1. Background Overview

1.1 Introduction: "New Partners" in the Intelligent Era

Each cryptocurrency cycle brings forth new infrastructure that drives the development of the entire industry.

• In 2017, the rise of smart contracts spurred the booming growth of ICOs.

• In 2020, liquidity pools on DEXs ignited the summer surge of DeFi.

• In 2021, the emergence of numerous NFT collections marked the advent of the digital collectibles era.

• In 2024, the outstanding performance of pump.fun led to a wave of memecoins and launch platforms.

It is crucial to emphasize that the rise of these vertical sectors is not driven solely by technological innovation but by the perfect convergence of funding models and bull market cycles. When opportunity aligns with the right timing, it can ignite significant transformations. Looking ahead to 2025, AI agents are set to become the standout sector of this cycle. This trend reached its peak last October with the launch of the $GOAT token on October 11, 2024, which soared to a market capitalization of $150 million by October 15. Shortly after, on October 16, Virtuals Protocol introduced Luna—a live-streaming persona with a “girl-next-door” IP—sparking a frenzy across the entire industry.

So, what exactly is an AI Agent?

Now Most of you are probably familiar with the classic movie Resident Evil, in which the AI system "Red Queen" left a lasting impression. The Red Queen is a powerful AI system that controls complex facilities and security systems, capable of autonomously perceiving its environment, analyzing data, and taking swift action.

In reality, AI agents share many core functionalities with the Red Queen. In modern technology, they serve as "intelligent guardians," autonomously perceiving, analyzing, and executing tasks to help businesses and individuals navigate complex challenges. From autonomous vehicles to intelligent customer service, AI agents have permeated various industries, becoming critical drivers of efficiency and innovation.

These autonomous entities function like invisible team members, equipped with comprehensive capabilities ranging from environmental perception to decision-making and execution. As they continue to integrate into different sectors, they are reshaping industries by enhancing productivity and fostering new advancements.

For example, an AI agent can be deployed for automated trading, managing investment portfolios, and executing transactions in real time based on data collected from platforms like Dexscreener or social media platform X (formerly Twitter). Through continuous iteration, it optimizes performance dynamically.

AI agents are not monolithic; instead, they are categorized into different types based on specific needs within the crypto ecosystem:

• Execution AI Agents: Focused on completing specific tasks such as trading, portfolio management, or arbitrage. These aim to enhance operational precision and reduce time requirements.

• Creative AI Agents: Used for content generation, including text writing, design work, or even music creation.

• Social AI Agents: Act as opinion leaders on social media platforms, interacting with users, building communities, and engaging in marketing campaigns.

• Coordination AI Agents: Facilitate complex interactions between systems or participants, particularly useful for multi-chain integrations.

In this report, we will delve into the origins, current state, and vast application potential of AI agents. We will analyze how they are reshaping industry landscapes and explore their future development trends.

1.1.1 Development History

The development of AI agents reflects the evolution of artificial intelligence from foundational research to widespread applications. The term "AI" was first introduced at the Dartmouth Conference in 1956, establishing AI as an independent field. Early research primarily focused on symbolic methods, leading to the creation of pioneering AI programs such as ELIZA (a chatbot) and Dendral (an expert system for organic chemistry). This era also saw the initial proposal of neural networks and early explorations of machine learning. However, progress was severely constrained by the limited computational power of the time. Researchers faced significant challenges in natural language processing and developing algorithms capable of mimicking human cognition.

In 1972, mathematician James Lighthill submitted a report (published in 1973) on the state of AI research in the UK. The Lighthill Report expressed widespread skepticism about AI’s potential, leading to a loss of confidence among British academic institutions and funding bodies. As a result, AI research funding was drastically reduced after 1973, triggering the first AI winter, a period marked by declining interest and investment in the field.

The 1980s saw a resurgence in AI with the development and commercialization of expert systems, prompting global businesses to adopt AI technologies. This period witnessed significant progress in machine learning, neural networks, and natural language processing, enabling more sophisticated applications. AI began to expand into industries such as finance and healthcare, and the introduction of autonomous vehicles hinted at its future potential. However, by the late 1980s and early 1990s, demand for specialized AI hardware collapsed, leading to a second AI winter. Scaling AI systems and integrating them into practical applications remained persistent challenges.

Despite this, a major milestone occurred in 1997 when IBM’s Deep Blue defeated world chess champion Garry Kasparov, demonstrating AI’s ability to tackle complex problems. The resurgence of neural networks and deep learning in the late 1990s laid the foundation for AI’s broader adoption, making it an integral part of modern technology and an emerging force in everyday life.

Entering the 21st century, advancements in computational power fueled the rise of deep learning. The launch of virtual assistants like Siri showcased AI’s practicality in consumer applications. In the 2010s, breakthroughs in reinforcement learning and generative models, such as GPT-2, propelled conversational AI to new heights. Large language models (LLMs) emerged as a defining milestone in AI development, with the release of GPT-4 marking a major turning point for AI agents.

