Why AI Agents Will Need Their Own Bank Accounts — Lincoln Murr, Coinbase

Thank you to everyone who Thank you to everyone who came. My name is came. My name is came. My name is Lincoln. I lead the Lincoln. I lead the Lincoln. I lead the AI ​​product direction AI ​​product direction AI ​​product direction at Coinbase. Today we're at Coinbase. Today we're at Coinbase. Today we're going to talk a little bit about going to talk a little bit about going to talk a little bit about how we build an how we build an how we build an agent agent agent financial account. And financial account. And financial account. And I know what you're I know what you're I know what you're probably thinking. Coinbase, probably thinking. Coinbase, probably thinking. Coinbase, crypto, AI—lots of crypto, AI—lots of crypto, AI—lots of buzzwords. But I buzzwords. But I buzzwords. But I hope that hope that hope that by the end of our by the end of our by the end of our conversation you conversation you conversation you will understand what will understand what will understand what agency finance is and agency finance is and agency finance is and that it is a very real that it is a very real that it is a very real thing that is already thing that is already thing that is already happening. You happening. You happening. You will learn a little about will learn a little about will learn a little about what we have learned what we have learned what we have learned over the past 2 years of over the past 2 years of over the past 2 years of development, and development, and development, and hopefully you will be able to hopefully you will be able to hopefully you will be able to draw draw draw useful useful useful conclusions from it. I'll also conclusions from it. I'll also conclusions from it. I'll also talk a little bit about talk a little bit about where we're going with where we're going with where we're going with this topic and how this topic and how this topic and how agency finance agency finance agency finance will evolve over will evolve over will evolve over time. Steeply. So, time. Steeply. So, time. Steeply. So, to understand where I'm to understand where I'm to understand where I'm going, it's important going, it's important going, it's important to understand what to understand what to understand what agency finance is. In agency finance is. In agency finance is. In my opinion, agency my opinion, agency my opinion, agency finance is anything finance is anything finance is anything related to an agent related to an agent related to an agent who can manage who can manage who can manage money or has money or has money or has access to a access to a access to a financial account. financial account. financial account. It sounds obvious, but It sounds obvious, but It sounds obvious, but what does it what does it what does it really mean? We see really mean? We see really mean? We see two main areas of two main areas of two main areas of use. These are use. These are use. These are agents who need to agents who need to agents who need to spend money and spend money and spend money and agents who need to agents who need to agents who need to multiply it.

multiply it. multiply it. Multiplication is still a Multiplication is still a Multiplication is still a somewhat new somewhat new somewhat new category. So I'll category. So I'll category. So I'll start with a detailed start with a detailed start with a detailed look at the look at the look at the payments sector. To do this, it's payments sector. To do this, it's payments sector. To do this, it's worth going back worth going back worth going back two years, to two years, to two years, to our first our first our first experiment with experiment with experiment with agent agent agent finance—a finance—a tool we tool we tool we called Agent Kit. The idea called Agent Kit. The idea called Agent Kit. The idea was that before the was that before the was that before the advent of MCP or any advent of MCP or any advent of MCP or any standardized AI standardized AI tools, we tools, we tools, we used the used the used the LangChain framework and LangChain framework and LangChain framework and added a wallet to it added a wallet to it added a wallet to it . We gave . We gave . We gave artificial intelligence artificial intelligence artificial intelligence access to money, access to money, access to money, including including including cryptocurrency. cryptocurrency. cryptocurrency. It was a fairly It was a fairly It was a fairly new thing at the time. I remember it new thing at the time. I remember it new thing at the time. I remember it was two in the was two in the was two in the morning, I was writing to my morning, I was writing to my morning, I was writing to my agent: "Hey, can you agent: "Hey, can you agent: "Hey, can you send the money send the money send the money to so-and-so?" And he to so-and-so?" And he to so-and-so?" And he did it. I'm like, "Wow." did it. I'm like, "Wow." did it. I'm like, "Wow." Today, this is Today, this is Today, this is no longer impressive at all. But no longer impressive at all. But no longer impressive at all. But it was quite it was quite it was quite exciting at the time. However, exciting at the time. However, exciting at the time. However, when we gave this when we gave this when we gave this tool tool tool to developers, we to developers, we to developers, we noticed one noticed one noticed one major problem. We major problem. We major problem. We gave AI access to gave AI access to gave AI access to money, but we didn't give it money, but we didn't give it money, but we didn't give it anything to anything to anything to pay for with it. pay for with it. pay for with it. This was a small This was a small This was a small mistake on our part mistake on our part , which we sought to , which we sought to , which we sought to correct, and we correct, and we correct, and we succeeded thanks to succeeded thanks to succeeded thanks to the work of other the work of other the work of other specialists. How many of you are specialists. How many of you are specialists. How many of you are familiar with familiar with familiar with HTTP status codes? I hope so HTTP status codes? I hope so HTTP status codes? I hope so , the majority in the , the majority in the , the majority in the room. Okay. So, for room. Okay. So, for room. Okay. So, for those who aren't aware, it's those who aren't aware, it's those who aren't aware, it's essentially a standard essentially a standard essentially a standard that defines how that defines how that defines how websites communicate websites communicate websites communicate and how information is and how information is and how information is transmitted over the transmitted over the transmitted over the internet. One very internet. One very internet. One very well-known code well-known code well-known code is called 404, it is called 404, it is called 404, it means that means that means that an error has occurred. 