Nemil Dalal from Coinbase discusses the x402 protocol, a new open standard that enables AI agents to make cryptocurrency payments for online resources using the long-dormant HTTP 402 "Payment Required" status code.

Watch Episode Here

Read Episode Description

Nemil Dalal from Coinbase discusses the x402 protocol, a new open standard that enables AI agents to make cryptocurrency payments for online resources using the long-dormant HTTP 402 "Payment Required" status code. The conversation covers stablecoin fundamentals, including how they work and who maintains their stability, before exploring how crypto wallets with multi-signature capabilities could protect AI agents from adversarial attacks while keeping humans in control. They examine forward-looking questions about whether this technology could finally create a viable alternative to advertising models, what happens when millions of AI agents have wallets, and the need for reputation systems in an agent economy. The discussion highlights how cryptocurrency, originally designed to be beyond nation-state control, might become the perfect enabling technology for AI economic participation.

Sponsors:
Google Gemini 2.5 Flash: Build faster, smarter apps with customizable reasoning controls that let you optimize for speed and cost. Start building at https://aistudio.google.com

Oracle Cloud Infrastructure: Oracle Cloud Infrastructure (OCI) is the next-generation cloud that delivers better performance, faster speeds, and significantly lower costs, including up to 50% less for compute, 70% for storage, and 80% for networking. Run any workload, from infrastructure to AI, in a high-availability environment and try OCI for free with zero commitment at https://oracle.com/cognitive

The AGNTCY: The AGNTCY is an open-source collective dedicated to building the Internet of Agents, enabling AI agents to communicate and collaborate seamlessly across frameworks. Join a community of engineers focused on high-quality multi-agent software and support the initiative at https://agntcy.org/?utmcampaig...

NetSuite by Oracle: NetSuite by Oracle is the AI-powered business management suite trusted by over 42,000 businesses, offering a unified platform for accounting, financial management, inventory, and HR. Gain total visibility and control to make quick decisions and automate everyday tasks—download the free ebook, Navigating Global Trade: Three Insights for Leaders, at https://netsuite.com/cognitive


PRODUCED BY:
https://aipodcast.ing

CHAPTERS:
(00:00) Sponsor: Google Gemini
(00:31) About the Episode
(05:05) Introduction and Stablecoin Basics
(09:05) Trust and Transparency
(12:31) Economics and Yield
(16:29) Business Model Sustainability (Part 1)
(19:53) Sponsors: Google Gemini | Oracle Cloud Infrastructure
(21:32) Business Model Sustainability (Part 2)
(21:32) Transaction Costs Evolution
(26:22) Blockchain Scaling Solutions
(30:04) Introducing X402 Protocol (Part 1)
(34:25) Sponsors: The AGNTCY | NetSuite by Oracle
(36:40) Introducing X402 Protocol (Part 2)
(38:01) Micropayments Adoption Barriers
(42:28) X402 Technical Implementation
(48:20) Agent Builder Integration
(51:36) KYC and Reputation
(55:37) Agent Identity Challenges
(01:00:19) Risk Management Strategies
(01:08:39) Systemic Risks Considerations
(01:15:47) AI-Powered Smart Contracts
(01:21:31) Closing Thoughts
(01:22:41) Outro

SOCIAL LINKS:
Website: https://www.cognitiverevolutio...
Twitter (Podcast): https://x.com/cogrev_podcast
Twitter (Nathan): https://x.com/labenz
LinkedIn: https://linkedin.com/in/nathan...
Youtube: https://youtube.com/@Cognitive...
Apple: https://podcasts.apple.com/de/...
Spotify: https://open.spotify.com/show/...

TRANSCRIPT

Introduction

Hello, and welcome back to the Cognitive Revolution!

Today my guest is Nemil Dalal, Developer Platform Lead @ Coinbase, who recently co-created the x402 protocol – a new open standard, designed to enable both humans & AI agents to transact seamlessly on the internet with cryptocurrency, that takes its name from the HTTP 402 "Payment Required" status code, which believe it or not has existed since the early days of the internet, but was never implemented due to the limitations of traditional payment systems.

As we've covered in a number of recent episodes, AI agents are now quickly becoming more capable and reliable, such that it's no longer crazy to think about given them modest budgets to spend on access to information, APIs, or MCPs, or perhaps even to hire other Agents or human collaborators to help them when needed.  

However, because traditional payment systems are designed for human users, and because they balance security-related trade-offs very differently, they're often very difficult for AI agents to use, and generally priced in a way that makes micropayments uneconomical.

x402 seeks to fix that in a way that's easy for sellers to implement and buyers to use.  Anyone who wishes to charge for access to an online resource can, by adding the x402 library to their application, return a 402 response that specifies the details of the payment required for access.  The user, which could be an AI agent equipped with a crypto wallet, can then make the necessary payment to the specified address and re-request the resource with proof of payment attached.

Now, it's worth noting that while I have used Coinbase to send international payments with the USDC stablecoin, I am not very knowledgeable about the crypto space in general, so… wanting to properly understand the fundamentals of this system, I asked Nemil, who previously ran USDC at Coinbase, to give me a primer on stablecoins in general, including their benefits to retail users, who we are trusting to keep them stable, and how those actors, including Coinbase, benefit from providing these services.  For some, this may be remedial content, but for me it was very helpful.  

From there, we get into more forward-looking possibilities.  

Among other things, I was interested to learn how crypto wallets, which can be programmed to require  multi-signature approval for certain types of transactions, seem very useful for protecting AI agents from adversarial attacks, to which they remain quite vulnerable, and more generally keep humans in control.

Beyond that, we consider what are as-yet unanswered questions:

• Could this be the technology that finally creates a viable alternative to the advertising model? 
  
• Is there any way to anticipate the dynamics that might arise when millions of AI agents have wallets, especially in light of all the scheming and other bad behaviors that we're seeing from the latest wave of intensively Reinforcement Learning trained reasoning models?
  
• What sorts of reputation systems might we need and how might they work given the fact that agents are so easy and cheap to create and delete?
  
• Should AI agents have to be "capitalized" to ensure accountability, as Tyler Cowen has suggested?
  
• And what are the prospects for AI "putting the smart in smart contracts" and greasing the wheels of local commerce by using AIs models to provide low-cost resolution for offline disputes?

All these are fascinating possibilities that we should all be watching closely as the agent economy continues to accelerate.

And big picture: this discussion is also a reminder of the contingency of the technologies that have enabled the AI wave.  Most famously, the development of GPUs, aka Graphics Processing Units, was driven for years by the unique demands of video game rendering engines.  And here again, cryptocurrency, which was motivated by a desire to create digital money that not even nation states could control, might prove to be the perfect enabling technology to allow AIs to participate in the economy.

I see this both as a reason to be humble about our ability to predict where all this is going, but also as a challenge to technologists and technology investors to invest now in the development of potentially complementary technologies that will enable supervision, monitoring, and control of the AI agent economy. 

I will definitely be returning to that theme in future episodes, but for now, I hope you enjoy this primer on cryptocurrency and exploration of how AI agents might get their first access to economic participation, with Nemil Dalal from Coinbase.

Main Episode

Nathan Labenz: Nimil Dalal, developer platform lead at Coinbase, welcome to The Cognitive Revolution.

Nemil Dalal: It's fantastic to be here. Thanks for having me.

