Very important
According to PANews, the potential adoption of A2A protocols from Google and MCP from Anthropic as standards for communication for AI agents in web3 faces significant challenges due to the clear differences between web2 and web3 environments.
The first challenge
lies in the maturity of applications. While A2A and MCP gained rapid momentum in the web2 space by optimizing already mature application scenarios, AI agents in web3 are still in early stages of development, lacking deep application contexts like DeFAI and GameFAI.
This makes it difficult to directly apply and effectively use these protocols in a web3 environment.
For example, in web2, users can seamlessly update code on platforms like GitHub using the MCP protocol without leaving their current workspace. However, in a web3 environment, executing on-chain transactions with locally trained strategies can be confusing when analyzing on-chain data.
Another significant barrier is the lack of foundational infrastructure in the web3 space.
To build an inclusive ecosystem, AI agents in web3 must address the lack of essential components such as a unified data layer, an oracle layer, an intent execution layer, and a decentralized consensus layer. In web2, A2A protocols allow agents to collaborate easily using standard APIs. In contrast, web3 environments pose significant challenges even for simple arbitrage operations across DEX.
Consider a scenario where a user instructs the AI agent to buy ETH from Uniswap when the price drops below $1600 and sells when it rises. This seemingly simple task requires the agent to deal with specific issues in web3 such as real-time on-chain data analysis, optimizing dynamic gas fees, controlling slippage, and protecting MEV.
In web2, such tasks are simplified through standard API calls, highlighting the clear difference in infrastructure maturity between the two environments.
Furthermore, AI agents in web3 must address unique requirements that differ from web2 protocols and functionalities.
For instance, in web2, users can easily book the cheapest flight using A2A protocols. However, in web3, when a user wants to transfer USDC across chains to Solana for liquidity mining, the agent must understand the user's intent, balance security, consistency, and cost efficiency, and execute complex operations on-chain. If these operations increase security risks, the perceived convenience becomes meaningless, rendering the demand a false need.
In conclusion, while the value of A2A and MCP protocols cannot be denied, the expectation that they will seamlessly adapt to the landscape of AI agents in web3 without modifications is an unrealistic expectation.
Gaps in infrastructure deployment provide opportunities for builders to innovate and fill these voids.