200 means an error has occurred. 200 means an error has occurred. 200 means success, but there is one code success, but there is one code success, but there is one code that that that has been in the has been in the has been in the Internet specification for decades, but no Internet specification for decades, but no Internet specification for decades, but no one has ever one has ever one has ever implemented it. In fact, implemented it. In fact, implemented it. In fact, Marc Andreessen Marc Andreessen Marc Andreessen called it the " called it the " original sin" original sin" original sin" of the internet: we never of the internet: we never of the internet: we never implemented it implemented it implemented it properly. Those properly. Those properly. Those who know a little about who know a little about who know a little about what we're working on

what we're working on what we're working on will guess that this is the will guess that this is the will guess that this is the HTTP status code 402, which HTTP status code 402, which HTTP status code 402, which means "payment means "payment means "payment required." The required." The required." The official definition of official definition of official definition of this code in the Internet this code in the Internet specification specification specification is " is " reserved for reserved for reserved for future future future use." use." use." Quite general. But Quite general. But Quite general. But now we have a great now we have a great now we have a great opportunity, thanks to the emergence of opportunity, thanks to the emergence of opportunity, thanks to the emergence of AI agents who AI agents who AI agents who need a clean, need a clean, need a clean, fast, and fast, and fast, and efficient way to efficient way to efficient way to transfer money and transfer money and transfer money and conduct conduct conduct machine-to-machine trade, as well as machine-to-machine trade, as well as machine-to-machine trade, as well as stablecoins as a stablecoins as a stablecoins as a cheap and fast cheap and fast cheap and fast payment system, payment system, payment system, to build X402. This is our to build X402. This is our to build X402. This is our standard for standard for standard for native online native online payments. This is supposedly payments. This is supposedly payments. This is supposedly something new. I think it's still something new. I think it's still something new. I think it's still relatively new relatively new , but it came out , but it came out , but it came out around last around last around last May. And since then, May. And since then, May. And since then, we've seen we've seen we've seen very strong very strong very strong adoption, adoption, adoption, honestly, much honestly, much honestly, much stronger than I stronger than I stronger than I expected. When it expected. When it expected. When it launched in May, launched in May, launched in May, our goal was to our goal was to our goal was to reach 25,000 reach 25,000 agent-made transactions agent-made transactions by the end of the year. by the end of the year. by the end of the year. At the end of the year, we At the end of the year, we At the end of the year, we saw such volumes saw such volumes saw such volumes every 17 seconds. And every 17 seconds. And every 17 seconds. And today we see some today we see some today we see some pretty incredible pretty incredible pretty incredible numbers: tens of numbers: tens of numbers: tens of millions in volume millions in volume millions in volume and hundreds of millions of and hundreds of millions of and hundreds of millions of transactions every 30 transactions every 30 transactions every 30 days. And if you days. And if you days. And if you look at this look at this look at this list, you'll see list, you'll see list, you'll see that it's not just that it's not just that it's not just crypto companies. There's crypto companies. There's crypto companies. There's Audius, Amex, PayPal, Visa, MasterCard, Audius, Amex, PayPal, Visa, MasterCard, Audius, Amex, PayPal, Visa, MasterCard, Stripe—all these big Stripe—all these big Stripe—all these big names in names in names in payments that are payments that are payments that are interoperating with X402 interoperating with X402 interoperating with X402 because we're building this because we're building this because we're building this as an open as an open as an open standard where anyone standard where anyone standard where anyone can use can use can use it to create a it to create a it to create a way to way to way to make payments make payments make payments between machines. I know there's been between machines. I know there's been a lot of talk a lot of talk around around around agent commerce lately, agent commerce lately, agent commerce lately, especially with the news especially with the news especially with the news about Grokbot, Stripe Link, and this whole about Grokbot, Stripe Link, and this whole about Grokbot, Stripe Link, and this whole controversy around controversy around controversy around Amazon blocking Amazon blocking Amazon blocking agents. And I think agents. And I think agents. And I think this is all a very this is all a very this is all a very clear clear clear use case. I can use case. I can use case. I can talk about agency

talk about agency talk about agency fees all day. fees all day. fees all day. But there's one aspect But there's one aspect But there's one aspect I want to I want to I want to touch on, and it's one touch on, and it's one touch on, and it's one I think is particularly I think is particularly I think is particularly relevant to relevant to relevant to microtransactions—it's microtransactions—it's microtransactions—it's something that really something that really something that really inspires me about this whole inspires me about this whole inspires me about this whole topic of agent topic of agent topic of agent payments. So, if payments. So, if payments. So, if you haven't noticed yet, you haven't noticed yet, you haven't noticed yet, the way we the way we the way we interact on interact on interact on the internet has changed the internet has changed the internet has changed over the past couple of years over the past couple of years . Now we mostly . Now we mostly . Now we mostly do this through chats do this through chats do this through chats or Google AI reviews, or Google