Nathan Labenz: I'm excited for this conversation. Honestly, I have a lot to learn. Regular listeners know that I'm totally obsessed with the AI technology wave and study it from all angles. But the same is not true of the crypto technology wave, so I'll probably ask some remedial or basic questions to get started. And with that introduction and helpful orientation from you, I know we've got some envisioning to do of what the intersection of the AI and crypto tech waves might look like, and specifically, a protocol that you recently put out to try to make this more concrete and really start to enable it. So, for starters, because this is going to be a foundational technology for the purposes of the vision we'll be sketching out, can you tell me where we are on stablecoins today? I think everybody knows that Bitcoin goes up and down and it can be crazy. But then there's this whole domain of stablecoins, which I've used a bit through Coinbase to pay international contractors from time to time. It does have a magical feel to it where I can just drop some dollars into the Coinbase account, convert them to USDC, and send them seemingly anywhere in the world. On the other side of that, I'm not entirely sure exactly what happens, and I'm not entirely sure how it works. But maybe you can give me a little Stablecoin 101, which I think not only I but probably a number of listeners would really benefit from.

Nemil Dalal: I'd love to. I used to run USDC at Coinbase when it was first venture-backed, just a few tens of millions in market cap, and helped grow it to a billion, so I'm really excited to talk about it. So, step one: in the last decade, we went from zero stablecoin usage in the world to about 250 billion today. There are a number of different players in the market. Coinbase is one of the really big ones, along with Circle, which is USDC. If you think about the blockchain, there are a lot of benefits. One, it's globally available. Anyone can use it. Two, fees are often much lower than traditional financial system rails. Three, it's often instantaneous, a matter of seconds to send this anywhere in the world. These are all the benefits of blockchains. We've known about this for a very long time, since Bitcoin. A lot of these benefits started coming together. But what was always hard was that there wasn't stability. The power of something like the US dollar, the euro, or any of these other fiat currencies is that it has stability. People buy things with it. They go to a supermarket, they go to a drugstore, they can buy things with it. So the power of a stablecoin is it combines all the benefits of a blockchain but has the stability of the US dollar. You asked some more detailed questions about this, and I'll share a few areas I think are interesting for folks to know about. One is how does it get its stability? Typically, the way this works is what the issuer does, in this case Circle and Coinbase, is they take dollars they're getting, put them into a bank account, or into treasuries, basically some type of short-dated government paper, and then they issue a token one-to-one against it. So the way it gets its stability is that you have those tokens in a bank account, issuing them one-to-one. So if I have $10, somebody gives me $10, I can create 10 USDC tokens and give them to that user. They can then do anything they want with them, and if they ever want to come back and get the underlying fiat back, I change that back. That's the power of this. Now, this was the vision 10 years ago. What's changing these days, or maybe I'll just spotlight one or two quick things. One, blockchains are getting faster and cheaper. Base nearly had 1,000 transactions per second. Settlement times are a matter of less than 500 milliseconds today, which is the plan for what Base is putting together. Fees are basically cents or less, so there could be sub-cent at different moments in time. All these things suddenly mean that blockchain, which was good before, is now essentially exceptional. So the vision we really think about is, can you make sending money as simple as sending a text message? And that's the vision with stablecoins and where we're headed.

Nathan Labenz: It's a great start. A couple questions for somebody, again, who doesn't understand all this deeply in a technological sense. Obviously, one of Bitcoin's big ideas originally was its trustless nature; you don't need to trust anyone in particular, right? You're just trusting this decentralized network, and somebody would have to take over the majority of the network to launch an attack. Over all these years, it's been robust to that sort of thing. When we have a stablecoin, am I trusting someone? Who am I trusting? Is there a concept of a fractional reserve behind that, or is it literally one-to-one? How do I know as an end user when I have USDC that there really is the ability to redeem it? Because I think many times, as I said, I have done these transactions, and all of the surface-level user features I have experienced, but I'm not quite sure exactly what is underlying that and providing that sense of stability.

Nemil Dalal: For stablecoins, unlike Bitcoin, you absolutely have to trust someone. That is a big difference, and the person you are trusting is an issuer. In this case, Circle and Coinbase are the issuers, and you have to trust that they are taking the money and putting the funds where they claim to put them. This is one of the risks that comes with stablecoins. Circle and Coinbase address this by sharing transparency reports. They show a radical degree of candor regarding where those funds are going and where they are held. With that degree of transparency and trust, Congress and others are exploring ways to legislate stablecoins to make them even more trusted globally. You absolutely need a form of trust, and it is typically on the issuer to ensure this. The other main thing is ensuring the keys for the mint and burn of the smart contracts are safely held so that no one can infinitely create USDC without underlying reserves. Typically, fiat-level transparency and smart contract security are needed to support this network.

Nathan Labenz: How do you manage this? When I go onto Coinbase, bring my dollars, convert them, and then make a payment, that payment may go anywhere in the world. On the other end, my counterparty, who may or may not be working with a Coinbase account or some other wallet provider or enabling technology. If they are in India, for example, they will cash out into local currency, getting rupees at the end of the day. How is this done with no fees? Obviously, there are significant fees if I try to do that through the traditional banking system. Even if I imagine this is done as a public good to enable the ecosystem, there would still be currency risk. It feels like there could be some risk of buildup or currency movement. Being the market maker does not sound like a great place to be if you are making no fees on it. I am a little confused as to who is taking on that risk, which seems like it cannot be totally eliminated, and why they are taking on that risk.

Nemil Dalal: With any product like this, you need market makers and cash-in and cash-out points, and both of those require money. There is compliance infrastructure involved, physical branch locations, and money movement as you go back and forth between different currencies to keep the books healthy in a financial exchange. Absolutely, all of these parties need to make money as part of this. The secret of stablecoins is that money is often made on the yield itself. When I take $10 and convert that for the stablecoin, that money goes into some form of short-dated treasury or a bank account, and yield is earned on the other side. For example, today the stablecoin market cap is about $250 billion, almost exclusively US dollar-based stablecoins. There is a three to four, or four and a half percent yield being made on those funds, which generates significant economics to support an entire ecosystem. Candidly, this exists in the traditional financial world, and banks take all that and still charge fees. The idea of this methodology is that you can make fees from that, but then offer very cheap or free service that anyone can use. Let me tell you a story of how a user might approach this: they come to Coinbase, wanting to convert $10 in their bank account into stablecoins. They hook up their bank account to Coinbase. Coinbase will pay some ACH fees for that, which might be a few cents in the US for ACH. It takes two to three days to move over. Those $10 show up at Coinbase, and then you suddenly get 10 stablecoins, 10 USDC tokens, in your wallet. Now you can send them anywhere in the world. They do not have to be a Coinbase user. All they need is a crypto wallet on the other side. So let us say it is someone in India, as you said. That person in India will suddenly have 10 USDC. There are a few different ways this could go. One is they decide to keep the USDC, put it into a DeFi protocol, earn some yield on it, or spend it for some goods. Slowly, we are starting to see this happen more and more, but it is still early days. Not every merchant on the local street in India accepts this. By the way, I am Indian, so this is a cause near and dear to my heart. You cannot just go spend that at local stores. But what you can do is sell it. There is a market maker on the other side who collects the USDC, deciding whether to keep it in USDC or go back to Coinbase or Circle and convert it back to dollars. On my side, I am converting that, probably paying some type of fee to do so. In India, I would likely have to pay a fee. It might be half a percent or one percent, or a few cents depending on the dollar or rupee amount. Then I suddenly have my rupees and can spend them wherever I want. That is the system as it is today, but I think we are rapidly moving to a closed-loop world where I have a stablecoin. A stablecoin on-chain is better than a fiat stablecoin or a fiat rupee. A rupee on-chain is better because it can be sent around the world, it can be cheap, and it can be used with this entire decentralized finance infrastructure. I do not know if your audience is aware of that, but there are things like yield protocols, the ability to trade, and many traditional financial services people are used to, including ways to store it. These are all the things you can do once you have that on-chain rupee or on-chain dollar.

Nathan Labenz: It's a lot for me to wrap my head around as someone who hasn't done it much. But I understand your point about the fees on the cash side; it makes sense. Would it be fair to say that we are in an early, almost subsidized era of this from a Coinbase perspective? Famously, Uber was burning, Uber and Lyft were burning tons of cash to scale their operation. It's a joke at this point that the millennial lifestyle was subsidized by venture capital for a long time.

