How does Vitalik view the AI agent economy and what specific application directions would he suggest? At Devconnect's Trustless Agent Day, Vitalik and Davide Crapis from the Ethereum Foundation delved into the fundamentals of AI agents. The host is @ tzhen from Flashbots. This conversation is not only about technology, but also a deduction for the future digital society. The core is built on two puzzle pieces: x402 that solves "how to pay" and ERC-8004 that solves "who to pay". @Vitalik Buterin believes that the intervention of AI has truly made "micro payments" feasible. Humans cannot tolerate judging whether to pay a few cents for each service, but for agents, this is only a millisecond level decision. However, high-frequency payments must be built on top of privacy. He emphasized the use of ZK technology to achieve "non associative" query credentials: privacy should not be an additional feature of a product, but a necessary hygiene habit. Solved the payment issue, while the ERC-8004 promoted by the @ DavideCrapis team attempted to address the trust problem. This is not just a whitelist, but a decentralized service discovery mechanism that breaks down trust into "soft trust" based on reputation accumulation and "hard trust" based on cryptographic proofs. But there is still a gap in between: there is a crisis of "role-playing" in the current industry, agents understand the theory of making cakes but have never been in the kitchen, and the inference cost is high, all of which are urgent "elephants in the room" that need to be solved. Why do we have to build agents on the blockchain? Vitalik's perspective is very sharp: this is the native habitat of the agent. They do not require human authentication and are more suitable for handling complex 'Synthetic Assets' in anonymous environments. More importantly, 'restricted delegation' allows agents to manipulate funds for arbitrage, but the underlying code is written to 'prohibit withdrawals', providing security that Web2 API cannot match. The discussion about the future of privacy and computing power has become extremely sci-fi. Vitalik proposed the principle of "user privacy>service privacy" and believed that TEE (Trusted Execution Environment) is a practical solution to balance privacy and computing power bottlenecks. He even predicted the 'demise of laptops': future UIs will be extremely fragmented (glasses, brain computers), computing cores will be decoupled from terminals, and we need a new architecture to securely use remote computing power. Finally, land on specific opportunities. Davide revealed that ERC-8004 will soon be launched on the mainnet, and the "first season" of infrastructure construction has been completed. Where are the next killer applications? Vitalik gave a very specific suggestion: Live Translation. Although current encrypted communication software protects transmission privacy, the translation function often relies on centralized cloud services, which is a huge gap for data leakage. Building a localized or trustless AI translation agent would be the perfect entry point to fill this gap. For developers, the "second season" of the Agent economy has just begun, from infrastructure to specific applications. More details https://substack.chainfeeds.xyz/p/vitalik-ai-agent-erc-8004