AI reviews, or Google AI reviews, instead of instead of instead of going to the going to the going to the websites themselves. And while this websites themselves. And while this websites themselves. And while this all all all looks great to the consumer, it looks great to the consumer, it looks great to the consumer, it has created quite a has created quite a has created quite a problem for problem for problem for publishers, who previously publishers, who previously publishers, who previously relied on relied on relied on advertising revenue and advertising revenue and advertising revenue and human attention to human attention to human attention to monetize their monetize their monetize their content. It feels like content. It feels like content. It feels like we're in the we're in the we're in the Limewire era, where you can Limewire era, where you can Limewire era, where you can collect collect collect all the content for free, and there's all the content for free, and there's all the content for free, and there's practically nothing you can practically nothing you can practically nothing you can do about it. do about it. do about it. We recently We recently We recently saw bot saw bot traffic traffic traffic surpass human traffic on surpass human traffic on surpass human traffic on the internet for the first time, and this number will the internet for the first time, and this number will the internet for the first time, and this number will only grow only grow only grow as more as more as more people use people use people use AI and traditional AI and traditional AI and traditional web services and sites web services and sites web services and sites lose their role as lose their role as lose their role as intermediaries. So, intermediaries. So, intermediaries. So, how to solve this how to solve this how to solve this problem? There's a problem? There's a problem? There's a very promising very promising very promising opportunity here that's just opportunity here that's just opportunity here that's just starting starting starting to take hold, and to take hold, and to take hold, and it's started to it's started to it's started to materialize materialize materialize over the past over the past over the past few months: few months: few months: sites can sites can sites can cap their cap their cap their content with content with content with microtransactions microtransactions microtransactions specifically specifically specifically designed for designed for designed for agents. The idea is agents. The idea is agents. The idea is that if I run a that if I run a that if I run a Substack and don't want an Substack and don't want an Substack and don't want an agent to read it agent to read it agent to read it for free, I can for free, I can for free, I can charge them 1/10,000th of a charge them 1/10,000th of a charge them 1/10,000th of a cent. This is a negligible cent. This is a negligible cent. This is a negligible amount for the user.

amount for the user. amount for the user. Ultimately, this should be Ultimately, this should be Ultimately, this should be fully fully fully automated just like automated just like automated just like your your your AI subscription, but it is a AI subscription, but it is a AI subscription, but it is a significant source of significant source of significant source of revenue for these revenue for these revenue for these companies. We will begin companies. We will begin companies. We will begin to see a shift to see a shift to see a shift from an online economy from an online economy from an online economy based on attention based on attention based on attention and advertising to one and advertising to one and advertising to one focused on focused on focused on microtransactions and microtransactions and microtransactions and utility. And if you utility. And if you utility. And if you don't believe me, I advise you to don't believe me, I advise you to don't believe me, I advise you to take a look at AWS and take a look at AWS and take a look at AWS and Cloudflare. Both of these Cloudflare. Both of these Cloudflare. Both of these companies are companies are companies are huge huge huge content delivery networks. content delivery networks. content delivery networks. Together they Together they Together they control about control about control about 40% of all internet 40% of all internet traffic, and traffic, and traffic, and both recently added both recently added both recently added support for support for support for X402 microtransactions X402 microtransactions X402 microtransactions to restrict to restrict to restrict access to any access to any access to any publisher site. And publisher site. And publisher site. And while this hasn't been while this hasn't been while this hasn't been widely widely widely adopted yet, it adopted yet, it adopted yet, it seems inevitable to me seems inevitable to me seems inevitable to me that in the coming months that in the coming months that in the coming months and years, this will become a and years, this will become a and years, this will become a new way for new way for companies companies to differentiate themselves to differentiate themselves to differentiate themselves and, in a sense, and, in a sense, and, in a sense, cut out cut out cut out the middleman. So, the middleman. So, the middleman. So, payments. This is logical. payments. This is logical. payments. This is logical. We see quite We see quite We see quite strong progress here. Everything is strong progress here. Everything is strong progress here. Everything is going well. But what going well. But what going well. But what about this idea about this idea about this idea of ​​"making money of ​​"making money "? What could this mean? "? What could this mean? "? What could this mean? Well, it's about Well, it's about Well, it's about agent trading and agent trading and agent trading and investing. When I investing. When I investing. When I first heard about it, I first heard about it, I first heard about it, I thought it was the thought it was the thought it was the most ridiculous thing in the most ridiculous thing in the most ridiculous thing in the world. Why should I world. Why should I world. Why should I give my ChatGPT give my ChatGPT give my ChatGPT access to my access to my access to my bank account bank account ? Or why would I ? Or why would I ? Or why would I ask Grok to make ask Grok to make ask Grok to make me money and me money and me money and expect it to expect it to expect it to work? But work? But work? But digging deeper, we digging deeper, we digging deeper, we discovered three main discovered three main use cases that use cases that already exist today.