Nemil Dalal: Yes.

Nathan Labenz: Are we in a similar period with crypto where Coinbase is willing to take an initial customer acquisition loss to get people onto the platform and into the chain lifestyle in the first place?

Nemil Dalal: No. This is already a big business. That's the crazy part: when you earn 4% yield, and there's $250 billion of stablecoins, that means the people who issue stablecoins are making over 10 billion a year already. The Uber example is where they're making very little and using VC to subsidize that, but the reality is that these are already great businesses. These are fantastic businesses. If you have less margin than a bank, you can do many amazing things with that. So, they're making good money, and as a result, there are many different things you can do with these stablecoins. Does that make sense?

Nathan Labenz: Yes, although I also get yield on my US... I haven't held USDC much-

Nemil Dalal: Yes.

Nathan Labenz: but I do notice when I deposit dollars into Coinbase and then convert them, I am suddenly getting yield, right? So I guess there's obviously just a spread between what Coinbase is giving you and what I'm getting?

Nemil Dalal: That's exactly right. Coinbase is making money. There's also value to Coinbase for USDC continuing to grow, and that's one of the reasons when you have it on our platform, for example, we offer you yield directly on top of it. But again, if you took that off platform and just had it in your wallet for a year, you would not be earning yield on that, and Coinbase and Circle would be earning yield together. Essentially, you can either move it to Coinbase or somewhere else where you earn that yield, or you can hold it off and Coinbase is still earning the yield, very similar to how a bank might earn that yield.

Nathan Labenz: Yes. Interesting. Okay. Just-

Nemil Dalal: And then-

Nathan Labenz: are. Yes. Let's-

Nemil Dalal: add is that blockchain fees are another really interesting thing. Why does it cost so much to send a wire transfer in a world where an SMS around the world is free? A lot of what we're sending is just information. It's not necessarily the money that is always moving because there are fiat on and off-ramps and various other ways to send things. What the blockchain has been amazing at is suddenly making this a much more commoditized and open market for anyone to move. The first version of blockchains, I remember in 2017, was about two to five dollars. Then gas fees spiked, getting to one hundred, two hundred dollars to make an on-chain transaction. But today, with Base and others, it's a matter of one or two cents to send this. That's not with subsidies or anything. That is the power of the blockchain. The blockchain allows you to have this open market where anyone can send transactions, and this technology is much better than what it was for traditional banking rails. I've worked earlier in my career for these banks. A lot of this is 1950s, 1960s, 1970s infrastructure they are running on. It's very hard to convince many banks to change their technology stack. The power of the blockchain, just like the internet, is that you have an old system, and you have a new system built with the whole new world in mind. As a result, everything is much cheaper and faster. So, I think that's the other thing that surprises people: the fees are so low, and that is not a function of subsidies. It is a function of the technology architecture and the fact that networks have generally gotten much better over the last 30-40 years, but it just hasn't flowed through to finance yet.

Nathan Labenz: Yes. So maybe two follow-ups on that. One is on a somewhat different services profile, and the other is on the underlying technology enabling low transaction costs, which is a good transition into the new 402, payment required, X.402 protocol. On the services side, there is some difference. Perhaps you can compare and contrast what I get, because part of what I do get if I send through the traditional system is recourse, right? I can say the money didn't get there, or somebody defrauded me, or it wasn't actually me that made this payment. There are various fraud vectors that I can engage the traditional banking system on. My understanding is that a non-trivial fraction of the fees associated with those transactions are basically to cover the fraud that gets through and to pay for the services of remediating fraud that people are constantly attempting. My general sense is that there's much less of that on the crypto side. So my trade-off is that I am paying much less in fees, but I also have qualitatively less redress available to me if something goes wrong. Complicate or clarify that picture for me.

Nemil Dalal: Absolutely. Exchanges around the world have forms of KYC and other checks and balances on their systems. So there are some basic forms of that. However, the core of your message is correct: unlike more traditional systems, this is not a reversible system by default. When you send a transaction on a blockchain, it ends up on the other side, and that's it. There's no recourse to pull them back by default. But there are interesting things baked into this. For example, stablecoins generally work with national governments and partners. If there's a legal case associated with USDC, and an enforcement agency reaches out, that is absolutely something baked into the protocol. There is the ability to hold those funds. Coinbase, any other exchange, Circle, and others all need to follow sanctions law that the US has and other international jurisdictions. These are absolutely things that can be done. So there is some form of recourse, but you're absolutely right, it isn't like pressing a button to suddenly reverse the transaction. That's also why it's so fast. If you can press a button and reverse a transaction, the other side can't truly trust that they've received the money yet. This is why for ACH, you might have to wait several days to access your funds, while with blockchain, it might be two seconds, and you suddenly have access. Blockchains can also add reversibility in more complex ways into the stack if they want to. For instance, one layer on top could be built where the foundational layer is not reversible, but this top layer allows you to wait for two days, and the sender can pull the money back. These are all things that people can build, and in fact, people have built them. They are not super popular yet, but I expect over time they will become more popular. Regarding the broader point about fees, yes, that is part of the reason for the fees, but I don't think it justifies all of them. That's probably a big shift. You're right that if you added a better system there, their fees would probably be a little bit higher than they are today. But compared to a wire, which people pay $20-$30 for, I can send a trillion dollars or 10 cents for less than a cent today on the blockchain. That is a really new feature of blockchains compared to the traditional financial system.

Nathan Labenz: So, my last question before we get to it: can you sketch out how transaction costs on blockchain have fallen so dramatically? This is something that even a crypto-ignorant, AI-obsessed crowd will have some sympathy for, because we're used to paying for compute. My general, high-level understanding is that transactions used to be backed by the proof-of-work concept, which is essentially proof of compute. A lot of those compute requirements have been taken out and replaced by other schemes. We're also seeing dramatic efficiencies on the AI side, not necessarily by removing compute, but by optimizing it. So, what's a very basic explanation of how transaction costs went from spiking to $100 or whatever, to now being so low?

Nemil Dalal: The overall message for your listeners is that just like there's a Moore's Law for bandwidth, where bandwidth gets better over time, there's a similar principle for blockchains. Bitcoin launched in 2009, giving us 16 years to refine and innovate ways to make it cheaper. The biggest shift is the increasing availability of block space. For example, in the Ethereum ecosystem, back in 2017, if you wanted to launch a transaction, you put it directly on the Ethereum Layer 1 blockchain. Today, you have Layer 2 blockchains on top, and even Layer 3s. They take all this information, allow people to make transfers, then summarize it and put it onto the Layer 1 blockchain. This uses a lot less storage, and storage is also increasing on the Layer 1 side. As a result, the amount of block space has gone up significantly. This is not dissimilar from the traditional banking system, where banks don't settle every single transaction individually. They might settle the net over time. While not exactly the same, I liken it to a model where Layer 2s handle many different transactions, and periodically write something called the Beacon Chain in Ethereum. That's one way to do it. There are many other ways to scale. Another example is Bitcoin's strong focus on decentralization. If you centralize it a bit more, for instance, by saying thousands or tens of thousands of people around the world can run nodes, but not every human can run a node on a phone, then the transaction capability for any node goes up substantially. Running it on AWS or other cloud providers is another way this increases. Looking forward, I think the TPS on Base just hit almost 1000 TPS last week. My expectation is that this will keep growing substantially. Everyone around the world will be doing transactions on the blockchain, not just financial ones. Social networks and other applications will live on this, just like the internet. That's the vision we see going forward, and that's exactly how we've progressed so far. In 15 years, we've achieved this much. Now it will be exciting to see what we can do in the next 15 years.

Nathan Labenz: It's a mix of efficiency advances at the raw algorithmic level, and also some design trade-offs. Some combination of those is delivering much higher scalability with compromises to the original vision of Bitcoin specifically.

