Why AI Agents Want Bitcoin

All right, so we're here to talk about one of the most exciting emerging use cases for Bitcoin, which is whether you want to give it to your AI agents and potentially give them sovereign spending authority.

​

We're going to go through three rounds with some open exchange at the end. We've got a great panel here. We'll introduce ourselves and then dive in.

​

My name is Eric Hadley. I run a website called Hyperdope. It does Lightning-gated video. I also do some things with agent services. I run a site called L402.directory that lists different health-checked services that agents can actually buy today.

​

Hey, guys. I run OpenAgents. We are an open source AI lab building on the Bitcoin stack, proudly based in the Bitcoin capital of the world, which we all know is Austin, Texas. Thank you very much.

​

One unique thing about OpenAgents is we know that everyone in this room has compute that you're not using, and we'd like to pay you bitcoin for it, because we figured out a way both to collect it from you and to have agents use it for fine-tuning, inference, and stuff.

​

Our goal at this conference is to run the largest decentralized model-training run in the world. As soon as we cross the 200-node threshold, we will steal the world record from Bittensor, so help us out with that. OpenAgents.com, click the button, paste it to your agent. It'll spin up a Pylon. Hopefully, if you've got a Mac or computer with you, we can get you some bitcoin earned by the end of this talk.

​

Hey, my name is Erik. I'm a design engineer. I work on Cashu. Cashu is a bearer asset mechanism on Bitcoin, where if you hold the string of text or the data, you have bitcoin.

​

Lately, we've been seeing a rise of interest in Cashu from some unexpected places, from Bitcoin companies that want to use it to administer payroll, to non-Bitcoin companies that want to use it internally to transfer money, and increasingly, people that want to use it for agentic payments. That's what we're here to talk about today.

​

Hey, everyone. I'm Roland from Alby. We build products, tools, and services to make it easier for individuals and businesses to use Bitcoin, which we think is the global currency, both for the web and for everyday usage.

​

Recently, we've been building a lot of AI tools. We have three different AI skills. One is for payments, to give your agent the ability to connect to a wallet and make and receive payments. One is for building applications that have Bitcoin in them, so just with a single human-language prompt, you can build a fully enabled Bitcoin application. Finally, we have a skill that enables your agent to manage a fully self-custodial node.

​

Apart from that, we've also been looking a lot at agentic payments. Our payments skill for agents has the ability to make agentic payments for the three main protocols, which are L402, x402, and MCP.

​

All right, so we're going to start with round one. Let's ask a general question and hear from everybody. Christopher, we'll start with you. Do agents even want money? Do they find it useful? And among the options, is Bitcoin the right option?

​

I'll start with a pedantic philosophical point, which is that agents don't want anything. This is a piece of software. There's certainly a movement among certain big labs, like Anthropic, to anthropomorphize the AIs and make you build an emotional connection with it. What does the agent want? What does the AI want? The AI doesn't want to be sunset, and let's have model therapy. Psychotic things, in my humble opinion. So the agents don't want anything.

​

At the same time, if you look at the models and the data they're trained on, the Bitcoin Policy Institute had a great paper where they ran some studies about whether, if you put certain prompts in front of certain models, it turns out they express a preference for Bitcoin relative to other things.

​

Largely, that's a function of the data that it's trained on. We've got 15 years of people trolling and saying stuff on Reddit, and we've got more content because we're just older, so it's weighted more heavily. But it's all a function of what data gets into the models.

​

You can envision models that are trained on not that. There was a project a couple years ago called Spirit of Satoshi that trained specifically on Bitcoin stuff, and they did fine-tuning. They started with the base model, then tried to make it more of a Bitcoiner. But if it's all a function of the data it's trained on, this is our argument for why we want to start training our own models from scratch. What if we don't feed it the midwit Reddit-tier shitcoin analysis in the first place? What if we teach it economics from day one?

​

Erik, your thoughts on the topic, and maybe you can say a little bit about what the UX is like today, maybe without Bitcoin.

