Bitcoin + AI: A Path Forward

Why Bitcoin needs deeper integration with AI.

There is no question that the two transformational technologies of our time are Bitcoin and AI. Simply put, Bitcoin is the new value layer for society, and AI is the intelligence layer. It is natural and important that these two technologies fit together, rather than remain apart.

There are two secular trends that I expect to continue in the next decade. First, Bitcoin will continue on its trajectory to gain worldwide adoption as the premium store of value. Second, machines will continue to get smarter, with more economic activity shifting to artificial agents rather than humans. It is certainly possible for these two technologies to mix, but I do not believe this is inevitable. Rather, this will take deliberate intention and commitment from intellectuals, innovators, entrepreneurs, and capitalists. Let’s break this down step by step.

Where we are today

Bitcoin is now emerging as the clear market leader among cryptocurrencies. With greater institutional adoption daily, a raft of new exchange-traded funds (ETFs), favorable regulatory approval, and more companies holding Bitcoin as their treasury asset, all signs point to greater use and adoption of Bitcoin. Bitcoin has fought its wars with other cryptocurrencies and won. The world is now realizing that the proof-of-work consensus and high security of Bitcoin give it an edge over the alternative cryptocurrencies (altcoins), that lack those key features.

For good or for ill, Bitcoin is unique in its immaculate conception, the perfect storm of events of a founder/creator who no longer guides its development. The anonymity and absence of Satoshi Nakamoto is one of Bitcoin’s core value propositions. Unlike every other cryptocurrency that is governed by a founder, founding team, or foundation, no one is in charge of Bitcoin. For this reason, changes to Bitcoin are difficult and involve and require near global consensus. With Bitcoin, there is no pre-mine, no hard fork, no pump and dump, no outside investors. Everyone comes to Bitcoin on the same terms, without permission and with proof-of-work. In this most poetic irony, the world trusts Bitcoin because it is the most trustless system that exists.

We are squarely in the middle of the store of value thesis for Bitcoin. Bitcoin’s fundamental scarcity, predictability, and unchangeability make it the best store of value for financial assets. This will attract billions and soon trillions of dollars of capital and will be Bitcoin’s primary contribution to the world in our lifetime. Many people dismiss store of value as trivial, but the numbers say otherwise. Private holdings of gold are five to six times Bitcoin’s current market cap, and that can easily flip in a decade. Bitcoin can eat market share from other assets like bonds, real estate, and equity.

Store of value is the first step in Bitcoin’s evolution. Medium of exchange is the next. How do you achieve Bitcoin adoption across 8 billion people? Not everyone will be able to afford to transact on chain. Bitcoin’s answer to this is second and third-layer protocols on top of the base layer. The Lightning network is the first and main second layer, allowing for fast instant payments between nodes over a payment network, completely separate but tethered to the underlying Bitcoin network. And now newer protocols like Fedimints and eCash allow for transfers of value on top of the Lightning network, where a Chaumian Mint issues eCash to other users that will ultimately settle on Lightning and therefore Bitcoin.

Both of these second and third-layer innovations were motivated by achieving Bitcoin adoption for human users. For Lightning, the modal use case was buying coffee at Starbucks. And for Fedimint and eCash, it is allowing a village in Africa to engage with Bitcoin. As is apparent from the protocols and their design, the ultimate end user is a human who wants to exchange in Bitcoin. But therein lies a challenge and an opportunity. As more economic activity shifts to artificial agents, the ultimate users of Bitcoin will be machines rather than humans. Machines will need to transact and operate in a digital world, so it is natural for them to use digitally native value.

The new AI agent economy

The growth of generative AI in just the last five years is the second most astounding innovation in technology. Large language models (LLMs) are a step change in the 50-year history of AI, in ways that no one anticipated. Deep learning of neural networks works in ways that few would have anticipated a decade ago. There is, of course, a lot of hype about AI, but I recommend you just try using ChatGPT to write some code for you. You will see that there is something here. Maybe not AGI yet, but real progress.

The frontier of generative AI today is the growth of autonomous AI agents. These agents will communicate, negotiate, and transact with each other. At first, with humans in the loop, but ultimately without. Agents are a natural next step for LLMs since two LLMs can engage with each other in faster ways than before. But more to the point, this agent-based approach is actually a throwback to classical AI. The original models of AI from as early as the 1980s all put an agent at its core: a learning agent, a reasoning agent, a search agent. LLMs can therefore build off of the intellectual infrastructure that is already in place.

Today, the frontier of AI agents is happening in AI startups, mostly out of SF. LlamaIndex, LangChain, OpenAI, DoubleO.ai, and Sierra.ai are all building tools for articulating complex multi-agent systems. The engineer designs a workflow and structures the interactions between these agents, and then unleashes the agents. This allows a live laboratory of agents interacting with each other, such as negotiating and trading.