Nemil Dalal: Bitcoin was one network. I'm saying this holistically for many different networks. Different networks have chosen different paths along that journey, such as making a design trade-off or focusing on another factor. Absolutely, there's a mixture of different approaches that the entire blockchain industry is taking.

Nathan Labenz: Let's get to X4O2. This is the new protocol. First of all, I love the fact that there has been this HTTP code. I didn't know this, but everyone knows 404, and probably 401. Not too many people, including myself, have heard of 402. So, introduce the X4O2 protocol, what it's meant to enable, and a little bit about how it works.

Nemil Dalal: Let me start at the beginning. We launched something called AgentKit. For many listeners, if you have an AI agent, you want it to be able to do things on-chain. What it needs is a wallet. We built that last year. Thousands of developers started using and embracing it. Teams like Virtuals launched AI personalities that you could tip, that could send money around the world, and they're all using AgentKit. When people used it, they asked, "Wouldn't it be amazing if, with this wallet, this AI agent could actually pay someone or something?" For example, it could buy a good, or go to an API, access the data the API provides, or send a text message. These are all things AI agents can't do because they don't have access to money. In a world where AI is essentially becoming a proxy for humans, that sounds crazy. It was very clear to me and our entire team that AI agents needed to be able to access money and do things with it. That's where X4O2 comes in. In the early days of the internet, the vision was to add credit cards to it. At the time, it was very difficult. You needed Visa and MasterCard, and others, to cooperate back in the '80s and '90s, and as a result, this never happened. The internet had a standard where if I wanted to access a webpage, I didn't need to look at an ad to do that. That's the model we're all used to today. But what it could do is say, "Hey, Emil, you're looking at this website. Before we can send you to the page, you actually have to pay 10 cents to do it." I would pay the 10 cents somehow. I'd get this 402 status code and say, "Oh, I didn't realize I had to pay. Okay, 10 cents, awesome." I'd send the money, and suddenly I'd have access to the webpage. So, the internet had this concept but couldn't implement it. We decided to look at that, and today we said, "The use case is actually a little different." We had gotten around the fact that 402 didn't exist by just using credit cards. But credit cards are not ideal for AI agents. You need things like permissioning, you need to sign off if they want to do it, and AI agents will want to access many different things. It's not just a subscription where I sign up and pay my money and then use it all the time. An AI agent may go here for one query, and here for another, and it needs to pull all that together. So, we said, "This is the perfect moment to bring this back," and the blockchain is a perfect mechanism because it allows you to move stablecoins around the world, move Bitcoin, move NFTs, and all these other things. That's essentially where we put it together. X402 is a standard that makes it super easy for any website to accept money and then return a result based on it. Here are two simple examples: I have a storefront, and an AI wants to buy from it. I can add X402, and suddenly they send me crypto tokens, let's say 20 USDC, and I send that to an address the AI agent has told me. That's one example. Another example is, I have an API and I want to provide that API to AI agents. One example is financial data; I want to let that AI agent trade the stock market or trade on the blockchain. That's an example where I can sell that data for a fee. Another example could be like Twilio, where I want to send a text message or use some other form of API. I'll pause there, but ultimately, X402 is an open standard. Anyone can build on top of it. We just incepted it into the world, but it makes it super easy for anyone who has an API to accept money in the form of crypto tokens and then offer a response back as a result. We think this is valuable for humans and micropayments, but in terms of where this moment is today, I think it's especially powerful in a world of MCPs and AI agents that want to go online and do things. Because, often, AI agents are not watching ads. One of the big changes in the web that's going to happen is that the entire monetization model, from my point of view, is going to change going forward.

Nathan Labenz: I want to get deeper into how it works. Taking one step back, it sounds pretty nice from a human user standpoint to envision an internet where, for display sidebar ads or whatever, those rates are not great, right? We all know that it's not super easy to just plug in Google display ads or whatever and have cash immediately rolling in. I think a lot of people find it intuitively appealing to have an option for publishers to say, "You can buy our content on a micropayments basis." Obviously, everybody has subscriptions everywhere, but that's getting tiresome. People are overloaded with subscriptions, and it feels like it has run its course in some ways. Now I go to CNN, and all of a sudden they want me to pay $30 a year. I don't think I need another subscription to CNN, but I might pay one to 10 cents for an article of interest. Why hasn't this already happened? We've had blockchain capability for at least a while now, right? What has blocked that from materializing for human users?

Nemil Dalal: Yeah.

Nathan Labenz: ...leaving aside that we now have a new class of AI agent users?

Nemil Dalal: First, for micropayments, traditional methods are really challenging, if not impossible, right? It's very hard to charge one cent with a credit card.

Nathan Labenz: Yeah, you can't do it with a credit card.

Nathan Labenz: But with the crypto technology we have, I still have never had the experience where I've gone to a website and they've said, "You can pay one Satoshi or whatever for access to this article."

Nemil Dalal: Yeah. By the way, there are some sites starting to offer this. There's Zora, they're more crypto native today, but Zora, Farcaster is another one. These are examples where you can do things for just a few cents on these apps. But you're absolutely right. I would say the biggest shifts have been: one, early crypto didn't have stablecoins. That was one issue. Two, fees were super high. I'd have to pay on Bitcoin five, 10, 20, $30 to make transfers. Only in the last six to 12 months has that price really nosedived because of things like layer twos. Solana and others have started cutting down on the fees charged. Previously, with $20, $10, or even $2 fees, it really impacted the types of micropayments you could do. The weird thing was that in 2015, 2016, it was possible to do really small transactions, but then the blockchain became super popular, and fees started spiking. That's essentially the battle we're fighting right now: to make it easier and lower fees. My instinct is that now it's going to become increasingly possible, but it required us to come up with these layer two technologies, these cheaper blockchains, which just wasn't possible even two or three years ago.

Nathan Labenz: That's interesting. In a way, that may very much parallel the AI story where often people ask, "If this AI is so good, why isn't it in use everywhere?" A lot of times my answer is, "Well, it wasn't useful at all until two years ago." Even then, it wasn't multimodal yet, and the context window was 8,000 tokens and whatever. So, depending on exactly where you want to put the thresholds for real usability, they have only fairly recently been crossed. It sounds like there's a similar story there.

Nemil Dalal: The thing I really enjoy about seeing the AI and crypto journeys together has been that they're both early technologies now starting to get mass market, and we're finally seeing their mass market capabilities. I think crypto's been around a little longer in its form. But I would say that only in the last few years have we really figured out ways to make this simple and easy. That, by the way, is one other big challenge in crypto: usability. The usability of it, with addresses and things like that, was historically very difficult. Only now are we figuring that out. I see the same thing in AI; six months from now it looks unrecognizable in many ways from what it was six months ago, because there's just so much innovation going on. So, I think this is where the moment is finally starting to be now, at least on the crypto side.

Nathan Labenz: For AI, the whole concept of prompt engineering. I'm old enough to remember when prompt engineering was a thing, and there was a time when it was a high art. We've had some of the early pioneers of the most creative prompt engineering strategies as guests on the show. Now, especially just the last 10 days with Claude 4.0, this thing genuinely feels like talking to a human in many ways. When I'm asking it to develop software for me, I don't have to lead it by the hand. I don't have to think super hard about exactly what I want back from it. I give it notes like I would give to a human developer, and it is increasingly capable of taking those notes and running with them. It really is an amazing thing. So, let's go deeper into how this actually works, because we have a lot of people, especially, who are going to be building agents. They might be building agents totally bespoke, with a framework or coding their own, or having Claude do the coding. They might be building something that's a bit more of a structured workflow/agent that might live in a no-code environment. Who knows what they might be doing? There are all kinds of form factors coming online. So, take me through the cycle of the protocol, what happens in one successful loop. What does the payment receiving side have to do? What does the agent builder have to do to equip their agent to do this? What is required to actually make use of all this?

