After further thinking about the direction of web3 AI Agent landing scenarios, we have extracted some forward-looking thoughts as follows:
1) The most original application function of web3 AI Agent may not be “trading”. Although the DeFi trading agent has been regarded as the endgame form of agent landing Crypto, AI itself is characterized by fuzzy reasoning. However, AI itself is characterized by fuzzy reasoning and illusionary process, which is naturally contrary to the precision and low fault tolerance required by trading scenarios.
In my opinion, in the short term, the advantage of web3 AI Agent is at the level of “data cleaning” and “intent analysis”, rather than immediately landing on the level of executing asset transactions with absolute accuracy. For example, it can clean on-chain and off-chain applicability data to build an effective information map; or for example, it can carry out modeling of on-chain users’ trading behavior and risk preference analysis to customize Smart Money trading decision-making assistant and so on;
(2) web3 AI Agent’s need for A2A agent communication protocol function may be greater than that of MCP, because MCP calls relatively mature functional API interfaces, if there is a mature agent application ecosystem in the premise, based on MCP, it can perfectly solve the problem of data silo, on the contrary, if its own application industry is immature, the standardized interfaces of MCP lack of utility. On the contrary, if the application industry is not mature, the standardized interface of MCP will be useless.
In contrast, A2A protocol can create a certain incremental market for agents, which will give rise to a number of specialized vertical agents, such as on-chain data analysis agents, smart contract auditing agents, MEV opportunity capture agents, etc. The built-in Agent Capability Registry and P2P messaging network of A2A will enable the vertical agents to better adapt to the linkage and complex interactions. Conditions such as A2A’s built-in agent capability registry and P2P messaging network will make each kind of pendant agent better adapted to the value of linkage and combination of complex interactions.
(3) Web3 AI Agent’s demand for infra construction > Application landing. The pursuit of practical value of Agent in web2AI context naturally has the highest priority, but for web3 AI Agent to build a complete ecosystem, it is necessary to fill in the seriously missing underlying infrastructure, including unified data layer, Oracle layer, intent execution layer, decentralized consensus layer, etc. The web3 AI Agent can be used to build a complete ecosystem, but it is not necessary to build a complete ecosystem for web3 AI Agent.
Instead of fighting with web2 in the application layer (which is destined to suffer), it is the right way to build infra with the differentiation of web3 in the infra layer. Although there is a lag in application landing compared to web2 AI, but building a decentralized consensus network for the operation of A2A, building a unified and interactive operation standard for the functioning of MCP and other basic infra are naturally highly compatible with the original characteristics of blockchain, so the urgency of building infra is not much worse than application landing.
(4) From Crypto Native to AI Native build mindset shift, looking back at the past many years of Crypto history, only one sentence “decentralized” framework adherence to derive a rich variety of tracks and innovation tide, the future of the AI + Crypto field, may be around the “AI autonomy” to take a longer way.
Whether it is Agentic or Robotic, the essence is to pursue a new set of AI-centric paradigm frameworks, such as a set of AI Agent clusters with self-funding management capability, a set of smart contract templates that can be upgraded according to the network environment and feedback, and a set of DAO governance frameworks that can be dynamically adapted and optimized based on the community’s contribution, etc. In the final analysis, it is important to detach from the simple AI+Crypto frameworks. In the final analysis, it is the hard way to get rid of the simple thinking of tool application, let AI have an autonomous evolution system, and let AI drive the progress of AI.
