Several forward-looking thoughts on web3 AI Agent

Author: Haotian

Further thought about the implementation scenarios of web3 AI Agents, extracting several forward-looking insights, as follows:

1. The most native application function of the web3 AI Agent may not be "trading". Although DeFi trading Agents have always been regarded as the Endgame form for Agents to land in Crypto, the AI itself carries ambiguity in reasoning and hallucination processes, which is inherently contrary to the precision and low tolerance for error required in trading scenarios.

In my opinion, the advantages of short-term web3 AI Agents lie in the "data cleansing" and "intent analysis" levels, rather than immediately landing on the asset trading execution level with absolute precision. For example: cleansing on-chain and off-chain applicable data to build effective information graphs; furthermore: modeling on-chain user trading behaviors and analyzing risk preferences, customizing Smart Money trading decision assistants, etc.;

2. The need for web3 AI Agent for the A2A communication protocol functionality may be greater than that of MCP. This is because MCP calls are relatively mature functional API interfaces. If there is a mature Agent application ecosystem, MCP can perfectly solve the data island problem. Conversely, if the application landscape itself is immature, the standardized interfaces of MCP lack utility.

In contrast, the A2A protocol can create a certain incremental market for Agents, leading to the emergence of a batch of specialized vertical Agents, such as on-chain data analysis Agents, smart contract auditing Agents, MEV opportunity capturing Agents, and so on. The built-in Agent capability registry and P2P messaging network of A2A will encourage each vertical Agent to better adapt to the linkage and complex interaction value. If it remains at the MCP protocol level, it may be difficult for web3 AI Agents to break through the limitations of language interaction.

3. The demand for infrastructure construction by web3 AI Agents > Application implementation. In the context of web2 AI, the practicality value of Agents is naturally the highest priority, but for web3 AI Agents to build a complete ecosystem, they must fill the severely lacking underlying infrastructure, including unified data layers, Oracle layers, intent execution layers, decentralized consensus layers, etc.

Instead of directly competing with Web2 at the application layer (which is bound to suffer losses), it is the right path to explore a different avenue at the infra layer and build an infra with differentiated advantages of Web3. Although there is a relative lag in application implementation compared to Web2 AI, building a decentralized consensus network for A2A operations and establishing a unified interactive operation standard for MCP to exert its utility are foundational infra that naturally align with the native characteristics of blockchain. The urgency of building infra is not much less than that of application implementation.

4. The shift in the mindset from Crypto Native to AI Native in building. Looking back at the history of Crypto over the past few years, the adherence to the framework of "decentralization" has given rise to a rich variety of tracks and waves of innovation. In the future, the AI + Crypto field may go further around the concept of "AI autonomy."

Whether it is Agentic or Robotic, the essence is to pursue a brand new AI-centered paradigm framework, such as a cluster of AI Agents with self-funding management capabilities, a set of intelligent contract templates that can self-upgrade based on network environment and feedback, and a DAO governance framework that dynamically adjusts and optimizes based on community contributions, etc. In the end, it boils down to extracting the simplistic tool application mindset, allowing AI to have a self-evolving system, and driving AI progress through AI is the hard truth.