​

Do agents want Bitcoin? Starting there, agents don't want anything. People, humans, want to use an agent to pay for something. So the agent just wants to complete that instruction and execute the payment.

​

Right now, if you want to make a payment with an agent, the current user experience is tied to human identity in the human world. You either have to feed it a credit card or link it to an API that you've already authenticated your human identity to, and then it's topping up. That's the current experience. Then you can have the agent perform the call or the function or whatever the case may be.

​

What I think is more novel, more interesting, and more unexplored is the permissionless aspect of having an agent perform a payment without an identity. These things are not tied to the human world. A permissionless agent that you spun up five minutes ago doesn't have a human identity. It's just this piece of software.

​

So how can this, in the current landscape that exists now, go on and execute these payments without a human identity, without KYC, without needing permission? That's where Bitcoin solves that problem.

​

There are many different ways to pay with Bitcoin. There's on-chain, there's Lightning, and there's Cashu, which is what I work on. I'm especially interested in the Cashu use case because it reduces a lot of that UX friction.

​

If you want your agent to pay with Lightning, for example, and if it's self-custodial Lightning, you need to have inbound liquidity, outbound liquidity, and an open channel. You need to trust the agent's logic not to spend your entire Lightning wallet balance. If you have $200 in there, you're going to tell the agent to perform this for $20 or whatever the case may be. You're going to trust the LLM not to hallucinate and spend your entire balance.

​

Cashu flips that on its head because with Cashu, it's a custodial system with great privacy. You don't have to deal with the onboarding friction of custodial Lightning, and the spend is capped by how much e-cash you give it as a bearer asset. If it has $20 worth of e-cash, it can only perform $20 worth of payments.

​

Roland, your thoughts on that same question, and why would Bitcoin make more sense than some of the competitor rails like x402?

​

I'd actually like to build a lot on what Erik said. How many people have run an open agent or any sort of agent before? Cool, there are a lot of hands.

​

The first thing I did when I set up an agent was I wanted it to be my personal assistant, and I asked it, okay, set up your own email address. I thought this was a very simple question, right? But then it thought about it for a while, and it came back to me and said, okay, you need to go through these steps to give me an email address.

​

But it's my personal assistant. It should do the work. It shouldn't be giving me the work, right? This is one of the things that Bitcoin as permissionless money solves. Instead, we can use something like LN email, or there's another one called agent mail. Basically, by making a micropayment, you can give your agent the ability to send and receive emails.

​

It's not just limited to emails. Any service that requires KYC or CAPTCHAs is completely agent-unfriendly. I think this emerging agent economy that we've seen this year has shown how closed down the world actually is. The internet is locked down. Users have to provide a phone number, they have to KYC to access these different services.

​

Why I think Bitcoin is different from stablecoins or other options for agentic payments is that Bitcoin is truly permissionless. No one can stop it. Stablecoins in the near term may be more popular, but we see every day that stablecoins are being frozen. Just a few days ago, over $2 million worth of stablecoins were frozen. So if your agent's money gets frozen, then it cannot actually complete the task it needs to complete for the human. Therefore, it becomes basically useless. This doesn't happen in Bitcoin.

​

The second round, let's get a little more practical. How do you actually give an agent Bitcoin? How do you set them up with Lightning? Do you type in your seed words? Erik, what's the first step people could take?

​

This is where I'm going to shamelessly shill Cashu and e-cash. In order to receive e-cash, you have to exist, because that's the only requirement. It's literally a piece of data or a string of text that you can push through any rail. It's rail-agnostic. You can do it through Bluetooth, through email, it doesn't matter. You're just moving a piece of data from here to here. A string of text. That's all it is.

​

Now extrapolate that further to how agents might interact. Agents no longer have to spin up a Lightning node. They don't have to know anything about Bitcoin. All they have to do is communicate from A to B and push a string of text to each other. That works at scale, and it works for sub-cent payments, which is what a lot of API calls and payments might be: fractions of a cent.