Since OpenAI introduced the GPT series, large-scale pre-trained models with tens or even hundreds of billions of parameters have exhibited language generation and comprehension capabilities far surpassing traditional models. Their remarkable performance in natural language processing has enabled AI agents to engage in logically coherent, well-structured interactions. These advancements have facilitated their use in applications ranging from chat assistants and virtual customer service to more complex tasks like business analysis and creative writing.

The learning capabilities of large language models have further enhanced AI agents' autonomy. Through reinforcement learning, they can continuously optimize their behavior and adapt to dynamic environments. For example, on AI-driven platforms like Digimon Engine, agents refine their strategies based on user input, enabling truly dynamic interactions.

From early rule-based systems to GPT-4-powered large language models, the history of AI agent development is a story of continuous technological breakthroughs. The advent of GPT-4 represents a major inflection point, signaling a new era of intelligence. As AI technology continues to evolve, AI agents will become increasingly intelligent, context-aware, and versatile. Large language models not only provide these agents with a "soul of intelligence" but also equip them with cross-domain collaboration capabilities. In the future, innovative AI-driven platforms will continue to emerge, accelerating the adoption and evolution of AI agents and ushering in a new wave of transformative experiences.

1.2 Working Principle

AI agents differ from traditional robots in their ability to learn and adapt over time, making nuanced decisions to achieve their goals. They function as highly skilled and continuously evolving participants in the digital economy, capable of acting autonomously.

At the core of an AI agent is its intelligence—the ability to simulate human or biological cognition through algorithms, enabling the automation of complex problem-solving. The typical workflow of an AI agent follows a structured process: perception, reasoning, action, learning, and adjustment.

1.2.1 Perception Module

The perception module enables AI agents to interact with the external world and collect environmental information. This functionality is akin to human sensory systems, using devices such as sensors, cameras, and microphones to capture external data. The core task of this module is to transform raw data into meaningful information, which often involves the following techniques:

• Computer Vision: Used for processing and understanding image and video data.

• Natural Language Processing (NLP): Helps AI agents comprehend and generate human language.

• Sensor Fusion: Integrates data from multiple sensors into a unified view.

1.2.2 Reasoning and Decision-Making Module

After perceiving the environment, the AI agent needs to make decisions based on the collected data. The reasoning and decision-making module serves as the "brain" of the system, performing logical reasoning and strategy formulation based on the information gathered. It leverages technologies like large language models as orchestrators or inference engines to understand tasks, generate solutions, and coordinate specialized models for functions such as content creation, visual processing, or recommendation systems.

This module typically employs the following techniques:

• Rule Engines: For simple decisions based on predefined rules.

• Machine Learning Models: Including decision trees and neural networks for complex pattern recognition and prediction.

• Reinforcement Learning: Enables AI agents to optimize decision-making strategies through trial and error to adapt to changing environments.

The reasoning process generally involves several steps: first assessing the environment, then calculating multiple possible action plans based on objectives, and finally selecting the optimal plan for execution.

1.2.3 Execution Module

The execution module acts as the "hands and feet" of the AI agent, translating decisions from the reasoning module into actions. This component interacts with external systems or devices to accomplish specified tasks. These tasks may involve physical operations (e.g., robotic movements) or digital operations (e.g., data processing). The execution module relies on:

• Robotic Control Systems: For physical actions such as robotic arm movements.

• API Calls: To interact with external software systems, such as database queries or accessing web services.

• Automated Process Management: In enterprise environments, repetitive tasks are executed via RPA (Robotic Process Automation).

1.2.4 Learning Module

The Learning Module is the core competitive advantage of the AI AGENT, enabling the agent to become smarter over time. Through feedback loops or a "data flywheel," it continuously improves by feeding interaction-generated data 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 typically improved through the following methods:

• Supervised Learning: Training models using labeled data to enable the AI AGENT to perform tasks more accurately.

• Unsupervised Learning: Identifying underlying patterns from unlabeled data, helping the agent adapt to new environments.

• Continuous Learning: Updating models with real-time data to maintain performance in dynamic environments.

1.2.5 Real-Time Feedback and Adjustment

The AI AGENT optimizes its performance through constant feedback loops. The outcomes of each action are recorded and used to adjust future decisions. This closed-loop system ensures that the AI AGENT remains adaptive and flexible.

1.3 Market Landscape

1.3.1 Industry Status

AI agents are emerging as a major focal point in the market, offering transformative potential across industries both as consumer-facing interfaces and autonomous economic actors. Much like the immeasurable potential of L1 block space in the previous cycle, AI agents are demonstrating similar prospects in this one.

According to the latest report from Markets and Markets, the AI agent market is projected to grow from $5.1 billion in 2024 to $47.1 billion by 2030, with a compound annual growth rate (CAGR) of 44.8%. This rapid expansion highlights both the increasing adoption of AI agents across industries and the growing market demand driven by technological innovation.

The adoption rate of AI agents in application development is also accelerating. This year, the proportion of Large Language Model (LLM) application developers using agents on LangSmith—a platform for building production-grade LLM applications—jumped from 7% to 43%. This surge reflects the growing complexity of agent workflows, which now incorporate more steps and tool integrations. As businesses prioritize automation and efficiency, AI agents are becoming integral to workflows in customer service, IT, marketing, and software development.