A big question is how these agents will interact with each other. Will they be interacting inside organizations (like Salesforce), or between organizations (like in the market)? My thesis is that the long-term future will be an entire economy of these AI agents. The core economic transaction in the economy occurs between separate individuals or corporations. Therefore, AI agents will need to transact with each other in a distributed, decentralized way. They will have budgets, objectives, and knowledge (or at least assumptions) about how to engage with other agents in the marketplace. And just as with human economies, these markets will need some light touch, not heavy-handed rules of the road through regulation or market infrastructure, like a way to transact and transfer value.

Enter Bitcoin

As the world’s distributed value layer, Bitcoin already has the tools for the transfer of value. What’s more, Bitcoin is a natural fit for artificial rather than human agents. The core elements of Bitcoin are private keys, digital signatures, and transactions on the blockchain. These are all computational objects that are more easily handled by machines than humans. Any user knows that the UI/UX of Bitcoin sucks. Developers over time have built tools to better make Bitcoin more usable for humans, like 24-word seed phrases to represent private keys. But machines don’t need the seed phrases; they can easily handle the large random numbers that are private keys. Humans fight to adopt Bitcoin, but machines are born into it.

The undiscovered country lies in unleashing machine intelligence on Bitcoin transactions. The last two major upgrades to Bitcoin, Taproot and SegWit, allowed for vast transactional complexity within Bitcoin transactions. Ordinals and inscriptions emerged in the last two years as the first use case of this greater complexity, allowing Bitcoin users to transfer data on the chain. But this is the beginning, not the end. Large Merkle trees on the order of 128 levels deep can now exist within a Bitcoin transaction. This offers computational complexity far more than any human could ever use.

A Simple Example: Degraded MultiSig

To fix ideas, consider a classic two-of-three multi-signature transactions between, say, Alice, Bob, and Carol. Alice, Bob, and Carol each have a private key. To send funds, a transaction requires two of three signatures. One problem is that the funds are trapped if two of the three lose their keys. Taproot allows for degraded multisig, which uses time locks to adapt multisig transactions over time. After 90 days the transaction can morph into a one-of-three multisig, in case two of the three people lose their keys. After a year, it can degrade into a one-of-one multisig to a separate person (like David) in case the original three all die in a plane crash. These are just some simple examples of the new power that Taproot unlocks.

Artificial agents can then optimize to discover the optimal multisig scheme. Is it two of three, three of five, five of seven? Should it degrade after 90 days or 180 days? And degrade to what specific configuration? We can design these by hand now, but a better approach is for the machines to figure it out optimally.

This is just the first step. Degraded multisig is a use case that we see and can understand now. But machines can discover their own use cases within Taproot that we are unable to even imagine. This is the true undiscovered country, where artificial agents will embed complex programs inside these large Merkle trees.

BitVM, recently released by Stanford Ph.D. student Robin Linus, gives a glimpse of the full power of what is possible under Taproot. It is now possible to run entire programs on Bitcoin, through formal verification systems from cryptography. In particular, you can represent an entire circuit on the Bitcoin blockchain. These early experiments are just the first steps in harnessing the full computational complexity that is now possible on Bitcoin. Artificial agents will unlock this to a whole new level.

A Roadmap

A path forward depends on our knowledge of technologies today as well as some experiments of new technologies that have not yet been invented. Today we know that LLM agents are the leading technology for AI. They already have the ability to communicate and negotiate with each other. The next step is to build the connective tissue between these LLM agents and the Bitcoin blockchain. For example a fairly straightforward goal would be to build two LLM agents that negotiate over a multi-sig transaction, find the optimal transaction, and push it to the Bitcoin blockchain through a pay-to-taproot (p2tr) transaction. This would be a fun experiment to run. And then we could repeat many of the existing use cases for Bitcoin transactions under the guide of these LLM agents.

The longer-term opportunity is to find ways for artificial agents to structure novel Taproot transactions with minimal oversight or design by humans. Engineers would just need to build the frameworks and systems for the agents to interact, and methods to assess their performance and test the outcomes. This is possibly the most vague but also the most exciting opportunity for research because we do not even know how agents will utilize Bitcoin’s new transactional complexity. They may independently generate new use cases that we cannot even imagine right now.

Why we need to fight for this

This fusion of Bitcoin and AI is possible but not inevitable. I’m confident that AI will continue to evolve on its own, as there are strong economic interests to see machine intelligence improve. On the Bitcoin side, there is strong momentum for a better store of value, and that also will proceed on its own. But AI agents using Bitcoin may or may not happen. Today, there are startups like Skyfire.xyz that are empowering artificial agents to use stablecoins. Absent any concerted effort from the Bitcoin community, it’s entirely possible that the agents of the future would simply continue to use stablecoins and, therefore, ossify the existing payment rails of the fiat system.

This would be one future, but it is not the best future. Ultimately, artificial agents should be transacting in the world’s most secure and sound money. And so let me issue this call to action. Let’s fight for machines to adopt Bitcoin. And in a poetic twist, thus allow machines to secure our human economic freedom.