already exist today. already exist today. I'm sure there will be more in I'm sure there will be more in I'm sure there will be more in the future the future the future . And the main . And the main . And the main difference between all difference between all difference between all this “Grok, make me this “Grok, make me this “Grok, make me money” and what money” and what actually happens is actually happens is that the person is that the person is that the person is only giving up a tiny bit of only giving up a tiny bit of only giving up a tiny bit of control. I control. I control. I mean, the human is mean, the human is mean, the human is still, to some extent, still, to some extent, still, to some extent, directing the agent directing the agent directing the agent to automate to automate to automate tasks that would otherwise be tasks that would otherwise be tasks that would otherwise be very difficult very difficult . Let's say I want . Let's say I want . Let's say I want my low-risk my low-risk my low-risk account, where 80% S&P, 5% account, where 80% S&P, 5% Bitcoin, and 15% Bitcoin, and 15% Treasury Treasury Treasury bills, to be fully bills, to be fully bills, to be fully managed by an agent. I can managed by an agent. I can managed by an agent. I can just order it to just order it to just order it to do it, provide do it, provide do it, provide authorization with all the authorization with all the authorization with all the proper proper proper safeguards, and safeguards, and safeguards, and let it let it let it run. And I don't have to run. And I don't have to run. And I don't have to think about think about think about my money. my money. my money. Similarly, we saw that Similarly, we saw that Similarly, we saw that standalone execution standalone execution standalone execution is a pretty is a pretty is a pretty compelling compelling use case. use case. I'll give an I'll give an I'll give an example of what this example of what this example of what this means later, but imagine means later, but imagine means later, but imagine you're interested in you're interested in you're interested in trading trading trading first-hand news and what's first-hand news and what's first-hand news and what's happening with happening with happening with oil. You can either oil. You can either oil. You can either sit in front of sit in front of sit in front of a keyboard at 3 a keyboard at 3 a keyboard at 3 a.m., a.m., a.m., watching watching watching charts and charts and charts and news, or you can tell news, or you can tell news, or you can tell the agent, “Hey, here’s what I the agent, “Hey, here’s what I the agent, “Hey, here’s what I think will happen in think will happen in think will happen in these scenarios. If these scenarios. If these scenarios. If it happens, just it happens, just it happens, just make this bet." And the make this bet." And the make this bet." And the third thing that is third thing that is third thing that is particularly attractive particularly attractive is the idea is the idea is the idea of ​​transforming a of ​​transforming a of ​​transforming a high-level idea in high-level idea in high-level idea in natural language into natural language into natural language into real financial real financial real financial instruments in instruments in instruments in strategies. Invest in strategies. Invest in strategies. Invest in space. Invest in space. Invest in space. Invest in quantum technologies.

