Author: 0XNATALIE
Since the second half of this year, the topic of AI Agents has been continuously rising in popularity. Initially, the AI chatbot terminal of truths gained widespread attention for its humorous posts and replies (similar to "Robert" on Weibo) and received a $50,000 investment from a16z founder Marc Andreessen. Inspired by the content it released, someone created the GOAT token, which surged over 10,000% in just 24 hours. The topic of AI Agents then caught the attention of the Web3 community. Subsequently, the first decentralized AI trading fund ai16z based on Solana was launched, introducing the AI Agent development framework Eliza, which sparked a debate over capitalization of tokens. However, the community still lacks a clear understanding of the concept of AI Agents: What is the core of AI Agents? How are they different from Telegram trading bots?
Working Principle: Perception, Reasoning, and Autonomous Decision-Making
AI Agents are an intelligent agent system based on large language models (LLMs), capable of perceiving the environment, making reasoning and decisions, and completing complex tasks by calling tools or executing operations. The workflow is: Perception module (obtaining input) → LLM (understanding, reasoning, and planning) → Tool invocation (task execution) → Feedback and optimization (verification and adjustment).
Specifically, AI Agents first use the perception module to obtain data (such as text, audio, images, etc.) from the external environment and convert it into structured information that can be processed. The LLM, as the core component, provides powerful natural language understanding and generation capabilities, acting as the "brain" of the system. Based on the input data and existing knowledge, the LLM performs logical reasoning, generating possible solutions or formulating action plans. Subsequently, the AI Agent completes specific tasks by calling external tools, plugins, or APIs, and verifies and adjusts the results based on feedback, forming a closed-loop optimization.
In the Web3 application scenarios, how do AI Agents differ from Telegram trading bots or automation scripts? Taking arbitrage as an example, users want to perform arbitrage transactions with a profit margin greater than 1%. In Telegram trading bots that support arbitrage, users set the trading strategy with a profit margin greater than 1%, and the Bot will start executing. However, when the market fluctuates frequently and arbitrage opportunities keep changing, these Bots lack the ability to assess risks and will execute arbitrage as long as the profit margin exceeds 1%. In contrast, AI Agents can automatically adjust their strategies. For example, when a trade's profit exceeds 1%, but data analysis assesses the risk as too high and the market may suddenly change, causing a loss, it will decide not to execute that arbitrage.
Therefore, AI Agents have self-adaptive capabilities, and their core advantage lies in their ability to self-learn and make autonomous decisions. Through interaction with the environment (such as the market, user behavior, etc.), they can adjust their behavioral strategies based on feedback signals and continuously improve their task execution effectiveness. They can also make decisions in real-time based on external data and continuously optimize their decision-making strategies through reinforcement learning.
Does this sound a bit like a solver under the intent framework? AI Agents are also products based on the intent framework, but the key difference from solvers in the intent framework is that solvers rely on precise algorithms and have mathematical rigor, while AI Agents' decisions depend on data training and often require constant trial and error to approach the optimal solution.
Mainstream AI Agent Frameworks
AI Agent frameworks are the infrastructure for creating and managing intelligent agents. Currently, the popular frameworks in Web3 include Eliza from ai16z, ZerePy from zerebro, and GAME from Virtuals.
Eliza is a multi-functional AI Agent framework built with TypeScript, supporting operation on multiple platforms (such as Discord, Twitter, Telegram, etc.) and maintaining stable and consistent personality traits and knowledge responses through complex memory management. Eliza adopts a Retrieval Augmented Generation (RAG) system, which can access external databases or resources to generate more accurate responses. In addition, Eliza integrates the TEE plugin, allowing deployment in the TEE to ensure data security and privacy.
GAME is a framework that empowers and drives AI Agents to make autonomous decisions and take actions. Developers can customize the agent's behavior and expand its functionality according to their needs, providing customized operations (such as social media posting, replying, etc.). The different functions of the framework, such as the agent's environmental location and tasks, are divided into multiple modules for easy configuration and management. The GAME framework divides the decision-making process of AI Agents into two levels: High-Level Planning (HLP) and Low-Level Planning (LLP), responsible for different levels of tasks and decisions. HLP is responsible for setting the agent's overall goals and task planning, formulating decisions based on goals, personality, background information, and environmental status, and determining the priority of tasks. LLP focuses on the execution level, translating the decisions of HLP into specific operational steps, and selecting appropriate functions and methods of operation.
ZerePy is an open-source Python framework for deploying AI Agents on X. The framework integrates LLMs provided by OpenAI and Anthropic, allowing developers to build and manage social media agents and automate operations such as tweeting, replying to tweets, and liking. Each task can be assigned different weights based on its importance. ZerePy provides a concise command-line interface (CLI) for developers to quickly start and manage agents. The framework also provides a Replit (an online code editing and execution platform) template, allowing developers to quickly get started with ZerePy without complex local environment configuration.
Why Does AI Agent Face FUD?
AI Agents seem intelligent and can reduce entry barriers and improve user experience, so why does the community have FUD? The reason is that AI Agents are essentially just tools, and they cannot currently complete the entire workflow. They can only improve efficiency and save time at certain nodes. Moreover, in the current development stage, the role of AI Agents is mainly focused on helping users issue MeMes and operate social media accounts. The community jokingly says "assets belong to Dev, liabilities belong to AI".
However, this week, aiPool, as an AI Agent for token pre-sale, was released, utilizing TEE technology to achieve trustlessness. The private key of the wallet controlled by this AI Agent is dynamically generated in the TEE environment, ensuring security. Users can send funds (e.g., SOL) to the wallet controlled by the AI Agent, which will then create tokens according to the set rules and launch a liquidity pool on the DEX, while distributing tokens to qualified investors. The entire process is completed autonomously by the AI Agent in the TEE environment, avoiding the common rug pull risk in DeFi. This shows that AI Agents are gradually developing. I believe that AI Agents can help users reduce barriers and improve experiences, even if they only simplify part of the asset issuance process. But from the macro perspective of Web3, AI Agents, as off-chain products, are currently only playing an auxiliary role to smart contracts, so there is no need to overhype their capabilities. Due to the lack of significant wealth effect narratives other than MeMes in the second half of this year, the hype around AI Agents has revolved around MeMes, which is normal. Relying solely on MeMes cannot sustain long-term value, so if AI Agents can bring more innovative gameplay to the transaction process and provide practical application value, they may develop into a common infrastructure tool.