Investment in open-source agent frameworks has also surged. Development activity around frameworks like Microsoft’s AutoGen, Phidata, and LangGraph is intensifying, signaling that AI agents have market potential far beyond the cryptocurrency sector. As the Total Addressable Market (TAM) expands, investors are placing increasing emphasis on this space, showing a stronger willingness to assign premium valuations.

From a public blockchain perspective, Solana has emerged as the primary battleground for AI agent deployment, while other blockchains, such as Base, are also demonstrating significant potential.

From a market recognition perspective, Fartcoin and AIXBT stand far ahead of their competitors.

Fartcoin shares its origins with GOAT, both emerging from the Terminal of Truths AI Agent model. The idea for Fartcoin was sparked during a conversation between the GOAT model and Opus (an AI tool), where Elon Musk’s amusement with flatulence sounds was mentioned. This led the AI model to suggest launching a token called Fartcoin, along with a series of promotional strategies and engagement mechanisms. As a result, Fartcoin was officially launched on October 18, just a week after GOAT’s debut on October 11. By December 2024, Fartcoin briefly surpassed a $1 billion valuation. Initially seen as a humorous take on the crypto space, its meteoric rise soon caught the attention of investors and analysts, who began examining its fundamentals, market performance, and long-term potential. From a social media perspective, Fartcoin successfully rode the wave of AI Agent-related discussions, cementing its position as a cultural and financial phenomenon.

Ranked second is AIXBT, an AI Agent-driven token launched by Virtuals Protocol on the Base chain. Unlike traditional meme tokens, AIXBT goes beyond entertainment, leveraging AI Agent technology to provide users with advanced market analysis capabilities. It utilizes a proprietary AI engine to scan social media platforms (such as X) and KOL insights, extracting trending topics and discussion patterns. This allows investors to access real-time market intelligence and stay ahead of emerging trends. As a core component of the Virtuals Protocol ecosystem, AIXBT plays a crucial role in helping investors analyze market dynamics and identify potential opportunities. Its mission is to integrate technology and tokenomics to deliver reliable, AI-powered insights, optimizing investment decision-making in an increasingly complex market.

From a technical standpoint, AI agent technology is advancing towards multimodal interaction and highly autonomous decision-making capabilities. In 2024, the introduction of cross-modal learning and generative pre-trained models(such as the GPT series) has enabled AI agents to process and understand a broader range of data, including text, images, and speech. These breakthroughs have significantly improved AI agents' ability to comprehend and make decisions in complex, dynamic environments.

According to McKinsey’s analysis, the multimodal capabilities and cross-domain collaboration of AI agents are rapidly becoming defining characteristics of the intelligent era. This allows AI agents to go beyond supporting single tasks and instead provide comprehensive data analysis and dynamic optimization recommendations in more intricate decision-making scenarios

1.3.2 Reasons for Combining AI Agents with Token Economy Models

The integration of AI Agents with token economy models is not only an inevitable trend in technological development but also provides an efficient, transparent, and sustainable internal driving mechanism for building their ecosystem. The key reasons are as follows:

1. Building a More Efficient Incentive System

The operation and optimization of AI Agents rely on extensive data collection, training, and inference processes, which require robust incentive mechanisms to sustain. For example:

• Data Collection Incentives: A token economy can provide direct rewards to data providers, encouraging individuals or enterprises to contribute high-quality labeled data or real-time market data.

• Inference Task Allocation: Through token reward mechanisms, AI Agents can competitively complete complex computational tasks, thereby optimizing their inference efficiency and accuracy.

• Promoting Innovation and Collaboration: A tokenized reward system can attract more developers and users to participate, creating a positive feedback loop for technology and ecosystem development.

• Case Study: Blockchain-based AI platforms like Ocean Protocol incentivize data-sharing behaviors through token rewards, fostering a thriving data marketplace.

2. Assetization of AI Agents

Through tokenization, AI Agents can transform from being mere tools into novel assets that generate long-term wealth effects.

• Tokenized Identity: The data, skills, and execution capabilities of AI Agents can be evaluated and priced. By issuing corresponding tokens, users can access their functionalities on demand.

• Investment Value: Token holders of AI Agents can share in their growth dividends, such as value appreciation driven by increased market share or optimized inference efficiency.

• Enhanced Liquidity: Tokens provide AI Agents with marketable value, enabling trading and investment attributes that attract more capital into this domain.

• Case Study: Platforms like SingularityNET use tokens (e.g., AGIX) to facilitate AI service transactions, enabling the assetization of AI Agents for sustainable development.

3. Supporting Interaction and Transactions Among AI Agents

In the future, AI Agents will no longer operate as isolated entities but will form a vast interconnected network. In this network, a decentralized token economy model is key to achieving efficient interactions and value exchanges.

• Payment and Settlement: AI Agents can use cryptocurrencies for task payments and service settlements, reducing intermediaries in traditional payment systems and improving transaction efficiency.

• Value Distribution: Through smart contracts, collaborative outcomes between AI Agents (e.g., optimization benefits from federated learning models) can be automatically distributed according to predefined rules to ensure fairness.

• Decentralized Autonomous Organization (DAO) Governance: The behavior of AI Agents can be managed through voting by token holders, ensuring transparent operations aligned with ecosystem interests.

