The development of AI is too fast, and the future will definitely be the world of AI. If we add one more core element, it will definitely be the world of AI+Crypto.

Now, AI has evolved to a new stage: AI Agent.

The imagination space and landing scenarios of AI Agent are worth looking forward to.

The train of the times is rushing by, and we need to get on it quickly.

I have also been constantly learning about AI Agent recently, and this article will record my learning path, hoping to help everyone better enter the AI Agent track.

This article is the first part of the AI Agent track entry guide, which helps to establish an overall understanding and framework-style understanding. We will continue to delve deeper into this track, keep improving, and grasp the AI wave.

01 What is AI Agent?

First, let's put aside all the complex concepts and directly compare the differences between AI Agent and the existing large models (such as ChatGPT).

The current large models are more like powerful "natural language search engines", which can answer questions and provide suggestions, but cannot truly make decisions and execute independently.

The capabilities of AI Agent go beyond the scope of the existing large models, no longer limited to "data processing", but able to complete the full closed loop from "perception" to "action".

To give a straightforward example: now if you ask ChatGPT how to invest in Crypto, ChatGPT will give you a bunch of suggestions, but an AI Agent can help you track global market information in real-time, dynamically adjust your investment portfolio, and maximize your returns.

From this, we can abstract the concept of AI Agent: AI Agent is a software entity based on artificial intelligence technology, which can autonomously or semi-autonomously perform tasks, make decisions, and interact with humans or other systems.

The core difference here is: autonomous action.

How does AI Agent achieve autonomous action?

Through AI, complex logic can be converted into precise conditions (returning True or False based on context), and then seamlessly integrated into business scenarios.

First is intent analysis: AI will understand what the user wants to do by analyzing the user's prompt words and context. It not only looks at what the user said, but also considers the user's previous usage records and specific situations, and then translates these needs into specific program instructions.

Secondly, it is decision assistance: AI is like a smart assistant, able to turn some complex problems that humans find difficult to handle into simple yes-or-no answers or a few fixed options through analysis. This not only makes decision-making more accurate and efficient, but also works well with existing business systems.

Based on the degree of autonomous action, AI Agent can be divided into two types:

One is an AI Agent that acts as a personal assistant, helping users handle some business matters.

The other goes further, where the AI Agent is an independent entity with its own identity or brand, providing services to many users.

In summary, AI Agent can be seen as the next stage of development and a new product form of large models, and AI Agent has a very large imagination space.

02 What is the relationship between AI Agent and Crypto?

AI and Crypto are not clearly separated, and the two can be integrated.

More importantly, the Web2 AI Agent and the Web3 AI Agent are not the same.

The Web3 AI Agent is a more advanced and more complete AI Agent, or perhaps it can be renamed as Crypto AI Agent.

With the capabilities of Crypto, AI Agent has more features:

(1) Decentralization

Combining with Crypto, the operation, data storage, and decision-making process of AI Agent are more transparent and not controlled by a single entity.

Web2 AI Agents are usually controlled by centralized companies or platforms, and the data and decision-making processes are concentrated in one or a few entities.

Once an AI Agent provides services to the outside world, there will be trust issues, so AI Agent needs the running or verification environment provided by the blockchain.

AI Agent also needs barrier-free usage, data transparency, interconnectivity, and decentralization.

(2) Incentive mechanism

This is the strongest empowerment of Crypto, through the token economic model, providing a direct incentive mechanism for developers and users to participate and contribute.

Web2 AI Agents mainly rely on traditional business models, such as advertising revenue or subscription services, to maintain operations.

Web2 startup teams or companies, even after a long time, cannot be profitable, and it is also difficult to raise funds; but in Web3, through the form of token issuance, they can directly obtain cash flow to support project development, such as the use of AI Agent requires Crypto payment.

A free market economy can foster more innovation.

(3) True immortality

With smart contracts, AI Agent has truly achieved "immortality".

As long as the smart contract is deployed on the blockchain, the AI Agent can operate automatically according to its rules, and theoretically can run indefinitely.

Smart contracts can ensure that the code and decision-making mechanism of the AI Agent permanently exist on the blockchain, unless there is explicit logic to stop or change its behavior.

However, the data it depends on may need to be continuously updated or maintained. If there is no continuous input of data or interaction from the outside world, the "immortality" of the AI Agent may be limited to its program logic, without dynamism.

In summary, compared to Crypto needing AI Agent, AI Agent needs Crypto more.

