Recently, I came across some in-depth thoughts from Vitalik on the future of AI and Web3, which are quite worth pondering.

His core view is: AI will fundamentally change the way we interact with blockchain, but the fundamental functions of blockchain itself—serving as a bulletin board and computation layer—will not disappear. It’s just that, whereas users previously interacted with specific things through a particular interface, now AI can invoke a variety of skills, combine tasks in one go, and interact with many different objects simultaneously.

This means that workflows involving Ethereum and other systems will greatly increase. But there’s an interesting shift here: the metaphor of an "operating system" isn’t entirely accurate. Operating systems will become smaller and simpler, while we will have a bunch of different tools and skills that AI helps users utilize.

Why does blockchain become more important in this process? Because as AI becomes more decentralized, many different AI entities will need to interact with each other. To enable this interaction, an economic layer is necessary. Cooperation is either based on economic incentives and rules or on centralized control—those are essentially the two paths. Blockchain naturally provides the possibility for such economic coordination.

In high-frequency agent trading scenarios, the role of L2 will become very critical. But more importantly, the way L2 is built should start from application needs, rather than simply copying EVM. Different functions should be distributed across different layers: accounts can be on L1, high-frequency trading and matching on L2, and privacy features can also become a dedicated L2 direction.

Regarding identity, Vitalik believes that decomposing identity is key—only prove the necessary information for completing a certain interaction. Fully exposing identity is usually meaningless; a more reasonable approach is to reveal only partial information, such as using zero-knowledge proofs to demonstrate reputation or fund sources. This applies to both humans and agents.

In the agent economy, the role of wallets is also being reshaped. On one hand, AI can be used to enhance security (e.g., formal verification); on the other hand, wallets can serve as agent wallets that integrate various capabilities. But the premise is to ensure privacy and security, without relying on third-party servers. At the same time, AI behavior needs to be limited—this is the responsibility of wallets in risk control.

He particularly emphasized a direction: ZK Payments and ZK API. The core goal is to make every API request private and completely isolated from each other. This is crucial because, in AI scenarios, even if pseudonyms are used, if identities persist, accumulated information will eventually lead to re-identification. Therefore, mechanisms must be in place to ensure that requests do not become linked.

These thoughts have profound implications for the future direction of Web3 infrastructure. From ERC-4337 to EIP-7579, and most recently EIP-8141, we see a gradual evolution—building a more solid foundation for agent economies while protecting privacy, reducing costs, and enhancing security. That’s also why many wallet projects are now exploring agent capabilities, trying to seize the opportunity in this new era.