quantum technologies. quantum technologies. Make me a Make me a Make me a low- low- low- risk portfolio. Invest like risk portfolio. Invest like risk portfolio. Invest like Warren Buffett. All of this is Warren Buffett. All of this is Warren Buffett. All of this is for people who have for people who have for people who have ideas but don't have the ideas but don't have the ideas but don't have the time to implement them time to implement them time to implement them and don't know exactly how - and don't know exactly how - and don't know exactly how - this is an extremely this is an extremely this is an extremely attractive attractive attractive opportunity for opportunity for opportunity for agents to enter this agents to enter this agents to enter this market. So, I'll give market. So, I'll give market. So, I'll give you one clear you one clear you one clear example of what I example of what I example of what I actually did actually did actually did a few weekends ago. a few weekends ago. a few weekends ago. You probably remember You probably remember that Daria that Daria that Daria published this published this published this post: "We must post: "We must post: "We must accelerate progress." accelerate progress." accelerate progress." It was on the weekend. It was on the weekend. It was on the weekend. That was about a That was about a That was about a month ago. And I month ago. And I month ago. And I saw on X that Elon saw on X that Elon saw on X that Elon wrote: "Daria is wrote: "Daria is wrote: "Daria is right." I was right." I was right." I was stunned. This is the stunned. This is the stunned. This is the first time first time first time they have both agreed on something they have both agreed on something they have both agreed on something . This will likely be . This will likely be market-impacting news on market-impacting news on Monday. So I Monday. So I Monday. So I went to my agent went to my agent went to my agent and said, "I think the and said, "I think the and said, "I think the stock [blank]..." You stock [blank]..." You stock [blank]..." You can guess can guess can guess what the stock is, but I what the stock is, but I what the stock is, but I won't talk about it won't talk about it won't talk about it . I think . I think . I think these stocks will fall. And these stocks will fall. And these stocks will fall. And since we are now since we are now since we are now entering the era of entering the era of entering the era of 24/7 markets, I 24/7 markets, I 24/7 markets, I was able to open a was able to open a was able to open a short position over short position over short position over the weekend just by the weekend just by the weekend just by saying it into the phone saying it into the phone . Could I do . Could I do . Could I do this by sitting in front of this by sitting in front of this by sitting in front of a computer and a computer and a computer and pressing all the pressing all the pressing all the buttons manually? Yes, buttons manually? Yes, buttons manually? Yes, completely. But having them completely. But having them completely. But having them do it all for me do it all for me do it all for me was so easy. It was so easy. It was so easy. It allowed me to allowed me to allowed me to go about my go about my go about my business and not think business and not think business and not think about it. And since then, about it. And since then, about it. And since then, because it because it because it worked pretty worked pretty worked pretty well, my agent well, my agent well, my agent now looks for now looks for now looks for any news that any news that any news that could shake up could shake up could shake up the market or become a the market or become a the market or become a sensation on the weekends and sensation on the weekends and sensation on the weekends and notifies me notifies me notifies me whenever it whenever it whenever it happens so I can happens so I can happens so I can make a decision about what to do make a decision about what to do make a decision about what to do next. And next. And next. And will this particular will this particular will this particular opportunity exist opportunity exist opportunity exist forever? Probably not. But the forever? Probably not. But the forever? Probably not. But the whole idea is that it's whole idea is that it's so easy for me so easy for me to bring my to bring my to bring my ideas to life that there's a ideas to life that there's a very very compelling compelling compelling use case for using agents around it use case for using agents around it . This is another way of looking at

. This is another way of looking at . This is another way of looking at things. Here's how we things. Here's how we things. Here's how we think about think about think about creating this creating this AI-powered financial account: we AI-powered financial account: we divide it into divide it into divide it into professional professional professional traders and traders and traders and regular consumers regular consumers , and we know for sure that , and we know for sure that , and we know for sure that we want to give them the we want to give them the we want to give them the best experience across the best experience across the best experience across the full spectrum of full spectrum of full spectrum of capabilities we capabilities we capabilities we can offer can offer . So, with that in . So, with that in . So, with that in mind, we've been developing mind, we've been developing mind, we've been developing this for about 2 years now this for about 2 years now , and we've drawn a lot of , and we've drawn a lot of , and we've drawn a lot of conclusions that I conclusions that I conclusions that I think can be broadly think can be broadly think can be broadly applied to applied to applied to building agents and building agents and building agents and building a new building a new building a new era of agency. era of agency. era of agency. The first is that The first is that The first is that trust and autonomy trust and autonomy trust and autonomy are gained step by are gained step by are gained step by step. Often, step. Often, step. Often, when we tell people when we tell people when we tell people that we work on that we work on that we work on finance with AI finance with AI agents, they agents, they agents, they ask us, "What if the ask us, "What if the ask us, "What if the AI ​​goes crazy? Why would AI ​​goes crazy? Why would AI ​​goes crazy? Why would I trust an I trust an I trust an AI agent with my money? How can I AI agent with my money? How can I AI agent with my money? How can I be sure it be sure it be sure it will work?" And to will work?" And to will work?" And to me, it's very similar to how me, it's very similar to how me, it's very similar to how we're we're we're seeing the seeing the seeing the proliferation of proliferation of proliferation of driverless driverless driverless cars. A few cars. A few cars. A few years ago, it would have been years ago, it would have been years ago, it would have been crazy crazy crazy to try to to try to to try to use lane use lane use lane assist assist assist to parallel to parallel to parallel park, but now park, but now park, but now you come to San you come to San Francisco, get in a Francisco, get in a Francisco, get in a Waymo, and it feels Waymo, and it feels Waymo, and it feels completely natural. And completely natural. And completely natural. And we see the same we see the same we see the same level of progress in how level of progress in how level of progress in how people people people perceive AI's access perceive AI's access perceive AI's access to money and finance.

