From AI Agent to DeFAI: Unraveling the Driving Forces Behind the New Wave of Encryption Bull Run

1. Background Overview

1.1 Introduction: "New Partners" in the Smart Era

Each cryptocurrency cycle brings new infrastructure that drives the entire industry forward.

• In 2017, the rise of smart contracts spurred the vigorous development of ICOs.
• In 2020, the liquidity pools of DEX brought about the summer boom of DeFi.
• In 2021, the emergence of numerous NFT series marked the arrival of the era of digital collectibles.
• In 2024, the outstanding performance of a certain platform led the trend of memecoins and launch platforms.

It should be emphasized that the emergence of these vertical fields is not solely due to technological innovation, but rather the result of a perfect combination of financing models and bull run cycles. When opportunity meets the right timing, it can lead to tremendous changes. Looking ahead to 2025, it is clear that the emerging field of the 2025 cycle will be AI agents. This trend peaked last October, with a certain token launching on October 11, 2024, and reaching a market value of 150 million USD by October 15. Immediately after, on October 16, a certain protocol launched Luna, making its debut with the IP live streaming image of a neighbor girl, igniting the entire industry.

So, what exactly is an AI Agent?

Everyone must be familiar with the classic movie "Resident Evil"; the AI system Red Queen is particularly impressive. The Red Queen is a powerful AI system that controls complex facilities and security systems, capable of autonomously perceiving the environment, analyzing data, and taking swift action.

In fact, AI Agents share many similarities with the core functions of the Red Heart Queen. In reality, AI Agents play a similar role to some extent; they are the "guardians of wisdom" in the modern technology field, helping businesses and individuals tackle complex tasks through autonomous perception, analysis, and execution. From self-driving cars to intelligent customer service, AI Agents have penetrated various industries and become a key force in enhancing efficiency and innovation. These autonomous intelligent entities, like invisible team members, possess comprehensive capabilities from environmental perception to decision execution, gradually infiltrating different sectors and promoting a dual enhancement of efficiency and innovation.

For example, an AI AGENT can be used for automated trading, managing portfolios and executing trades in real-time based on data collected from certain data platforms or social platforms, continuously optimizing its performance through iterations. The AI AGENT is not a single form, but is divided into different categories based on specific needs in the encryption ecosystem:

1. Execution AI Agent: Focused on completing specific tasks, such as trading, portfolio management, or arbitrage, aimed at improving operational accuracy and reducing the time required.

2. Creative AI Agent: used for content generation, including text, design, and even music creation.

3. Social AI Agent: As an opinion leader on social media, interact with users, build communities, and participate in marketing activities.

4. Coordinating AI Agent: Coordinates complex interactions between systems or participants, especially suitable for multi-chain integration.

In this report, we will delve into the origins, current status, and vast application prospects of AI Agents, analyzing how they are reshaping the industry landscape and looking ahead to future development trends.

1.1.1 Development History

The development history of AI AGENT showcases the evolution of AI from basic research to widespread application. The term "AI" was first introduced at the Dartmouth Conference in 1956, laying the foundation for AI as an independent field. During this period, AI research mainly focused on symbolic methods, giving rise to the first AI programs, such as ELIZA(, a chatbot), and Dendral(, an expert system in the field of organic chemistry). This stage also witnessed the initial proposal of neural networks and the preliminary exploration of the concept of machine learning. However, AI research during this period was severely constrained by the limitations of computing power at the time. Researchers encountered significant difficulties in natural language processing and the development of algorithms that mimic human cognitive functions. Additionally, in 1972, mathematician James Lighthill submitted a report published in 1973 about the status of ongoing AI research in the UK. The Lighthill report essentially expressed a comprehensive pessimism regarding AI research after the early excitement, leading to a significant loss of confidence in AI among UK academic institutions(, including funding bodies). After 1973, funding for AI research was drastically reduced, and the field experienced its first "AI winter", with increasing skepticism about AI's potential.

In the 1980s, the development and commercialization of expert systems led global enterprises to begin adopting AI technology. This period saw significant advancements in machine learning, neural networks, and natural language processing, paving the way for the emergence of more complex AI applications. The introduction of autonomous vehicles and the deployment of AI across various industries such as finance and healthcare also marked the expansion of AI technology. However, from the late 1980s to the early 1990s, the AI field experienced a second "AI winter" as demand for specialized AI hardware collapsed. Additionally, scaling AI systems and successfully integrating them into practical applications remained a continuing challenge. Meanwhile, in 1997, IBM's Deep Blue defeated world chess champion Garry Kasparov, marking a milestone event in AI's ability to solve complex problems. The resurgence of neural networks and deep learning laid the foundation for AI development in the late 1990s, making AI an indispensable part of the technological landscape and beginning to influence everyday life.

By the beginning of this century, advances in computing power drove the rise of deep learning, with virtual assistants like Siri showcasing the practicality of AI in consumer applications. In the 2010s, breakthroughs in reinforcement learning agents and generative models like GPT-2 pushed conversational AI to new heights. During this process, the emergence of large language models (Large Language Model, LLM) became a significant milestone in AI development, especially with the release of GPT-4, which was seen as a turning point in the field of AI agents. Since a certain company released the GPT series, large-scale pre-trained models with hundreds of billions or even trillions of parameters have demonstrated language generation and understanding capabilities that surpass traditional models. Their outstanding performance in natural language processing allows AI agents to exhibit clear and coherent interaction abilities through language generation. This enables AI agents to be applied in scenarios such as chat assistants and virtual customer service, gradually expanding to more complex tasks ( such as business analysis and creative writing ).