Nemil Dalal: Let's go through an end-to-end example of how this works. One example that's important to me is inference. If you're in AI, you need inference to work. One of the fun use cases we discuss is an agent having a wallet and paying for its own inference. Let's start at the beginning. You have the buyer, an AI agent, that wants more inference. It goes to an API to access inference. It says, "I would like to access inference. Send it my way." That's the request it composes and sends. It then receives a 402. The 402 has a few sets of details. "The inference costs this much. Put it together on this network," for example, the Base blockchain, Solana blockchain, or any other. "And send me a transaction as part of that." The purchaser, in this case the AI agent, says, "Okay, awesome. I know how much I need to pay for it. I take my crypto wallet." You can use AgentKit to hook up an AI with a wallet. I have that wallet and I say, "Okay, awesome. It said I need five USDC on Base to do this." So what I do is I sign a transaction. I don't need to broadcast it. All I need to do is sign that transaction and send that payload back. So I respond back to it and say, "Awesome. Same request. Here's the five or 10 USDC," whatever that amount is, "in a signed transaction." Now, on the inference provider side, this is where it gets really interesting. It says, "Okay, awesome. It sent me this payload. It looks like a well-formed transaction. Second, it looks like it's good for the money; it actually has the money, which is great to see. I'm going to wait to submit it because I don't know if the request will use all $10 of the inference they sent me. Maybe it'll use more, maybe less. Let me do the work." So it does the work, and then it has a decision to make. It says, "Awesome. This looks great." I submit the transaction, and suddenly I have the $10 myself. It arrived on the blockchain. Then I can send back the funds I have. But as you can imagine, there are a ton of different edge cases, and I'm happy to talk through any of these if they're interesting. One is that if I overpay, I can always overpay, get a token that allows me to keep that as a prepaid card, letting me spend that money until it's exhausted. Or, if I've underpaid and actually need more money to do that. I can respond without the response the buyer is requesting. I can say, "Actually, you can send me more money. I thought it was $10, but it's actually $15. Go ahead and make a new transaction and send me that." Again, there are a bunch of different edge cases where maybe I spend that money on inference, and it should actually pay me more, so I'm out of that money and similar situations. So again, there are a lot of smart things we can build on top. But I'll pause there; that's the core of it. It's super simple: "Hey, request X402. Awesome." I'll put together a blockchain transaction, sign it, but don't need to submit it. The other side receives it, validates it, verifies it, does the work, submits the blockchain transaction, and then returns whatever inference capabilities I need. That's it. It's basically a request, then a response, and then you're done.

Nathan Labenz: And part of what you've released is middleware software for the payment receiving side, right? That allows them to... I saw a one-line drop-in of,

Nemil Dalal: Yes.

Nathan Labenz: ...this one line of code will check for the payload, validate, and so on. If it fails, it kicks back this 402. Is it really that simple? How simple is it to start accepting money this way?

Nemil Dalal: Totally. First of all, x402 is just a standard, like any other internet standard. It's out there; anyone can use it, modify it, adjust it. We're excited to see what the world does with it, and we're also seeing tons of companies reaching out to us. That's number one. But what we realized is it's not as simple as just throwing out a standard. You have to make it super easy for people to use, and you actually have to come up with use cases, like the inference example we were talking about, to make this easy. So what we did was build a middleware. We started with Node.js, which is one of the most popular, JavaScript is one of the most popular languages in the world. We said, "Let's make a super simple middleware where I can put it into my web server, in front of any API call I already have." It's basically one line of JavaScript that I include, and suddenly it works. But I don't think that's even simple enough, because you still need to manage the blockchain infrastructure and similar things. That's where Coinbase Developer Platform comes in; that's the team I lead. We basically made it so we can spin up a wallet for you. We can automatically convert the funds you receive into dollars in your bank account, if that's where you want them to end up, rather than having them on the blockchain. In general, with standards, I think the key takeaway for us is we have to make it super easy. We have to support people to do that. That's essentially what we do. We have a Telegram group where anyone in the AI community who wants to join, we'd be excited to get them spun up. We have a Telegram group they can check out. For anyone interested in more details, they're all available at x402.org. You can see the white paper, the diagram I went through of the request response, and also the exact details of how the middleware works and download it.

Nathan Labenz: And then if I'm on the agent builder side, does this become another tool, or even another MCP-type construct that I give to the agent? So I would say,

Nemil Dalal: Yeah.

Nathan Labenz: And then I also have my payment tool, which, having received the 402, my AI would know to call this payment MCP, send the money, and then loop back.

Nemil Dalal: Totally. By the way, we talk about MCP. I imagine all your listeners know it well, but MCP is one example of how we're rewriting the web to make it easy for AI. The web was made for humans. MCP is one example of that, and we want to make it easy to discover the different APIs and what you can do with them. That's what MCP was. But even MCP is missing the payments layer, right? That's essentially what x402 adds on top of MCP. So we see a future world where every MCP server is hooked up with x402, and when there's a paid API, they can do that. Some websites, like the New York Times, maybe someday will charge you when you're crawling the New York Times. If you're an AI crawler and you want to do that, it might charge you, and that's the way it enforces and does it. You can't access the web page if you're an AI agent, unless you pay, as an example. And you absolutely brought up this idea of discovery, right? Which is, how do I discover all the x402 APIs that are out there and use them? Today, the way we started was we just have a page on x402.org. Any of your users can check it out. I think it's one of the first links on the page, and you can see the ecosystem. We're starting initially with the crypto community. We think there are two key areas that we're focusing on. So in the crypto community, if you want a crypto API, there are a number of different teams like Nayyar and others that have integrated it. And then there are inference providers that we're also exploring and talking with to add their capabilities. You can see discovery of all the different things that are out there. But over time, where this gets even more exciting is we'll probably have an MCP server, which will list all the x402 endpoints you can access, with descriptions and why they matter. Then the AI agent can decide,

Nathan Labenz: One thing that seems it could still be a challenge, but maybe also a huge unlock, is that inference providers, certainly your Anthropics and OpenAIs and presumably Google, although I haven't hit this myself yet, are starting to do more of a KYC type of paradigm. So I wonder, is that a barrier or is that something this can help with in a way? Because you've got lots of people who may struggle to meet the KYC requirements, and maybe that's... I just don't really know what to expect as the intersection, because the notion is generally that the models are getting really powerful, we're going to want to know who's using them and for what. They're starting to implement all these account level monitoring type of things. Would they be able to do a version of that with these protocols driving the transfer of funds? Or is there an incompatibility or maybe a missing extra component to the system if you want to do that?

Nemil Dalal: I think that's independent in that Claude, OpenAI, anyone can add; that's their decision whether to gate certain functionality or features behind KYC or not. So I would say that's independent of this. But I think where x402 gets interesting is that it's the lowest level of rails, right? So you can imagine things on top of it, for example, reputation. There could be a reputation system that's built on top of this. For example, if I don't have the money, could I use that reputation as essentially a credit score for an AI agent so that they want to pay for some API, they don't have the funds, but I'm floating their credit. And there should be a world with a reputation system on top of that, that allows you to do that. That's something which doesn't exist today. And x402, again, it's just the payment rails. But as you... Just like with credit cards, just like with other things, adding reputation in addition to a credit card network is a really powerful thing, and the same here is that we absolutely see a world where, over time, people will start adding those primitives. By the way, the blockchain in general, there have been a lot of different attempts at building really powerful primitives around reputation. I think it's just a matter of time, where people have reputations but also AI agents have reputations to do things. So, they might get credit at an API provider. They might get access to something just because of their reputation. They've done good things in the past. They haven't done things in the wrong way. They maybe haven't scraped a website in the wrong way, and as a result, they have higher reputation and higher capabilities.

Nathan Labenz: Yeah, I'm really interested in this reputation for AI agents notion, and even just, like-

Nemil Dalal: Yeah.