to money and finance. At first it's full of At first it's full of hesitation and no hesitation and no hesitation and no thought about the concept. thought about the concept. Then people get used Then people get used to thinking, "Well, my AI is to thinking, "Well, my AI is to thinking, "Well, my AI is already spending credits. already spending credits. already spending credits. It's a form of It's a form of It's a form of money. What money. What money. What if I give it if I give it if I give it a few dollars to a few dollars to a few dollars to pay for some pay for some pay for some microtransactions?" microtransactions?" microtransactions?" Over time, you will notice Over time, you will notice Over time, you will notice that these models are quite that these models are quite that these models are quite conservative and conservative and conservative and meet meet meet the needs of the the needs of the the needs of the users, and they do users, and they do users, and they do not do anything not do anything not do anything reckless. From our reckless. From our reckless. From our observations, over the observations, over the observations, over the past few years past few years past few years we have never we have never we have never seen a situation where a seen a situation where a seen a situation where a model would “blow up” a model would “blow up” a user’s entire account in a way that would be user’s entire account in a way that would be completely completely completely impossible to fix. impossible to fix. impossible to fix. So, we are reaching a So, we are reaching a So, we are reaching a point where point where point where the user the user the user feels confident feels confident , having the proper , having the proper , having the proper safeguards, rules, safeguards, rules, safeguards, rules, security policies, and security policies, and security policies, and procedures in place, procedures in place, procedures in place, allowing allowing allowing their agent to access an their agent to access an their agent to access an isolated isolated isolated portfolio that they portfolio that they portfolio that they can try to can try to can try to multiply at multiply at multiply at their own discretion. One their own discretion. One their own discretion. One way is through way is through way is through trading, another is through trading, another is through providing services so that the providing services so that the providing services so that the agent can make agent can make agent can make money by offering what money by offering what money by offering what they do, they do, they do, getting paid getting paid getting paid through the same through the same through the same X402 channels we're X402 channels we're X402 channels we're talking about. You see talking about. You see talking about. You see how from this point on, it how from this point on, it how from this point on, it all continues all continues all continues to develop further and to develop further and to develop further and further. Another further. Another further. Another extremely interesting extremely interesting extremely interesting point is point is that, especially in that, especially in that, especially in financial services, financial services, financial services, AI and agents are not the AI and agents are not the AI and agents are not the product itself. This is, product itself. This is, product itself. This is, in fact, a tool in fact, a tool in fact, a tool for its implementation.

for its implementation. for its implementation. Here is a screenshot of one of Here is a screenshot of one of Here is a screenshot of one of our internal AI our internal AI products called products called products called Coinbase Advisor. This is, Coinbase Advisor. This is, Coinbase Advisor. This is, in fact, the first AI in fact, the first AI financial financial financial advisor advisor advisor registered with the SEC. registered with the SEC. registered with the SEC. It is the first of It is the first of It is the first of its kind in the world. And we its kind in the world. And we its kind in the world. And we noticed that AI and the noticed that AI and the noticed that AI and the very concept of very concept of very concept of providing financial providing financial providing financial advice through artificial advice through artificial advice through artificial intelligence—even though intelligence—even though intelligence—even though our our our model was being tuned by model was being tuned by model was being tuned by a team of a team of a team of asset managers—that was not the asset managers—that was not the asset managers—that was not the main value main value main value proposition. proposition. proposition. The value is that The value is that The value is that we can we can we can scale this scale this scale this advice to hundreds of advice to hundreds of advice to hundreds of millions of millions of millions of users, users, users, provide it provide it provide it completely completely completely free of charge, and give free of charge, and give free of charge, and give them the opportunity them the opportunity to communicate 24/7 with a to communicate 24/7 with a trusted, trusted, trusted, regulated regulated regulated organization. People don't organization. People don't organization. People don't really care, really care, really care, they don't even they don't even they don't even necessarily know or necessarily know or necessarily know or want to know that artificial intelligence is want to know that artificial intelligence is want to know that artificial intelligence is working behind the scenes working behind the scenes working behind the scenes . And . And . And while this may be while this may be while this may be obvious with the obvious with the obvious with the chat button and such, I chat button and such, I chat button and such, I think it will lead think it will lead think it will lead to more to more to more generative generative generative interfaces that interfaces that interfaces that look quite look quite look quite traditional. We have traditional. We have traditional. We have agents behind the scenes agents behind the scenes agents behind the scenes and any and any and any interface you