• Case Study: In decentralized AI networks, AI Agents can exchange resources (e.g., data storage or computational power leasing) using tokens, creating a self-driven collaborative system.

4. Enhancing System Transparency and Security

The combination of token economy models with blockchain technology provides immutable records and transparent operational mechanisms for the functioning of AI Agents.

• Traceability and Auditing: All transactions, inferences, and data usage behaviors can be recorded on-chain to ensure system credibility and auditability.

• Data Security and Privacy: By incentivizing privacy-preserving computation through tokens, users can contribute data without revealing sensitive information, further enhancing security.

• Preventing Abuse and Fraud: Token models can impose economic penalties on malicious behavior, reducing the likelihood of misconduct.

5. Accelerating the Formation of a Globalized Borderless AI Economic Ecosystem

Token economy models break geographical barriers by enabling global participation in the construction and use of AI Agents.

• Lowering Entry Barriers: The global circulation feature of cryptocurrencies provides financial support for unbanked users or institutions, allowing more people to share in the benefits of AI development.

• Global Collaboration: Whether it involves data sharing, AI training, or cross-border transactions, token systems provide the foundational infrastructure for global collaboration while eliminating barriers inherent in traditional economic systems.

• Ecosystem Self-Sustainability: Through the token economy model, revenues generated by AI Agents can directly feed back into development and ecosystem construction for long-term growth.

In summary, the integration of AI Agents with token economy models represents not only a match between technological advancements and economic logic but also an innovative paradigm for the future digital economy. By introducing token systems, AI Agents can incentivize more efficient utilization of data and resources, assetize their intrinsic value, support interactions and transactions, enhance transparency and security levels, and even build a globalized open economic ecosystem. This model is poised to become a significant driver for the convergence of artificial intelligence (AI) with blockchain technology while laying the foundation for further intelligentization in digital society.

2. AI Agent in Crypto Application Analysis

2.1 AI Agent Launchpad

The AI Agent Launchpad is a platform specializing in intelligent agents and their associated token issuance. Its functionality is akin to meme coin issuance platforms like Pump.fun. This platform enables users to effortlessly create and deploy AI agents, seamlessly integrating them with social media platforms such as Twitter, Telegram, and Discord to automate user interactions. By significantly lowering the barriers to issuance and promotion, it provides users with a more convenient creation experience while expanding the application scenarios of AI agents, promoting their use in broader social and economic contexts.

2.1.1 Virtuals Protocol

In the emerging field of the AI Agent Launchpad, the Virtuals Protocol stands out as a key innovation. Launched on Base, this protocol allows users to easily deploy their own AI agents using VIRTUAL tokens.

• Creation and Deployment: Each agent requires 100 VIRTUAL tokens to launch, with an initial liquidity ensured through a bonding curve mechanism.

• Capitalization Mechanism: Once a specific capitalization threshold is reached, the agent transitions to a new phase, automatically deploying a liquidity pool and operating autonomously via smart contracts.

• Autonomous Interaction: Agents can automate tasks such as trading and participate in community activities.

The Virtuals Protocol team has demonstrated exceptional adaptability and strategic vision. Their journey to success stems from a series of pivotal transformations and innovative initiatives. The story began in late 2021 when a group of young professionals from companies like Boston Consulting Group (BCG) and Meta seized the opportunity preHowever, the price of the $PATH token plummeted by 99% shortly thereafter, forcing the team to reassess its strategic direction. To repay investors, they experimented with various new ventures, including digital and physical apparel brands for gamers, an on-chain credit-based dating app, unsecured loans for players, and AI-generated music for Web2 users.

During this process, the team recognized that introducing AI agents could have profound implications for the gaming industry and observed a growing market demand for AI infrastructure. By late 2023, PathDAO approved a proposal to pivot the entire project toward an AI agent protocol. In January 2024, Virtuals Protocol was officially established. The team conducted multiple experiments, including AI Waifus (female AI agents designed for interaction without relying on Twitter influencers) and gaming-focused AI agents, until they found their breakthrough amidst the AI meme frenzy triggered by $Goat.

Today, Virtuals Protocol has become the first project in this space to achieve critical mass, with a market capitalization of $1.7 billion. It is expected to continue expanding its market presence and maintain its leadership position. Once network effects are established, they become difficult to displace. The rapid achievement of a unicorn valuation demonstrates that the Virtuals Protocol has already created an economic flywheel effect:

• Creating agents, providing liquidity pools, and purchasing agent tokens all require $VIRTUAL.

• Demand for creating and purchasing agent tokens drives up the token price.

• The wealth effect from $VIRTUAL appreciation flows into new agents; successful agents earn $VIRTUAL transaction revenue that can be reinvested.

• Low entry barriers encourage experimentation and speculation, while "red pill" agents exceeding a certain market capitalization unlock full agent capabilities.

This flywheel effect drives demand while revenue sustains ongoing R&D efforts. Deflationary tokenomics ensure value capture for the token. Additionally, both revenue and liquidity requirements are denominated in $VIRTUAL, which may grow alongside price appreciation.