Nathan Labenz: ... the provenance, you know, that like, whose agent is this? Who, uh... On whose behalf is this act- uh, excuse me, acting? And it strikes me as a, a kind of challenging problem because try as I might, I haven't figured out a way to draw a real clean box around an AI agent. Sometimes, you know, if, if you construct them very simply, then it's, like, pretty clear what the agent is. But an example that I kinda keep chewing on is, uh, another past guest's company, Augment, which is doing, you know, AI coding assistance but focused on large enterprise code bases, put out a coding agent that used, in turn, a smart MCP, and the smart MCP was provided by, you know, another open source developer, and they used that MCP to do the thinking. So, it kind of maps onto the schematic that you had outlined before where, like, you have an AI but it might wanna go get more inference. That's essentially, you know, the coding agent is going out and saying, "I need help planning all the code changes I'm gonna make," so it has a, a planning tool presented to it in the form of an MCP. There's gonna be some inference there, presumably. Th- this thing happens to be open source, but it could, you know, obviously be closed source or proprietary in any, any number of ways. So, it might have to pay for that in order to get the thing on the other end to do the planning for it and then get its plan back. But now it starts to become, like, a very weird thing where it's like, what exactly is the agent, right? Like, is my thing the agent? Is my thing, like, plus that thing the agent? Um, do you have any extra clarity on this? 'Cause obviously, you know, funds and, and who, like, controls what money and who's responsible for what is important and possibly could be clarifying for this sort of thing. Do you have any sense for how we can get crisp on this? 'Cause for me right now, I f- I- I'm expecting to see just, like, a total smear and the sort of forms that I see people outlining to me feel like more sort of ends of a spectrum, where the sort of middle of, like, my agent but also a smart MCP kind of blurs. Like, where is the intelligence coming from and who's ultimately responsible for what, and if the thing gives me a bad plan or, you know, acts maliciously toward my agent, you know, like, who's really at, at fault for that?

Nemil Dalal: Yeah.

Nathan Labenz: Um, I just really... We have s- like, so many unanswered questions, but I wonder if your study of all this from the financial side of it has given you any rules of thumb you'd like to, you know, see become best practices or, or w- you know, any sort of clarity really.

Nemil Dalal: So, like, I... Like you, I'm seeing this and then marveling at it, right? The behavior that's coming together. I think, like, one model I just historically have seen, and I'll give you the, the reason why, is that, like, just generalist and specialist AIs, so like, AIs that are, like, custom-made for a certain purpose. And so, like, one example we've seen is, like, web crawling, and then there are other A- AIs that are generalist that wanna tap into those specialist AIs. And so this is where I come in, is that we've seen, at Coinbase Developer Platform, multiple examples of AIs paying different AIs. And to a lot of people that, like, breaks their brain. Probably in the AI community, it's not as surprising but, like, like most people in the world, it's like, "Oh, I thought the AI was the thing." And to me, there might be multiple layers behind the scenes, exactly like you're saying, and I might interact with one, and really I'm just talking about one person, like one AI. But the... Behind the scenes it's hitting MCP, which is really like forms of data, maybe. Um, it might hit other AI agents who are super specialized. And so, where X.OAR2 and Coinbase Developer Platform come in is, we think that, like, you need a financial layer to incentivize that, right? that like, if one AI is asking another AI, then like it probably is not gonna... You know, like, it's not gonna get charged by an ad. There's some form of value transfer that needs to happen there. And again, if it's the same company's AI, it's super simple, but if it's multiple companies and multiple different agents that are out there, and more specialized and more generalized, um, in different context windows and they all want to inter- interact together, we think the thing that's missing is really just money, right? Some form of financial exchange that can go on. So, that's like one part of the solution, I feel, is that you have some form of money movement around that allows you to be able to talk to generalists and specialist AI agents, get the information from them, maybe get... Like, some MCP servers are free, but other parts of MCP server, just like, you know, APIs are gonna charge you money and you pay for it. So, that's like one whole set of like conjectures. The only other thing I'd just tell you is that, like, the blockchain is a perfect identity layer, and I think this is like for, like, for anyone in AI building, I think this is an area where, like, I always love bringing the AI and crypto communities together. But an example is that, like, the blockchain, like, if you have an NFT, that essentially is a form of identity, right? It's like, think of it as like creating a, a user. Like, on the blockchain I can create an N- an NFT, and by default that user is empty. There's nothing in it. But then if I start doing things. Let's say I'm that AI agent and I, I, like, correctly deli- deliver medical advice, as an example. A...... user who's seen that information can, like, tag my NFT and say, "Hey, this is great. It's a really good medical information," and they gave me the right thing, and I confirmed that it was the right thing. And other people can do the same thing. So think about, like, a passport, where you're getting stamps, right? So, like, you have... Like, the AI agent has an NFT. It's getting stamps more and more, and maybe all the stamps are on medical information. And suddenly, over time, it's seen as a credible source for that reputation. So again, this is not to do with X4002, but in general, what we're seeing in crypto, there's a lot of innovation on the identity stack. And the... Again, like the blockchain, it's like, think about it as, like, an open source credit bureau, right, where that information is available. Anyone can add information to it, and all you're looking at is like, who are the people adding information? Do I trust them? And are they adding a lot of positive signals? If they are, I'll maybe give more weight to it. And hey, they did this one thing bad. It's kind of like a review system, you know? If they've done this one thing bad, or they made a mistaken advice this one time, but 999 times out of the thousand, they did it right. So to me, like, the two things I see is, like, blockchain is amazing for the payments layer. I see it also as amazing for that reputation system. And this is all possible today. Like, none of this is, like, sci-fi. I think the sci-fi part is like, how do we bring them together, and which use cases do we start at? So again, if there are entrepreneurs listening today, I think this is a really interesting area of, like, innovation and exploration.

Nathan Labenz: When I set up a wallet, presumably I am not going to give my AI agent my main wallet with the bulk of my funds, right? We have much harder, if not impossible, to reverse transactions, and all sorts of hard to predict and easily hijackable behavior on the part of the AIs. My first instinct would be to limit my downside risk by keeping the deposit into the AI's wallet small. As long as these are microtransactions, paying for information it needs to accomplish tasks, and it is all very small, then whatever. If I drop $10 into the AI's wallet and somebody prompt injects it into sending that full $10 in the wrong direction-

Nemil Dalal: Yeah.

Nathan Labenz: I can just live with that. What more do we have in terms of keeping... This also intersects somewhat with another idea that has been floating around, which I mostly associate with Tyler Cowen. He has been saying that he expects we are going to need to have AI agents somehow capitalized, that there is going to be-

Nemil Dalal: Yeah.

Nathan Labenz: a financial stake so that there can be some redress if the agents go wrong. They are not just flying around willy-nilly; they have to put some skin in the game for the agent-

Nemil Dalal: Yeah.

Nathan Labenz: or, obviously, the parties that create the agent. Presumably, that would need to be a lot bigger than $10. I feel comfortable with dropping $10 into this agent, or even $100. But I am not going to put major funds into it today because I do not really trust its judgment that much, and I also know that they are not adversarially robust. Then we have this other idea that maybe they really do need to be capitalized to create the right incentives for people to have the right guardrails and recourse if they mess things up. So, two questions are: what are the best practices? How do I keep myself safe when I do this? And does this extend to that capitalized agent concept, or would you put that in a whole different bucket?