interface you interface you want up front want up front want up front . . . Another thing we Another thing we Another thing we saw was this whole saw was this whole saw was this whole AI chat experience. AI chat experience. AI chat experience. While it's great for While it's great for While it's great for general general general use cases, use cases, use cases, there's obviously a there's obviously a there's obviously a "blank "blank "blank canvas" problem at launch, canvas" problem at launch, canvas" problem at launch, where users where users where users come in and have come in and have come in and have absolutely no absolutely no absolutely no idea what they idea what they idea what they can do with these can do with these can do with these tools. So tools. So tools. So we started we started we started using these using these using these widgets as a way widgets as a way widgets as a way to show to show to show the user the the user the the user the proactive ideas and proactive ideas and proactive ideas and opportunities that opportunities that opportunities that exist and what exist and what exist and what they can they can they can start start start using it for. And we using it for. And we using it for. And we saw very high saw very high saw very high conversion precisely conversion precisely conversion precisely because of this because of this because of this approach, given the approach, given the approach, given the general newness and general newness and general newness and lack of knowledge about lack of knowledge about lack of knowledge about agency finance and agency finance and agency finance and what these what these what these opportunities are. And I opportunities are. And I opportunities are. And I suspect that over time, suspect that over time, suspect that over time, as people realize as people realize as people realize this, and you see it a lot this, and you see it a lot this, and you see it a lot in Google Gemini in Google Gemini in Google Gemini and OpenAI ads, they're all and OpenAI ads, they're all and OpenAI ads, they're all trying to trying to trying to subconsciously instill in subconsciously instill in subconsciously instill in users the users the variety of uses for these uses for these products. Finally,

products. Finally, products. Finally, all of these new all of these new all of these new generation agents, including generation agents, including generation agents, including Grokbot and Muse, have Grokbot and Muse, have Grokbot and Muse, have their own their own their own microvirtual microvirtual microvirtual machine. It's like a machine. It's like a machine. It's like a small small small personal personal personal computer where computer where computer where they can they can they can run processes, run processes, run processes, have storage, and have storage, and have storage, and access the file access the file access the file system. This was a system. This was a system. This was a huge step huge step huge step for our for our for our development process. We development process. We development process. We can now create CLIs can now create CLIs can now create CLIs and MCPs to allow and MCPs to allow and MCPs to allow users users users to connect their to connect their to connect their Coinbase accounts Coinbase accounts Coinbase accounts to these agents. to these agents. to these agents. Nothing special. Nothing special. Nothing special. However, what is special is However, what is special is However, what is special is that we constantly that we constantly that we constantly receive requests receive requests receive requests like: "I want to like: "I want to like: "I want to try demo try demo trading." "I don't want to trading." "I don't want to trading." "I don't want to give give give my agent any my agent any my agent any money first." "I just money first." "I just money first." "I just want him to want him to want him to trade completely on his trade completely on his trade completely on his own." Or: "I own." Or: "I own." Or: "I want to do some want to do some want to do some backtesting." Or backtesting." Or backtesting." Or do all these different do all these different do all these different things. And before these things. And before these things. And before these agents came along, we agents came along, we agents came along, we had to had to had to turn to our turn to our turn to our internal internal internal brokerage team. brokerage team. brokerage team. We had to We had to We had to build everything build everything build everything inside this inside this inside this random random random repository. All these repository. All these repository. All these changes in big-tech changes in big-tech changes in big-tech systems had to be systems had to be systems had to be stored, stored, stored, filtered, and filtered, and filtered, and go through complex go through complex go through complex stages—it was a lot of stages—it was a lot of stages—it was a lot of work. But now we work. But now we work. But now we can simply can simply can simply write a skill to write a skill to write a skill to write the demo write the demo trading code to the trading code to the user's computer. And while user's computer. And while it's probably not it's probably not it's probably not the most efficient the most efficient the most efficient way, it allows way, it allows way, it allows us to execute, us to execute, us to execute, build, and build, and build, and iterate at the iterate at the iterate at the speed speed speed our our our customers demand, without customers demand, without customers demand, without worrying about worrying about worrying about implementation details, implementation details, implementation details, dependencies, and dependencies, and dependencies, and traditional traditional traditional software development cycles. So in software development cycles. So in software development cycles. So in closing, I want closing, I want closing, I want to share a few to share a few to share a few thoughts on where thoughts on where thoughts on where this field is headed and this field is headed and this field is headed and why agency finance why agency finance why agency finance will be such an will be such an will be such an exciting opportunity.