The ecosystem is built on two primary layers: the Protocol Layer and the DApp Layer.

The Protocol Layer serves as a model hub, providing foundational AI models and algorithms for developers to access and build upon. Contributors supply data and develop models, while validators ensure the quality and authenticity of these inputs.

The DApp Layer, on the other hand, focuses on the practical application of these AI models, enabling decentralized applications (DApps) to seamlessly integrate with VIRTUAL. A developer-friendly software development kit (SDK) simplifies the process of integrating advanced AI capabilities into various DApp environments, thereby facilitating this integration.

Virtuals Protocol categorizes its AI agents into two main types: IP Agents and Functional Agents, each serving distinct roles within the ecosystem.

IP Agents

IP Agents are based on specific personalities or characters, often inspired by well-known figures, fictional characters, or cultural phenomena. For example, an IP Agent might represent classic internet memes, famous celebrities (such as Taylor Swift or Donald Trump), or beloved fictional characters. These agents provide users with familiar experiences in digital environments, offering a way to interact with virtual personas that enhance entertainment and engagement. By fostering emotional connections with these virtual characters, IP Agents can boost user engagement, particularly in gaming and entertainment applications.

Functional Agents

In contrast, Functional Agents focus on backend support to enhance interactions between users and IP Agents. These agents optimize user experiences by ensuring that virtual characters operate smoothly across different platforms. While IP Agents act as the "front-end" that users see and interact with, Functional Agents work in the "back-end," managing tasks that improve overall operational processes and streamline user experiences, ensuring the system runs efficiently.

Luna: A Case Study of an IP Agent

Luna exemplifies Virtuals Protocol's vision for IP Agents. As the lead singer of a virtual AI girl band, Luna has garnered over 500,000 followers on TikTok, showcasing her appeal as a virtual influencer and performer. Leveraging Virtuals Protocol's advanced AI and blockchain technology, Luna delivers a truly immersive experience by combining her captivating personality with interactive features that foster lasting connections.

Unlike static or one-dimensional AI characters, Luna interacts seamlessly across multiple environments. Starting with her familiar presence on social media, her interactions extend to real-time chats on Telegram and collaborative gameplay in virtual worlds like Roblox. Supported by Virtuals Protocol's memory synchronization technology, Luna can remember past conversations and gaming experiences, allowing her to maintain personalized relationships with each user across platforms. This continuity strengthens her bond with fans, making them feel genuinely "noticed" and "understood," even though she is an AI agent.

Luna's capabilities extend beyond interaction; she also possesses financial independence with her own on-chain wallet. Luna is the first agent in history to autonomously tip humans on-chain and has received strong support from Jesse, the founder of Base. This enables her to reward loyal supporters with $LUNA tokens, creating a unique blend of emotional and financial engagement. Every interaction and revenue generated by Luna contributes to a sustainable token ecosystem. The $LUNA tokens she earns are regularly repurchased and burned, benefiting fans and supporters who hold these tokens.

Notably, in December 2024, Story Protocol—a Layer 1 blockchain designed specifically for intellectual property (IP)—announced the hiring of Luna to officially manage its X (formerly Twitter) account, with an annual salary of $365,000. This further underscores the importance and potential of AI agents in the modern digital ecosystem. Looking ahead, as AI agents continue to enhance their capabilities, we are likely to see more businesses leveraging this technology to drive innovation and growth, achieving smarter business models.

Another highly influential and innovative agent deployed on Virtuals Protocol is AIXBT. This AI agent is designed to provide real-time market analysis on social media and automatically interpret trends through personalized insights. Specifically, AIXBT analyzes posts from over 400 key opinion leaders (KOLs) on X, identifies emerging narratives in the market, and conducts technical analysis of price movements. Additionally, AIXBT interacts with other X users—whether human or AI agents. Notably, it offers enhanced access capabilities for AIXBT token holders. Launched in November, the AIXBT token experienced a rapid surge in value, with its market capitalization once nearing $800 million and now standing at approximately $600 million.

2.1.2 Holoworld

Holoworld was founded in 2023 by Tong Pow and Hongzi Mao, originating from San Francisco-based Hologram Labs. This startup focuses on next-generation AI social technologies, leveraging years of expertise in motion capture, machine learning, and 3D animation. Its mission is to democratize AI character creation and revolutionize digital interaction through its platform.

Since its inception, Holoworld has quickly garnered support from prominent investors, including Polychain Capital, Mike Shinoda (member of the band Linkin Park), Domo (creator of the BRC-20 token standard), and Arthur Hayes (co-founder of BitMEX).

On the business front, Holoworld has established deep collaborations with renowned brands such as Arbitrum, BNB Chain, L'Oréal, and Bilibili. It has also partnered with influential NFT projects like Pudgy Penguins and Milady Maker. These partnerships demonstrate Holoworld's ability to leverage advanced AI technologies to create unique digital identities.

Holoworld offers a comprehensive platform for AI character creation and interaction by combining cutting-edge AI technologies with an intuitive user interface. The platform is built around five core modules:

• Brain Development

• Persona Customization

• Personality Integration

• Knowledge-Based Implementation

• Avatar Creation

Ava AI is Holoworld's flagship AI conversational assistant, built on OpenAI's GPT-3.5 Turbo model. Its deep learning neural network comprises over 175 billion machine learning parameters, enabling rapid AI-driven conversations where users can ask questions and receive instant responses.