The learning ability of large language models provides AI agents with greater autonomy. Through reinforcement learning (Reinforcement Learning) technology, AI agents can continuously optimize their behaviors and adapt to dynamic environments. For example, in a certain AI-driven platform, AI agents can adjust their behavioral strategies based on player input, truly achieving dynamic interaction.

The development history of AI agents, from early rule-based systems to large language models represented by GPT-4, is an evolutionary history of continuously breaking through technological boundaries. The emergence of GPT-4 is undoubtedly a significant turning point in this process. With further technological advancements, AI agents will become more intelligent, scenario-based, and diverse. Large language models not only inject the "wisdom" soul into AI agents but also provide them with the ability to collaborate across fields. In the future, innovative project platforms will continue to emerge, driving the implementation and development of AI agent technology and leading a new era of AI-driven experiences.

1.2 Working Principle

The difference between AIAGENT and traditional robots is that they can learn and adapt over time, making detailed decisions to achieve goals. They can be seen as highly skilled and continuously evolving participants in the encryption field, capable of acting independently in the digital economy.

The core of the AI AGENT lies in its "intelligence" ------ that is, simulating human or other biological intelligent behavior through algorithms to automatically solve complex problems. The workflow of an AI AGENT typically follows these steps: perception, reasoning, action, learning, and adjustment.

1.2.1 Perception Module

The AI AGENT interacts with the external world through the perception module, collecting environmental information. This part of the functionality is similar to human senses, using sensors, cameras, microphones, and other devices to capture external data, which includes extracting meaningful features, recognizing objects, or identifying relevant entities in the environment. The core task of the perception module is to transform raw data into meaningful information, which often involves the following technologies:

• Computer Vision: Used for processing and understanding image and video data.
• Natural Language Processing ( NLP ): Helping AI AGENT understand and generate human language.
• Sensor Fusion: Integrating data from multiple sensors into a unified view.

1.2.2 Inference and Decision Module

After perceiving the environment, the AI AGENT needs to make decisions based on the data. The reasoning and decision-making module is the "brain" of the entire system, which conducts logical reasoning and strategy formulation based on the collected information. Utilizing large language models, acting as orchestrators or reasoning engines, to understand tasks, generate solutions, and coordinate specialized models for specific functions such as content creation, visual processing, or recommendation systems.

This module typically employs the following technologies:

• Rule Engine: Simple decision-making based on predefined rules.
• Machine learning models: including decision trees, neural networks, etc., used for complex pattern recognition and prediction.
• Reinforcement Learning: Allow AI AGENT to continuously optimize decision-making strategies through trial and error, adapting to changing environments.

The reasoning process usually consists of several steps: first, assessing the environment; second, calculating multiple possible action plans based on the objective; and finally, selecting and executing the optimal plan.

1.2.3 Execution Module

The execution module is the "hands and feet" of the AI AGENT, putting the decisions of the reasoning module into action. This part interacts with external systems or devices to complete assigned tasks. This may involve physical operations ( such as robotic actions ) or digital operations ( such as data processing ). The execution module relies on:

• Robotic control systems: used for physical operations, such as the movement of robotic arms.
• API Call: Interacting with external software systems, such as database queries or network service access.
• Automation Process Management: In a corporate environment, repetitive tasks are performed through RPA( robotic process automation).

1.2.4 Learning Module

The learning module is the core competency of the AI AGENT, enabling the agent to become smarter over time. Continuous improvement through feedback loops or "data flywheels" feeds the data generated from interactions back into the system to enhance the model. This ability to gradually adapt and become more effective over time provides businesses with a powerful tool to improve decision-making and operational efficiency.

The learning module is usually improved in the following ways:

• Supervised Learning: Using labeled data for model training, enabling the AI AGENT to complete tasks more accurately.
• Unsupervised learning: Discovering potential patterns from unlabelled data to help agents adapt to new environments.
• Continuous Learning: Update models with real-time data to maintain agent performance in dynamic environments.

1.2.5 Real-time Feedback and Adjustment

The AI AGENT optimizes its performance through continuous feedback loops. The results of each action are recorded and used to adjust future decisions. This closed-loop system ensures the AI AGENT's adaptability and flexibility.

1.3 Market Status

1.3.1 Industry Status

AI AGENT is becoming the focus of the market, bringing transformation to multiple industries with its enormous potential as a consumer interface and autonomous economic actor. Just as the potential of L1 blockchain space was hard to estimate in the last cycle, AI AGENT is showing the same prospects in this cycle.

According to the latest report from a certain organization, the AI Agent market is expected to grow from $5.1 billion in 2024 to $47.1 billion by 2030, with a compound annual growth rate of 44.8%. This rapid growth reflects the penetration of AI Agents across various industries and the market demand driven by technological innovations.

The investment of large companies in open-source proxy frameworks has also significantly increased. The development activities of frameworks such as AutoGen, Phidata, and LangGraph from a certain company are becoming increasingly active, indicating that AI AGENT has greater market potential beyond the encryption field, and the TAM is also