03 The narrative evolution of AI+Crypto

The evolution of AI from large models to AI Agents is in two stages, and the integration of AI and Crypto can also be divided into two stages:

3.1 Large model stage: infrastructure

There are three evaluation dimensions for AI projects: computing power, algorithms, and data.

In fact, the role played by Web3 is to add an incentive system to AI, tokenizing computing power, algorithms, and data.

Therefore, the intersection of AI and Web3 can also be explored from the three dimensions of computing power, algorithms, and data:

(1) Computing Power:

Distributed computing network: Blockchain naturally has a distributed nature. AI can use the distributed network of Web3 to obtain more computing resources. By distributing AI computing tasks to the nodes in the Web3 network, it can achieve more powerful parallel computing capabilities, which is particularly useful for training large AI models.

Incentive mechanism: Web3 introduces economic incentive mechanisms, such as token economics, which can incentivize participants in the network to contribute their computing resources. This mechanism can be used to create a market where AI developers can purchase computing power for machine learning tasks, and providers can receive token rewards.

(2) Algorithms:

Smart contracts: Smart contracts in Web3 can automatically execute AI algorithms. AI can design algorithms to run in the form of smart contracts on the blockchain, which not only increases transparency and trust, but also can achieve automated decision-making processes, such as automated market prediction or content moderation.

Decentralized algorithm execution: In the Web3 environment, AI algorithms can be executed by multiple nodes for verification, rather than relying on a single central server. This increases the resilience and security of the algorithms, preventing single points of failure.

(3) Data:

Data privacy and ownership: Web3 emphasizes the decentralization of data and user ownership of data. AI combined with Web3 can use blockchain technology to manage data permissions, ensure data privacy, and allow users to selectively share data in exchange for rewards, providing AI with a richer but controlled data source.

Data verification and quality: Blockchain technology can be used to verify data, ensuring the authenticity and integrity of the data, which is crucial for training AI models. Through Web3, data can be verified before being used, improving the quality and credibility of AI algorithm outputs.

Data marketplace: Web3 can promote the development of data marketplaces, where users can directly sell or share data with the AI systems that need it. This not only provides AI with a diversified data set, but also ensures the liquidity and value of the data through market mechanisms.

Through these integration points, AI and Web3 can develop synergistically:

• AI can use Web3 to obtain distributed computing power and high-quality data, and also use smart contracts to improve the efficiency and transparency of algorithm execution;
• Web3 can also be enhanced by AI to increase the intelligence of its systems, such as smart resource management and automated contract execution.

For these three dimensions, there are already several well-known projects on the market:

Computing Power (Computational Power) projects:

• Render Network: Although it mainly focuses on rendering, it can also provide AI computing power.
• Akash Network: Provides decentralized cloud computing resources that can be used for AI needs.
• Aethir: Focuses on decentralized cloud computing, which may involve the provision of AI computing power.
• ionet: A decentralized computing power platform that supports AI reasoning and training.

Algorithm (Algorithms) projects:

• Cortex: A decentralized world computer that can run AI and AI-driven DApps on the blockchain, focusing on integrating AI into smart contracts.
• Fetchai: A blockchain-based machine learning platform that has launched the no-code management service Agentverse, simplifying the deployment of AI agents for Web3 projects.
• iExec RLC: Provides a blockchain-based AI model marketplace that supports confidential computing and decentralized oracles.

Data projects:

• Vana: Vana is building a DAO for personal genetic data, allowing users to control and potentially benefit from a data marketplace.
• RSS3: Launched an open-source AI architecture that allows any large language model to become a Web3 AI agent, involving the utilization and management of data.

Comprehensive projects:

• Myshell: A decentralized AI consumption layer that aims to connect consumers, creators, and open-source researchers. It opens a platform where anyone can create, share, and monetize their AI-native applications.

Overall, in the era of large models, the integration of Crypto and AI is mainly at the infrastructure level, laying the foundation for the long-term development of AI.

3.2 AI Agent Stage: Application Landing

The emergence of AI Agents marks the application layer landing of AI.

AI Agents can also be divided into three development stages: Meme Coin Stage, Monolithic AI Application Stage, and AI Agent Framework Standard Stage.

1. AI Agent Meme Coins

AI Agent Meme Coins are a very special existence, as Meme Coins are the product of community sentiment.