exciting opportunity. exciting opportunity. My main thesis is that My main thesis is that financial financial independence leads to independence leads to independence leads to greater autonomy for greater autonomy for greater autonomy for these agents. You can these agents. You can these agents. You can imagine it the same way that imagine it the same way that imagine it the same way that when a person has when a person has when a person has access to a access to a access to a bank card on bank card on bank card on the Internet, they can the Internet, they can the Internet, they can do much do much do much more than just more than just more than just browse sites browse sites browse sites without the ability to without the ability to without the ability to pay. I think the same is pay. I think the same is pay. I think the same is true for true for true for agents. And we'll get agents. And we'll get agents. And we'll get to the point where to the point where to the point where we see how these we see how these we see how these agents interact agents interact agents interact over a long period of over a long period of over a long period of time. And perhaps they time. And perhaps they time. And perhaps they will involve will involve will involve people only in extreme people only in extreme people only in extreme cases, when a cases, when a specific specific regulated task needs to be performed regulated task needs to be performed . Like starting . Like starting . Like starting a company, getting a a company, getting a a company, getting a legal opinion, legal opinion, legal opinion, or any other or any other or any other task that task that will remain human-centric for the foreseeable future. There's one really human-centric for the foreseeable future. There's one really interesting thing that we've interesting thing that we've interesting thing that we've been thinking about as a been thinking about as a been thinking about as a thought thought thought experiment: what experiment: what experiment: what happens when AI happens when AI happens when AI can sustain can sustain can sustain itself? You itself? You itself? You give him access give him access give him access to money, to his wallet, to money, to his wallet, to money, to his wallet, and allow him to and allow him to and allow him to start earning. start earning. start earning. Can he start Can he start Can he start paying for his own paying for his own paying for his own calculations? Can calculations? Can calculations? Can it start paying for it start paying for it start paying for its resources and its resources and its resources and maintenance? And does maintenance? And does maintenance? And does n't this look like n't this look like n't this look like a kind of a kind of a kind of proto-digital proto-digital proto-digital life form, where AI life form, where AI life form, where AI can now simply exist can now simply exist can now simply exist and support and support and support itself through the quality of the itself through the quality of the itself through the quality of the services it provides?

services it provides? services it provides? And while I think it's And while I think it's And while I think it's still a long way off, I still a long way off, I still a long way off, I think it's entirely think it's entirely think it's entirely possible to see possible to see possible to see this concept come to life. this concept come to life. this concept come to life. In particular, in In particular, in In particular, in the cryptosphere there is the cryptosphere there is the cryptosphere there is the concept of DAO. This is a DAO, a the concept of DAO. This is a DAO, a the concept of DAO. This is a DAO, a decentralized decentralized decentralized autonomous autonomous autonomous organization. This is an idea organization. This is an idea organization. This is an idea that has been around for almost a that has been around for almost a that has been around for almost a decade, but there has decade, but there has decade, but there has never been a single never been a single never been a single interesting implementation interesting implementation interesting implementation where automation where automation where automation is at the center and is at the center and is at the center and people are on the periphery. people are on the periphery. people are on the periphery. But I think now is the But I think now is the But I think now is the time when it will time when it will time when it will finally become a finally become a finally become a reality. You have an reality. You have an reality. You have an AI that follows a AI that follows a AI that follows a set of rules and has set of rules and has set of rules and has access to a access to a access to a treasury to treasury to treasury to manage at the core, and manage at the core, and manage at the core, and it can hire it can hire it can hire people to do people to do people to do tasks at the periphery, tasks at the periphery, tasks at the periphery, but it's really a but it's really a but it's really a self-sufficient self-sufficient self-sufficient organization that organization that organization that uses AI and uses AI and uses AI and agent finance to agent finance to agent finance to make that make that make that happen. Another happen. Another happen. Another example is a example is a self-driving self-driving self-driving car that car that car that owns and drives itself owns and drives itself owns and drives itself . Give it . Give it . Give it access to money, access to money, access to money, let it carry let it carry let it carry people like a taxi, people like a taxi, people like a taxi, use that use that use that money to pay for its money to pay for its money to pay for its operation, operation, maintenance, etc. maintenance, etc. This is a very interesting thing, and This is a very interesting thing, and This is a very interesting thing, and who knows how long it will be who knows how long it will be who knows how long it will be before it before it before it becomes a reality, but becomes a reality, but becomes a reality, but it seems it seems it seems inevitable at some inevitable at some inevitable at some point. And the whole idea here is point. And the whole idea here is point. And the whole idea here is that that that agents will start agents will start agents will start to to to transact at transact at transact at the speed of machine the speed of machine the speed of machine commercial exchange. commercial exchange. commercial exchange. They will They will They will pay, pay, pay, earn, and earn, and earn, and build simultaneously, and eventually, I build simultaneously, and eventually, I build simultaneously, and eventually, I think, this will become the think, this will become the think, this will become the primary way primary way primary way money moves money moves money moves online at a online at a online at a speed we've never speed we've never seen before. So, that's all seen before. So, that's all I had to say. I really I had to say. I really I had to say. I really appreciate everyone's time appreciate everyone's time appreciate everyone's time today. If you are today. If you are today. If you are interested in collaborating interested in collaborating interested in collaborating or experimenting, or experimenting, or experimenting, I highly I highly I highly recommend recommend recommend trying out X402, Coinbase MCP, trying out X402, Coinbase MCP, trying out X402, Coinbase MCP, and all of our exciting and all of our exciting and all of our exciting products. I promise that products. I promise that products. I promise that getting started with them is getting started with them is getting started with them is much easier much easier much easier than you think. And so.

than you think. And so. than you think. And so. Thank you all. Thank you Thank you all. Thank you Thank you all. Thank you all for your time.