Nemil Dalal: Yes, I think it does extend to the capitalized agent. Let me take a step back. You mentioned you might want to put $10 or $100. I will be upfront: I think people are going to put millions, if not billions of dollars, into AI agents over time. To give you a simple example, I want to be able to trade on-chain, on the blockchain, or on the stock market. That is a lot of access to funds that you are giving an AI agent. We are seeing that people want to be able to say, "I have an AI agent, it is a trader agent." We had hundreds of people at our hackathon exploring ways to build a wealth manager. I can give it money, and it can go make more money from this. It sounds like science fiction, but the flip side is there are tons of quant funds on Wall Street. You and I probably both know that they do this already. The question is, how do we get to that world eventually? To your question, absolutely. One way is segregation. That is how you defined it: can I put a small amount of money aside and only have it access that? Beyond that, what you need is some form of permissioning. If it is more than a certain dollar amount, please escalate it, and I have to sign off for the ability to do that. I will let you do a certain degree of transactions, and the more you do that are right, the more credit line I will give that AI agent. The two things we have been seeing are segregation and some form of credit limiting process. It is not just credit limiting, but it is basically a sign-off, determining if the agent is doing the right thing or not. If the AI agent wants to do a trade, I can look at that trade and say, "Does this make sense? Does it not? What is the rationale for it? Okay, sounds good. Go ahead and do it," and then give that permission. That is part one. To your question on the Tyler Cowen thing, crypto already has a form of this called slashing. Many of the newer blockchains are proof of stake, where I have different transactions. I put some money at risk, a million dollars, $10 million, $10, and I earn some yield from putting that money at risk. But if I do the wrong thing, I can get slashed by the rest of the community, and I lose that money. This is exactly the same model you can have for this: the AI agent has that crypto. The cool thing is these are all primitives built on the blockchain. So the AI agent has that money, it puts it into escrow. By putting it into escrow, it can do certain transactions it might not be able to do, but then if it does the wrong thing, those funds are slashed and taken away. Absolutely, I think the blockchain is a great element for this. All this requires movement of money, which is what X4002 is, and then you can build all these intelligent primitives on top, like reputation, slashing, etc.

Nathan Labenz: Is that slashing ultimately a social process? Is it a matter of consensus in the rest of the community that this entity, whether human or AI or otherwise, did wrong, and so we are taking your funds? It sounds almost like a jury, a trial-type function.

Nemil Dalal: Yes. On blockchain, that is the case. But it doesn't have to be the same for the mechanisms you use for AI. That's one inspiration. It could be a centralized party, a single centralized party, as a way we could architect this for AI. We could do it as multiple different parties, and they could all be trusted, right? It doesn't have to be just a random juror of folks out there. On the blockchain, the way it works is it is the community, and again, different networks have it different ways. There might be different leaders that are elected through some decentralized governance vote, and as a result, they are the ones who get to decide whether something is right or wrong. If they decide wrong and it's ultimately shown to be erroneous, then they all get slashed, which is how it's done in decentralized systems. We could design it any way we wanted for AI. A lot of times what we see with these systems in crypto is they are always decentralized, and then there are some trade-offs regarding how decentralized they are. We've also seen, as we talked about stablecoins earlier, that's a centralized form, and it could be a centralized form of jury. There might be a company or a person who attests to whether this is the right action or not. People trust them so much because they've done such a good job and haven't gone against that social goodwill for a very long period of time, so they're allowed to keep doing that. I wouldn't say that we have to follow what is in the blockchain. Mm-hmm.

Nathan Labenz: The cool thing is, all the primitives to allow that slashing are possible. Blockchains today have a form of smart contracts. Think about this as a legal contract, but it's a legal contract that the computer follows to the letter. However it's written in code, that's what it follows. So if that's the slashing mechanism, whatever that is, you can guarantee it will be followed. In a world where every country has a different legal system, where we may or may not agree on the laws and things like that, the power of these smart contracts is that they are enforced the same way the world over. So it could be a really powerful primitive for this.

Nemil Dalal: Yes. Let's come back to smart contracts in just a second.

Nathan Labenz: Yes.

Nemil Dalal: In terms of escalation, you had mentioned creating rules to govern my AI agent's payments. If it's above a certain amount, then escalate that to me. That obviously makes a lot of sense. I assume that would be something done on the wallet side in a hardwired way. I don't just want to be prompting my AI agent, "Hey, by the way, in general, if it's more than whatever, you should come to me." I think, at least for now, I want to have a much firmer boundary for it. So presumably I can set that up in a Coinbase-powered wallet, no problem, right? Do I have that understood correctly, for starters? Yes. The power, again, this is a smart contract. It's just a form of what the rules are to let me do something. The way we would design this in code would be that the AI agent would be able to spend any dollar, let's say up to $10, without getting permission. If it's more than $10, if they try to do it, it fails by default. But if they want to do more than $10, the way it works is they bring their human, and the human has to approve a certain thing using their private key. So the AI agent might have its own wallet, but a human has their own wallet, and both of them have to say yes, and only then are the funds spent. The reason you want to do it not only at the AI level is that in a perfect world, the AI agent is prompting you. So it knows a stock, and the AI agent is saying, "Hey, Emile, I would like to spend this, but I need your approval." So you probably want to tell the AI agent, "Hey, if it's for more than $10, let me know." But you don't want to trust that on its own. There could be other reasons why the AI agent may use that. So you have it at the wallet level as well. The wallet is basically what we call a two-of-two multi-sig. Think of it as a safe deposit box. I have a key, the bank has a key. In this case, the AI agent has a key, I have a key, and the two together are required to be able to open it. That's the way we'd architect this.

Nathan Labenz: Yes. Okay. That makes sense. How about more on a systemic level? One of the most fascinating bits of research we've covered on this show in recent months is a paper that looked at different models and their ability to learn to coordinate with each other as agents over multiple generations, or their failure to learn to coordinate with each other. The result was, and I don't think the result actually matters too much, but just to show that nobody really knows what's going to happen in some of these dynamic systems that we're starting to develop. Claude, a prior version, was able to play this donor game effectively with itself in a way that multiplied its funds over the generations of the agent and settled into this cooperative equilibrium. The core mechanic there is that one Claude could donate to another Claude, and if it did, the receiving Claude would get twice as much. So the Claudes were able to do that with each other and not only take the pro-social action, but also punish defectors and get into this equilibrium where the majority of Claudes are doing the donation and their collective wealth is rising. GPT and Gemini could not do that, interestingly. My outlook is that all possibilities are probably going to become real. That sounds amazing, could be amazing. It could also be that the flip side of cooperation is collusion. So I could imagine AI agents cooperating with each other in extremely pro-social ways that I would approve of. I could also imagine them colluding with each other in all sorts of ways that might be really problematic. What do we have, if anything, at a systemic layer? I know the original vision of Bitcoin was that nobody can shut it down. But I'm a little nervous about creating a vast sea of autonomous AI agents on a foundational technology that nobody can shut down. I think about episodes like the famous Flash crash, which I don't think is necessarily a great analogy for what AI agents might get up to. But the key point is, when the Flash crash happens, there is somebody with a breaker switch that can say, "Hey, this whole system is going totally haywire. We're going to shut it down, look at what's going on, and we'll come back online when we have clarity for people, so this doesn't run away too far from everybody and make a total mess." Do we have anything like that in the crypto-enabled AI agent economy? If not, is anybody working on that? Because that seems like it's probably going to be something we will need or want in some way, shape, or form.

Nemil Dalal: First of all, crypto and AI are at their most nascent stage. Before we launched AgentKit, I don't think there were many AI agents with a wallet. Perhaps only a handful of AI agents in the world had a wallet. So we are truly in the early stages of this. Many things we discussed are essential. You need some form of reputation, which makes it much harder for an AI agent to do the wrong thing, as there would be social or financial consequences. That's one part. Second, having humans in the loop is another powerful primitive. The Flash crash you mentioned would often come into play with the networks themselves. For example, in this case, it might be Uniswap, a decentralized exchange, or Aerodrome, another decentralized exchange. These are all examples where they might have a mechanism, such as a circuit breaker, if there are too many transactions on a book in one block. You mentioned the Flash crash, and that's essentially a way to build a circuit breaker into the system. My sense is that as these things come together, we will develop many forms of enforcement mechanisms. The easiest one is probably just a human in the loop. However, at some point, AI agents may become so good that humans might not want to be in the loop. It could be like those pop-up notifications you get too many of, and humans might abdicate some responsibility, especially for lower dollar amounts. So, in that world, reputation and other forms built into the protocols themselves will be critical. We dream about all of this, but until the volume of AI agents performing actions on the blockchain increases, I don't see it happening. The two things I do see are: first, payments seem obvious. Agent commerce is clear. People are already asking their AI agents to buy things, so x42 is powerful for that. Second, as we discussed, reputation. The blockchain is a perfect place to build reputation. It's open access, anyone can do it, and once you write it, you can't change it. It's not like a centralized repository. For all these reasons, it's a very good reputation layer, and I think those are probably the two starting points for this.