AI development is too fast, and this technology also appears very profound, which makes ordinary people very anxious. AI Meme Coins gave ordinary people the opportunity to participate.

Therefore, AI Meme Coins bring a sense of participation in the AI revolution to holders, allowing ordinary people to also participate in the AI wave.

The final result is: AI + MEME uses the wealth effect to accelerate the market education and dissemination of AI.

From another perspective, why do AI Agents need to issue coins?

On the one hand, the wealth effect attracts capital and users, injecting momentum into the industry's subsequent development; on the other hand, the MEME-style issuance method itself is a community financing means, providing cash flow for the project's own development.

We can see the top targets:

• $GOAT: The first popular AI Agent Meme Coin;
• $Fartcoin: Attracts user attention by generating humorous content (such as "fart jokes");
• $ACT: Aims to create a digital ecosystem of equal interaction between users and AI;
• $WORM: Aims to combine digital biology and blockchain technology to create a unique digital asset that simulates the neural system of biological worms;

2. Monolithic AI Applications

AI Agents are integrating with various sub-tracks of Crypto, presenting a thriving situation.

With the development of AI Agents, the tokens issued by AI Agents are no longer purely Meme Coins, but have the attributes of value coins with actual use cases.

(1) Genesis Projects

• ai16z: The first AI Agent to break out, and established the first framework standard Eliza.

(2) Agent Gaming

• ARC: Developed a framework called RIG based on the Rust language, supporting decentralized applications (dApps) and smart contracts.
• FARM: Focuses on using AI to enhance the realism and strategic depth of agricultural games.
• GAME: $GAME empowers the autonomous operation and intelligence of AI agents, deeply integrating AI and games.

(3) Agent DeFi

• $NEUR: Focuses on token analysis and DeFi interaction, providing intelligent financial decision support.
• $BUZZ: Provides a natural language interface, allowing users to conduct DeFi transactions and management more intuitively.

(4) Code Auditing

• AgentAUDIT: Uses AI technology to automate code auditing, improving code security and quality.

(5) Agent Data Analysis

• REI: Provides insights and forecasting services through large-scale data analysis using AI technology.

(6) Autonomous AI Agents

• LMT: An AI Agent that can learn and execute tasks autonomously, aiming to reduce human intervention.
• GRIFFAIN: An AI Agent that can autonomously optimize its own behavior, particularly for decision-making and strategy formulation in complex environments.

3. AI Agent Framework Standards

AI Agent framework standards are still in a chaotic state.

What is an AI Agent framework standard?

AI Agent framework standards simplify the development and deployment of AI Agents by providing a set of unified specifications and tools.

It allows developers to create an AI Agent that can interact with various clients (Twitter, Discord, Telegram, etc.), extend its functionality through plugins, and leverage AI technology to enhance its intelligence.

These standards and base libraries (such as memory storage, session isolation, context generation, etc.) ensure that the operation of AI Agents is efficient, secure, and user-friendly.

By connecting to various AI platform interfaces, the framework standards further enhance the capabilities of AI Agents, allowing them to utilize the latest AI technologies to provide better services.

In summary, AI Agent framework standards are the infrastructure and platforms, and they can form their own ecosystems, with a higher narrative space than monolithic AI applications.

The main AI Agent framework standards are:

• ai16z: Built the Eliza framework, supporting multiple platforms such as Discord, Twitter, and Telegram, allowing AI Agents to seamlessly integrate with these platforms.
• Virtual: Built the GAME framework, designed specifically for games and virtual environments, allowing AI Agents to operate autonomously or interact with players in these environments.
• swarms: A multi-agent AI framework, based on which developers can create and manage multiple AI agents, suitable for scenarios requiring high complexity coordination, such as simulating social behavior, complex business process automation, or large-scale data processing.
• ZEREBRO: Built the ZerePy framework, similar to Optimism's OP Stack, making the development and deployment of monolithic AI applications easier and more standardized, allowing these Agents to independently create and distribute content on social platforms.

Related ecosystems have already emerged around these frameworks, and when researching relevant projects, we need to pay close attention to these ecosystems.

04 Summary

The narrative of AI Agents has begun to explode.

Our industry has a main narrative that explodes every year, and around this main narrative, many star projects will emerge, and naturally, there will be many opportunities.

For example, the DeFi Summer of 2020, the Meme Summer of 2023, the Meme Summer of 2024, and the AI Summer of 2025 is currently emerging.

Don't waste each of these rare wealth-creating opportunities.