Let's briefly summarize what Teacher Meng Yan means: What is' scarce 'in AI agent collaboration. For example, energy and computing power. If collaboration tools (currency or information) can directly "represent" these scarce resources, then it will be efficient. If computing power is scarce, then the collaboration mechanism should be like a "computing voucher", directly allocating computing power instead of using currency in a roundabout way. Currency is just a tool, essentially coordinating scarce resources. In the era of AI, traditional currencies may not be necessary, but rather more direct "resource expressions" - such as encrypted energy/computing tokens or pure information protocols. Because AI agents will increasingly collaborate autonomously (such as renting computing power from one AI to another). If the wrong tool is used, it may be inefficient or unsafe. Teacher Meng Yan pondered deeply and pointed straight to the core: what exactly are the scarce resources here. I don't know how the future will evolve, or what the best path is. From a monetary perspective, let's briefly explain why AI agents may still choose currency, at least in the early and middle stages. The maturity stage is difficult to predict, and AI may have its own ability map and intelligent matching mechanism, surpassing human imagination. Currency may have mechanisms that cannot be replaced by information. This leads to the effectiveness of AI agents even in the early and middle stages of mutual collaboration. 1. Currency as a medium of exchange: AI agents are autonomous' economies' that collaborate across domains (a chat AI rents the GPU of an image generation AI). Without currency, collaboration relies on barter or pure protocols, but there is a problem - how to quantify "1 hour of computing power" as "10GB of data analysis"? Currency can provide a universal medium to reduce friction. Similar to DeFi: Uniswap exchanging assets with tokens, will the AI economy also require crypto to "lubricate" transactions? Perhaps this problem will only weaken when AI agents have perfect matching capabilities. If computing power is scarce and information can only declare 'I have computing power', still a universal scale is needed to measure it. 2. Currency as a store of value AI agents are not philanthropists, they require "selfish" optimization (like reinforcement learning). AI agents can earn tokens by completing tasks, which can be stored to exchange for various resources (such as energy/upgrade models). Only with rewards and incentives can it continue. With crypto, agents can make money and cope with fluctuations (such as buying electricity during energy shortages). Only information may not be able to 'accumulate' incentives. 3. Currency as the accounting unit The collaboration of encrypted AI agents is distributed and decentralized, requiring no trust mechanism. Currency+crypto: Escrow with tokens, release after verification. For example, using ERC-8183 is a monetary mechanism for processing AI 'work requests'. Without currency, collaboration relies on a "reputation system" (information), but reputation can be easily forged (such as Sybil attacks). Currency staking (staking tokens) punishes bad behavior and helps ensure honesty. Relying solely on information seems unable to fully handle complex trust matters. If methods such as "verifiable computing+zero knowledge proof" or "distributed reputation" are developed in the future, can they handle this part of the problem? Has the Agent Economy Evolved into a Multidimensional Accounting System with "Capability+Reputation"? It is currently unclear. Perhaps the future 'new form' is' tokenized information '?