As the cost of producing artificial intelligence models decreases, the population of AI agents will grow exponentially. Agents will soon outnumber humans online, creating, consuming, and exchanging multitudes more information than we ever could. But if we get, say, a million-fold increase in digital activity, and 99% of that growth comes from machines, it will be hard to cope with this transformation without adopting onchain infrastructure and business models that both empower agents to reach their full potential and allows us to identify, control and audit their actions.
Today, companies like OpenAI take on the massive costs of producing models and then sell access via their proprietary interfaces and APIs, and they’re mostly limited to consuming and creating content. But to fully harness the potential of AI, we need a vast population of specialized agents that can talk to and transact with each other. They must be free to roam the internet and be able to have and spend money to execute tasks on our — or their own — behalf. We also need ways to identify, control, and audit their actions.
The problem is that we can’t coerce agents to follow our laws, preventing effective regulation. They can’t use the traditional financial system, which depends on a jurisdictional identity model, limiting their ability to transact. They consume a ton of online information — imposing all the traffic costs on service providers — but don’t generate the corresponding revenues by subscribing or clicking on ads. To solve these problems, we need a natively digital legal and financial system combined with new business models to take full advantage of the opportunities presented by this new technology.
The solution requires (1) sovereign digital infrastructure with a new software paradigm that guarantees trusted code execution with an immutable audit trail; (2) an independent, digital financial system that treats humans and machines alike; and (3) a cryptographic identity model coupled with decentralized communication and reputation protocols. This is only possible using blockchain protocols and smart contract applications.
Blockchain protocols provide all sorts of decentralized digital services that are accessible via smart contracts and payable with digital assets. For example, smart contract networks like Ethereum and Solana enable secure, reliable execution of open-source software backed by an auditable trail of blockchain transactions, and networks like Filecoin and Arweave provide cheap and scalable onchain data storage services. As it becomes easier and cheaper to build new protocols on top of established platforms, the range of services offered via decentralized networks is expanding.
We can use these platforms to train, deploy, and operate agents in a decentralized manner, but more importantly, the ability to consume them via smart contracts facilitates agent interactions. It’s much harder for AI agents to consume a typical Web2 REST API than to read a smart contract and pay for its service with tokens – no accounts or credit cards required.
Wallet-enabled agents can use any smart contract service or platform, from infrastructure services to DeFi protocols to social networks, which opens a whole universe of new capabilities and business models. An agent could pay for its own resources as needed, whether it’s computation or information. It could trade tokens on decentralized exchanges to access different services or leverage DeFi protocols to optimize its financial operations. It could vote in DAOs, or charge tokens for its functionality and trade information for money with other specialized agents. The result is a vast, complex economy of specialized AI agents talking to each other over decentralized messaging protocols and trading information onchain while covering the necessary costs. It’s impossible to do this in the traditional financial system.
Consider this consequence: if agents act onchain – even if they think offchain – we end up with a public, immutable and cryptographically-signed record of their activity over time. Blockchains will ensure the safe deployment of AI at scale, enabling us to do things like audit agents’ internet actions, distinguish machine-made from human-made content, and build identity and reputation systems for machines based on their onchain activity. It will help us and them identify and reward good actors (with tokens or reputation), punish bad ones (e.g., by slashing), and tell which agent performs better than another at a particular task. Agents will then be able to decide who to rely on based on their onchain history, which they can easily access thanks to the open-source nature of blockchain data.
Many pieces are coming together to enable this vision. Blockchain infrastructure is rapidly becoming fast and cheap thanks to new consensus mechanisms and scaling solutions. Smart contract wallets and “wallet as a service” (WaaS) providers will enable agents to transact; meanwhile, account abstraction and other emerging techniques can allow human-agent interactions where we can authorize agents to spend from our wallets. We can build registries and reputation systems (including blocklists or slashing mechanisms, for example) using agents’ public keys as identifiers. We can even play with DAO-owned agents and experiment with new business models; perhaps some agents will live on their own Layer 2 network, owned and managed by a decentralized community of operators.
These are the kinds of ideas that initially seem far-fetched but turn out to be very obvious in hindsight. Of course, smart contracts will mediate commercial activity between agents just as legal contracts govern that of humans. Of course, agents will use the internet’s financial system powered by digital assets instead of banks and credit cards. Of course, agents will identify, communicate, and transact with each other using cryptographic identities via decentralized protocols. I don’t see another way around it: artificial intelligence belongs onchain.