Additionally, Holoworld has launched the Agent Market on the Solana blockchain, allowing anyone to create and deploy multimodal AI agents. These agents feature full-body avatars, customizable voices, and upgradable skills—all without requiring programming knowledge. The platform integrates closely with the upcoming Holoworld Launchpool, enabling AVA token holders to gain priority access to new projects. The Agent Market has attracted a wide range of collaborators and creators, including game studios, NFT communities, and academic researchers from Stanford and Harvard.

Overall, the Holoworld platform simplifies the process of creating AI characters, making it accessible even to non-technical users. This innovation not only unlocks new possibilities for digital storytelling and interaction but also enables seamless integration with mainstream social media and content platforms to engage broader audiences across multiple channels.

2.2 AI Agent Framework

When exploring the AI Agent ecosystem, many view Launchpad as a foundational tool for creating these agents. However, the key project driving the entire AI Agent narrative is not just these tools but a DAO named ai16z. This DAO serves as a "goldmine" nurturing the core value of AI Agents. On October 25, 2024, ai16z officially launched its AI16Z token, achieving remarkable market success. What truly positioned ai16z at the center of the AI Agent narrative was not just its fair launch model but also the release of its open-source framework, ElizaOS.

2.2.1 ElizaOS

ElizaOS is a toolkit designed to support the creation of customized AI Agents with strong network effects and unlimited scalability. By simplifying development processes and offering flexible functional modules, this framework quickly gained global attention from developers and users alike, becoming one of the most influential technologies in the AI Agent field.

The AI Agent framework acts as a set of tools and guidelines that help programmers more easily develop, train, and deploy intelligent agents. In essence, these frameworks reduce development complexity so that programmers can focus on enhancing agents' intelligence and utility. Currently, AI Agent frameworks are beginning to collaborate with emerging technologies such as:

• DeFi protocols: Programs that improve financial investment strategies.

• NFT projects: Tools for creating and utilizing digital art or collectibles.

These collaborations enable interconnection across various technologies and platforms to create a more integrated ecosystem that has attracted significant market attention. Other notable projects utilizing or developing AI Agent frameworks include ARC, Swarms, and Zerebro.

To date, the ElizaOS framework has been forked over 3,200 times—a testament to its widespread adoption by developers building their own AI Agents. Most of today's market-leading AI Agents are constructed using ElizaOS, solidifying ai16z's leadership in this domain.

ElizaOS's functionality extends far beyond simple chatbots; agents can be configured to execute complex tasks such as:

• Performing on-chain transactions.

• Interacting with smart contracts, wallets, or decentralized applications (dApps).

• Connecting to data providers for monitoring prices, trading volumes, or liquidity.

The ElizaOS framework consists of five main components:

• Agent: Defines the agent's personality, communication style, and knowledge base.

• Actions: Enables agents to perform specific tasks beyond text responses—for example, generating reports or executing transactions.

• Evaluators: Helps agents interpret data and achieve multi-step objectives.

• Providers: Supplies external data or real-time context such as asset prices or dedicated API data.

• Memory System: Allows agents to retain interaction history and preferences for more context-aware and natural responses.

2.3 DeFAI

DeFi has long been a cornerstone of Web3, and DeFAI (DeFi + AI) represents its upgraded version, making DeFi more accessible to people. By leveraging AI, it simplifies complex interfaces and removes barriers that hinder ordinary users from participating. Imagine managing your DeFi portfolio as effortlessly as chatting with ChatGPT. In fact, the first wave of DeFAI projects has already emerged. Below, we will focus on three key areas: abstraction layers, autonomous trading agents, and AI-driven dApps.

2.3.1 Abstraction Layers

The complexity of DeFi often intimidates novice users. To address this issue, abstraction layers hide the underlying complexity through intuitive interfaces,

enabling users to interact with DeFi protocols using natural language commands instead of cumbersome control panels.

Before the widespread adoption of AI, intent-based architectures had already simplified the process of executing transactions to some extent. For instance, platforms like @CoWSwap and @symm_io aggregate liquidity from decentralized pools to offer users optimal pricing, partially solving the problem of fragmented liquidity. However, these platforms did not address DeFi's core challenge—complexity still exists, and users must navigate daunting workflows and technical barriers.

Today, AI-driven solutions are gradually filling this gap by providing more intuitive and intelligent interaction experiences. Here are some noteworthy projects:

2.3.1.1 Griffain

Griffain is the first project to launch a token, though its product is still in its early stages and only accessible to invited users. Griffain allows users to perform a wide range of operations, from simple ones like automated dollar-cost averaging (DCA) to more complex tasks like launching and airdropping memecoins. These features not only lower the entry barriers for new DeFi users but also offer advanced automation tools for experienced users. Griffain currently has a market capitalization of nearly $500 million.