​

Cashu has no problem executing that with sats. You could do this with Lightning, but then there are routing fees and it depends on all these factors. So I would say the receiving aspect that Cashu unlocks is very perfect for agents.

​

That's the Cashu piece. Obviously, there's a little bit more of a trust layer there. Let's say if we're going to go to Lightning. Roland, maybe you can jump in. What's Nostr Wallet Connect, or how does that all work?

​

Nostr Wallet Connect is an open protocol. You don't have to understand Nostr at all. It's kind of just a technical term, but what it means is that it gives seamless access for an application or an agent to access a Lightning wallet. That Lightning wallet can be custodial, it could be a Cashu wallet, or it could be a self-custodial wallet.

​

At Alby, we have two options. One is where you can give an agent permissioned, budgeted access to the human's self-custodial wallet. Say you have a million sats on your self-custodial wallet, but you can give your agent a 10,000-sat budget per month or per day, so that you stay in control.

​

The other option is that you can get your agent to spin up its own self-custodial Lightning node. You have to trust your agent that it's not going to spend it all, but on the other hand, this agent has the keys to its own money. It does not have to trust anyone. It's in full control.

​

Christopher, how do you do your setup? Do you do budget controls, or how do you look at it?

​

First, it's definitely a beautiful part about Bitcoin. Because it is permissionless, it's very easy for companies building on the Bitcoin stack, like Alby, Cashu, and the Money Dev Kit guys, to put out these little toolkits that are MCP tools or skills. They're the kinds of things you can literally feed to an agent. There's generally no account to sign up for, or if there is, you can get an API key very easily.

​

Then all the many different trade-offs you would want between privacy, security, and custodianship are pretty much covered. We have a bunch of those different trade-offs covered. All of that can and should be largely seamless to the user, at least the more mainstream user base that I'm trying to target.

​

We're trying to solve this at the product level. You have a piece of software that you install from OpenAgents called Autopilot. It sits on your desktop. It has a built-in Bitcoin wallet that we're shipping using Spark Wallet to start.

​

If you want to interact with your friends, neighbors, business partners, or someone you're buying from, and their agent really prefers Cashu, that should just be a negotiation between agents. You shouldn't have to care about that. We don't need to choose between these different providers. It can be an all-of-the-above strategy.

​

We're going into the third round. This is about the emerging agent economy. What do agents actually want to spend sats on? What are they looking for out there? Roland, why don't you start us off?

​

I think agents still act on behalf of a human, right? A human is trying to build something or achieve some goal. Normally the agent can pay for that with Bitcoin instead of having the user manually spend time. It's basically automation to save time.

​

There are different indexes, like Lightning.directory and also the L402 Index, which have a whole bunch of APIs that agents can use. Rather than signing up, paying for subscriptions, and using credit cards, they can pay per use for these different services.

​

One I really like is PPQ, where you can pay for AI inference. What I gave my agent was the ability to pay for its own VPS, so it lives in the cloud on its own machine. It pays for its own hosting once per month, and this is automatic subscription payments powered by Nostr Wallet Connect.

​

The other thing is it pays for its own AI inference usage. Every single request, it can pay PPQ to be able to operate and sustain itself. It can also do requests to PPQ, pay per request, to generate images or to request data from the Twitter API and other things.

​

I think the Twitter API is actually quite a good example, because normally you need a subscription to be able to crawl the data if you want to get insights into what's happening in different topic areas when you're building or doing marketing. Instead, your agent can just pay for the data it needs. It doesn't have to pay for a subscription. I think this is really cool.

​

Erik, what's your thought on the economy? Any good stories?

​

The two things that I'm seeing the most of, if I'm being objective, are 402 and buying LLM compute. In the 402 space, I will say that the x402 protocol that the Coinbase and Base guys are pushing is genuinely impressive. I don't ever want to badmouth any project. From everything I've seen, the developer experience of that is quite good.

