this issue. Can AI help me? And that is the things I will never learn, and that's a domain expert. And we were so excited say, Yes, we can solve all Hey, AI has a limitation, but we can build a system together on top of the LF and make sure it's happening. So this is a fascinating journey. And I said, I'm sure if you try out yourself, you will Speaker 2 59:49 feel it, and it really doesn't enable anyone technical background or not, to become a builder and a creator on their own. Speaker 3 59:54 Yeah, this time really for I think the plane is a metal to us, and we. They should be able to I must say, even if you're not engineered, this isn't really a time for you to be Speaker 2 1:00:06 Yeah, because you know the problems that your function faces better than anyone ever will rely on you have the ability to build your way out of them anyway. Please. Thank you so much for talking with me today, and thank you all for being here. Applause, Unknown Speaker 1:00:30 lots of Unknown Speaker 1:00:41 material. A magazine You Reis. Oh, I Speaker 4 1:02:00 You Fernando, for the reservations on the end of the Hall, sorry, Unknown Speaker 1:03:01 end of the hall To the left side. To be Reis. Unknown Speaker 1:04:02 Always you Good Unknown Speaker 1:04:42 question. Unknown Speaker 1:05:20 And kick the door wide Open. Unknown Speaker 1:05:41 I wish I could Unknown Speaker 1:05:49 Ready to say that i i You I You Unknown Speaker 1:07:20 I You one? Do you need an extra Hand? Unknown Speaker 1:08:40 In capital, I i Speaker 4 1:09:22 would go I feel like I have tomorrow And then Wednesday, Speaker 4 1:09:37 like I'm not counting on Wednesday as counting on Wednesdays. They're Unknown Speaker 1:09:59 fun. And have Speaker 4 1:10:42 now Don't Know You You You You You Thank You. You You I You. Unknown Speaker 1:15:20 You You You You You You You Good. Thank. You You You Getting over You You You Speaker 4 1:21:20 You You You You You You Unknown Speaker 1:23:40 You that's it, right? Yeah, you you. You I I You So You I You Thank so. You You