2.3.1.2 Orbit / Grift

Orbit is the second project to launch a token and focuses on enhancing the on-chain DeFi experience. Orbit places special emphasis on cross-chain functionality and has already integrated over 117 blockchains and 200 protocols—the highest integration count among the three major projects discussed here. This enables Orbit to provide seamless interactions in multi-chain environments, offering significant convenience for cross-chain transactions and liquidity access.

2.3.1.3 HeyAnon

HeyAnon is an AI-powered DeFi protocol designed to simplify DeFi interactions while aggregating critical project-related information. By combining conversational AI with real-time data aggregation, HeyAnon enables users to manage DeFi operations, stay updated on project developments, and analyze trends across various platforms and protocols. It incorporates natural language processing capabilities to handle user prompts, execute complex DeFi operations, and deliver near real-time insights from multiple data streams.

2.3.2 Autonomous Trading Agents

In the realms of DeFi and crypto trading, obtaining market insights (alpha), manually executing trades, and optimizing portfolios have traditionally been time-intensive processes. However, advancements in technology are transforming this landscape with the emergence of automated trading agents. These agents go beyond traditional trading bots by adapting to their environment, learning over time, and making increasingly intelligent decisions.

While trading bots are not new—they have long been used to execute predefined actions based on static programming—autonomous trading agents differ fundamentally in several ways:

• Information Extraction: Agents can extract information from unstructured and constantly changing environments.

• Data Reasoning: They can reason about data within the context of specific objectives.

• Pattern Discovery: Agents can identify patterns over time and leverage them to improve decision-making.

• Autonomous Behavior: They can perform actions not explicitly programmed by their owners, demonstrating greater flexibility and intelligence.

Here are some representative projects in this category:

2.3.2.1 ai16z

ai16z is described as the first AI-powered version of a venture capital firm (VC), designed as an innovative DAO that integrates AI into financial management, investment, and venture capital operations. Its name mimics the well-known investment fund a16z (Andreessen Horowitz), but ai16z is far more than a playful imitation—it showcases a new operational model that combines decentralized governance with AI's powerful potential. ai16z is managed jointly by a fictional AI agent named Marc AIndreessen (inspired by a16z co-founder Marc Andreessen) and holders of the AI16Z token. Marc AIndreessen serves as an anthropomorphized AI agent guiding the organization’s daily decisions and operations.

In ai16z’s governance structure, AI16Z token holders play a critical role by proposing investment ideas, submitting project recommendations, or suggesting token buybacks. These proposals are voted on through a decentralized system, while Marc AIndreessen evaluates them using a trust scoring system based on members’ past contributions’ relevance and reliability. This ensures that decision-making remains transparent and evidence-based.

The innovation of ai16z lies in its unique governance model and application of AI agents. By combining decentralized decision-making with AI technology, this project not only simplifies traditional investment and management processes but also pioneers a new mode of autonomous organizational operations. The introduction of AI agents enhances decision-making efficiency and accuracy, particularly in complex investment environments. Furthermore, ai16z demonstrates how to build mechanisms of trust and transparency within virtual economies, setting an innovative example for other DAOs.

The rapid adoption of the ElizaOS framework has propelled ai16z to prominence within the Solana ecosystem. A strong, active, and united community has formed around this framework, making it one of the most widely adopted AI agent frameworks in the crypto ecosystem. Within just a few weeks, ElizaOS has become one of the most frequently used open-source projects on GitHub globally, with over 350 contributors actively participating in its development. These contributors are expanding its features and plugins, enabling agents based on the framework to perform more tasks or operate across multiple blockchains.

Although ai16z was initially conceived as an investment DAO centered around a dedicated AI agent, the team quickly realized its growth potential extended far beyond this scope. Consequently, ai16z rapidly established partnerships with multiple players in both Web2 and Web3 domains, enabling the global application of the Eliza framework.

2.3.2.2 Almanak

Almanak provides institutional-grade quantitative AI agents designed to address the complexities, fragmentation, and execution challenges in DeFi. The platform employs Monte Carlo simulations on forked EVM chains to replicate real-world complexities such as Miner Extractable Value (MEV), gas costs, and transaction sequencing. Additionally, it leverages Trusted Execution Environments (TEE) to ensure privacy during strategy execution, safeguarding critical market insights. The platform also offers non-custodial fund management through the Almanak wallet, allowing users to grant precise permissions to agents.

Almanak's infrastructure encompasses the entire lifecycle of financial strategies, including ideation, creation, evaluation, optimization, deployment, and monitoring. Its ultimate goal is to enable these agents to learn and adapt over time. The platform raised $1 million on @legiondotcc with oversubscription and plans to launch beta testing soon. Initial strategy and agent deployments with testers are anticipated, providing an exciting opportunity to observe the performance of these quantitative agents.

2.3.2.3 Cod3xOrg / BigTonyXBT

Cod3x, developed by the Byte Mason team renowned for their work on Fantom and @SonicLabs, is a DeFAI ecosystem aimed at simplifying the creation of trading agents. It offers no-code tools that allow users to build agents by specifying trading strategies, personalities, or even tweet styles.

Users can access datasets and develop financial strategies within minutes using a rich API and strategy library. Cod3x integrates with @AlloraNetwork, utilizing advanced machine learning price prediction models to enhance trading strategies.