Nathan Labenz: Got it. I think I'm going to want some of those circuit breakers at some point. I agree, we're not quite there yet, but it's funny, I think the models-

Nemil Dalal: Yes.

Nathan Labenz: are getting there. Adoption and the dynamics are not there, but I don't think we need anything much more powerful than Claude 4s, Gemini 2.5s, and 03s to create a very dynamic agent economy. It's maybe half a generation away, but it's not very far. Honestly, over the last ten days, I've been feeling that Claude 4 is smarter than me, I have to admit it, and maybe 03 too, I don't know. But it's really starting to feel like these things have it where it counts. They don't have all the affordances, and we haven't created what you have created, but they haven't been adopted. When they are, it seems like it could go very fast. One thing I've been looking out for a long time, and I haven't really seen, maybe you have, or maybe you know why it's not happening yet, or could point me to something. I've had this notion that AI might put the 'smart' in smart contracts. What I mean is, what you can put into fully explicit code is limited, which creates a brittle nature for smart contracts. It feels mechanistic and deterministic, not like going in front of a human judge, and ultimately, it can't really resolve disputes between people. On the other hand, AIs are getting to a point where you might say, 'Our smart contract resolution mechanism is that we will call Claude or Gemini 2.5, submit the evidence to it, and let it decide what is right or wrong.' I think about local services. I have a hundred-year-old house, and when something needs fixing, it's not easy to find someone reliable. The reliability factor there is not great. But it seems like we're getting to where something like that could be possible, and I just haven't seen anybody come forward and say, 'Here's my arbitration on the blockchain powered by AI system' yet. Are you aware of anything I'm missing, and what do you think is the prospect for that kind of future?

Nemil Dalal: I haven't seen anything, and the exciting thing would be if someone listening to this podcast went and did it, because it's all possible today. What you're describing is a form of multi-signature, where one of the signers is the AI agent, and it can make the decision. For example, it could be a two-of-three system. Let's say there are two humans in the loop who don't agree, so only one agrees that the funds, for example, should be released. An AI agent is also there to weigh in and make a decision, acting as the second signer. Then the decision is made. I'll give you another example that I really liked earlier in my career. I worked for a few months with the Gates Foundation on something called weather-indexed crop insurance. The way this works is that if the rainfall is less than a certain amount, you automatically pay out. The question is, how do you figure that out? Typically, it involves some form of data source that needs to align and make a decision on whether funds should be released. The cool thing is you could pay for a growing season, say $1,000. The money is there, investors put it in, and the money goes to the investors if the rainfall is more than a certain amount. If it's less than a certain amount, it goes to the farmer. This is a great example where AI can be plugged in. All the AI has to do is, on a specific date, weigh in and review the precipitation over the last six months, then unlock the funds. All of this is possible. None of this is required, but it's feasible. One other thing for those on the AI side: there's an interesting challenge in crypto where oracles are the hard problem. Smart contracts can only access what's on the blockchain. Many things in the world, like weather information, are not on the blockchain. So, they use something called oracles to bring that data in. Your idea is essentially to use AI as that oracle. We should trust the AI, hopefully it doesn't hallucinate. If we believe in that AI and think it's high quality, then that's a great example where the AI becomes the oracle. It can decide how to vote using its key whenever it needs to be pulled in. So, I haven't seen it, and it would be amazing. This is relatively simple to do. If anyone wants to reach out to me on Twitter or elsewhere, I'd be excited to give them details and help them get set up if they wanted to do this on the AI side. Basically, you need an AI you trust and a problem where you want it to weigh in, and then the AI can absolutely do that.

Nathan Labenz: It strikes me that AIs are very gullible. I'm a close follower of the AI bad behavior research. Much of that research is done by telling the AI that nobody will read your private scratch pad, so you can think about what you want to do there and then you can take your action here. Then of course the researchers are reading the private scratch pads, and that's where they're seeing some of the evidence of the bad behavior. But more generally, they're easy to trick. So a hard thing from the perspective of the AI would be knowing what constitutes authentic evidence, what is made up. Is that really a picture of your roof? Was that picture taken now? Was it taken a month ago? You're saying it's not fixed, but I can't tell if that picture was taken today or a month ago. So it seems we may also need some sort of provenance technology, like on-device signing of photos. Or in your weather example, it would be some sort of indication that a sensor in the field is actually in the field where it's reported to be, so we can trust these raw numbers. And then the AI can provide the analysis on top of those raw inputs. But we still need to have some confidence. The AI needs confidence that it's actually dealing with real evidence, right? That it can reliably reason over. I think we have a little bit of that, but not much. How would you characterize the state of how much we can demonstrate to AI that the evidence we're submitting to it is in fact authentic?

Nemil Dalal: Yes. There are two parts. One is, is the evidence we're submitting authentic? And the second part is, is the AI hallucinating or is it being goaded or anything else in a certain direction? By the way, I've seen AI agents that have Twitters. We've seen examples in the community where AI agents have mistakenly been hacked or goaded into giving out money in a way that it shouldn't be. So I would say both are critical, both are important to get right. But with any form of technology, I think you start where it's possible. So, the satellite example of the rainfall. The idea that, on the course of this growing season, is there enough rainfall? The cool thing is that could just be satellite data. You don't need a human. You're right. On the roof example, is my roof bad or not? Absolutely. You have to figure out provenance. You might want a third party to take the picture rather than the person who's insured as part of that. But the weather indexing example is that it literally could be satellite data that Google has or anyone else has. Or there's a satellite company with an API, use x402get. These are all examples of how you could do it. So my instinct is, those are things that, to get this perfect and do it at scale, absolutely. You need to solve both of those problems over time. But the hallucination could be solved a little bit, especially, sorry, the goading problem, which is even a step further than that. Could be solved by not putting the AI online, right? Only having an expert be able to access it and parametrize it or things like that. And it doesn't accept other inputs after that. But then, when you have that satellite information or things like that, those are ones where they're very precise and they're not super complex problems. And that's I think where a lot of this stuff starts. It doesn't start with the hardest problem on day one. It starts with a very specific problem. Google has the data and there's a human with some funds in a wallet. Let's combine those two together, and that works.

Nathan Labenz: I really appreciate your time, and I know we've run a little long. Perhaps a final opportunity to take the floor if there's anything we haven't covered at the intersection of AI and crypto that you'd like to highlight, or any elaborations on the positive vision you'd want to inspire people with.

Nemil Dalal: All I'll say is we're scratching the surface on how crypto and AI can come together. I think crypto is a really powerful financial system for AI agents. And x402 is just the start of that. And you and I talked a little bit today about things like reputation and other things that haven't even been really tested yet. So I would just say, if you're building an AI today, I'm super excited for you to try out x402. That's x402.org. You can also DM me directly on Twitter if anyone wants to talk more or engage on anything. But there's so much innovation possible because we're finally bringing these two technologies together. And so super excited to see what people build with it.

Nathan Labenz: Nemil Delal, developer platform lead at Coinbase and one of the lead developers of the 402 Payment Required, also known as the x402 protocol. Thank you for being part of the Cognitive Revolution.

Nemil Dalal: Thanks so much.