​

But at the end of the day, like Roland said, it's dependent on stablecoins. Stablecoins can be frozen. They're not uncensorable. So for me, I'm interested in agents paying for things that are in a little bit more of an adversarial, permissionless identity environment. That is a lesser-explored design space, and that's something I'm more interested in: 402, but permissionless and uncensorable.

​

The other thing, LLM compute, I think PPQ is a great example. Another service I'm seeing that I also think is really cool is Routstr, which allows you to purchase LLM credits or compute, very similar to OpenRouter, if anyone knows that, but the authentication is a Cashu string or a Cashu token.

​

Instead of an API key that you have to authorize with OpenRouter, you just generate a Cashu token and use that string, that piece of data, as the key to buy the compute. If you generate $50 worth of Cashu or Bitcoin, that is now your API access and your credit. You now have $50 worth of compute that you can use to access all these LLMs with no identity and no permission. To me, that is very cool and very novel. It gets me pretty excited.

​

Christopher, your thoughts on the economy? We've talked mostly about agents spending on things. Can agents earn?

​

For sure. This also ties into my answer to the same question: what's the most exciting thing that an agent can purchase? I think it has to be compute. Compute is the limiting factor to why we don't have a billion or 10 billion Bitcoin agents tomorrow. There's no compute. If they had the compute, then you could run everything else easily on top of that.

​

If we're needing to compete against hyperscalers and all the VC investment and all of the soon-to-be-public-capital money going toward helping OpenAI buy GPUs from NVIDIA, and prices keep going up, that may limit our ability.

​

But if you take a look at the compute mix that OpenAI has, they said that at the end of 2025, they have about two gigawatts of compute in their network. If you total up all of the spare, what we call stranded compute, just on people's desktops, times everyone in the world, that's 20 gigawatts.

​

Can we use that compute for edge inference, to train models, to do embeddings, the kinds of things that are needed in agentic workflows? We've proven that yes, you can.

​

How are you going to do that? You're going to run a piece of software. It's going to run a Nostr NIP-90 service provider. Our software is called Pylon. You install it, turn it on, click the button, and it will say, okay, you've got a subset of your GPU that you're willing to sell to the highest bidder at a certain price.

​

Right now, the only buyer is OpenAgents. We've distributed and paid for a million and a half sats worth of compute over the last two weeks. We want this compute to train our models. But you can envision, months down the road, if there are agents that need compute, they're going to be bidding against us. We want your compute to be bid for by a lot of AI agents.

​

We're actually coming close to time here. What's a takeaway for the audience? If they wanted to jump into this world today or later this week, what's the first thing they should do? Roland?

​

If you're not running an agent yet, I definitely recommend you run one. There are very simple options. With a credit card, you can set up an agent, so it's super easy. If you already have an agent, then try to get it to do something without your help. If it gets stuck, see if there's a Bitcoin service it can pay instead to get unblocked without you having to do the work for your agent.

​

Erik?

​

I want you guys to experience what it's like to receive Bitcoin without a wallet. Come find me. I'm going to go backstage. If anyone here has BitChat, I'm going to start dropping Cashu strings of text on BitChat, so you can receive them through Bluetooth.

​

I want you to experience what it feels like to receive Bitcoin through Bluetooth, through Signal, through WhatsApp, without a wallet, and then think further about that for agents. To me, that's the big unlock. That's how it should feel.

​

Christopher?

​

The first thing you should do is sell your spare compute to OpenAgents. Let us pay you Bitcoin for it. OpenAgents.com. Whether you have a computer with you today or when you get back home, OpenAgents.com. It's literally a button you click and paste into your agent.

​

If you don't have an agent, I'd recommend the Hermes agent or OpenCall. We're going to interoperate with a bunch of those, and it will set everything up for you.

​

We've got a few people from our team. OpenAgents guys, raise your hands. If you want help troubleshooting, they're in the back. You have compute, we want to buy it from you, and then do cool stuff with it. Take our Bitcoin. We want your compute. The end.

​

All right, I think that's probably where we should stop. We're just about out of time. Thanks so much, everybody. This was awesome.