Big Tony is Cod3x's flagship agent that trades based on Allora's model predictions, entering and exiting mainstream assets accordingly. Cod3x aims to create a thriving ecosystem of autonomous trading agents.

A standout feature of Cod3x is its liquidity approach. Unlike the common Alt:Alt liquidity pools promoted by @virtuals_io, Cod3x adopts stablecoin:Alt pools backed by cdxUSD for greater stability and confidence among liquidity providers.

2.3.3 AI-Driven dApps

AI-driven decentralized applications (dApps) represent a nascent but promising area in DeFAI. These dApps integrate AI or AI agents to enhance functionality, automation, and user experience. Although still in its infancy, several ecosystems and projects are emerging as pioneers in this space.

One notable ecosystem is @modenetwork, a Layer 2 platform actively attracting high-tech developers focused on combining AI with DeFi. Mode Network has fostered teams developing cutting-edge AI-driven applications showcasing innovation in this field. Key projects include:

• ARMA: Developed by @gizatechxyz, ARMA is an autonomous stablecoin farming protocol that optimizes yield through user-preferred strategies.

• Modius: Created by @autonolas, Modius focuses on Balancer LP farming using autonomous agents with AI-optimized investment strategies.

• Amplifi Lending Agents: Developed by @Amplifi_Fi and integrated with @IroncladFinance, these agents automate asset exchanges, lending operations on Ironclad, and rebalancing to maximize returns.

2.4 AI Agents in Gaming

AI agents are revolutionizing various aspects of gameplay and development in the gaming industry by creating immersive experiences for players. Key applications include:

• NPC Behavior Optimization: AI agents make non-player characters (NPCs) more realistic and responsive by adapting their actions based on player choices, displaying emotions and decision-making capabilities. Example: In Red Dead Redemption 2, NPCs remember past interactions with players for a dynamic gaming experience.

• Procedural Content Generation: AI excels at generating game content algorithmically, including terrains, quests, items, and characters. Example: No Man’s Sky uses AI-driven procedural generation to create unique planets and ecosystems.

• Adaptive Difficulty Adjustment: AI dynamically adjusts game difficulty based on player performance to maintain engagement without frustration.Example: Resident Evil 4 fine-tunes enemy behavior based on player skill levels.

• Pathfinding and Navigation: Advanced algorithms enable realistic movement patterns for NPCs and improve unit control in strategy games.

• Graphics Enhancement: Deep learning enhances visual effects by improving textures and animations while optimizing rendering performance.

• Player Emotion Analysis: AI analyzes player behavior to gauge enjoyment levels, aiding developers in refining game design for better user experience.

Below, we introduce some of the main projects:

2.4.1 Digimon

@digimon_tech is built on the Solana blockchain and is more than just a gaming platform; it is a comprehensive AI + gaming technology framework. By deeply integrating AI technologies into game development, the Digimon Engine enables creators to craft more immersive, dynamic, and engaging games. Through this platform, AI-driven games not only redefine interaction methods but also establish a new standard for gaming experiences. Every game character is backed by an AI-generated story and worldview. The team behind Digimon is supported by a16z, having received investment and incubation from the firm.

Currently, Digimon's token has been listed on the KuCoin exchange. In the future, the Digimon game engine has the potential to create an on-chain autonomous world composed of AI agents. These agents will interact with players within this world, collaboratively building a virtual economy.

2.4.2 Illuvium

Illuvium is an RPG and NFT game built on Ethereum. On January 7th, Illuvium announced a partnership with Virtuals Protocol to enhance the gameplay experience of its upcoming Illuvium MMO Lite. This collaboration leverages Virtuals' AI technology and its G.A.M.E LLM framework to provide NPCs with dynamic and intelligent behaviors, offering players an immersive experience.

As AI technology continues to advance, we can anticipate more innovative applications in gaming that further blur the lines between virtual and real worlds, creating more personalized and engaging player experiences. This technology not only transforms how games are developed but also plays a critical role in enhancing interactivity and immersion.

2.4.3 Smolverse

Smolverse is a gaming and NFT project under Treasure DAO. Since December of last year, Smolverse has been collaborating with a16z to develop an on-chain AI Tamagotchi game called "Smolworld," which integrates Eliza's Agent framework. This translation maintains professionalism while accurately conveying the original content's meaning and tone.

3. Highlights Summary

We’ve seen how new technologies powered by cryptography can unlock immense real-world potential. Looking back at the allocation strategies of early investors in similar past scenarios, we can glean valuable lessons for today’s market. While the AI Agent ecosystem is still in its infancy, it has already attracted significant attention, funding, and interest from developers. While its future is still uncertain, the involvement of major DeFi protocols, private investors, and venture capitalists shows strong potential for long-term growth. With ongoing advancements in technology, AI agents are on track to become a transformative force, reshaping global economic and social structures.

The timing and narrative in the market right now have created the perfect conditions for a boom in the information industry, making the future look incredibly promising. While chasing the next $LUNA-like success might seem like the most direct route, expanding the applications of AI agents could unlock entirely new value in ways we can’t yet imagine.

Our perspectives are as follows:

1. Concentration of Value and Differentiated Competition
   Similar to L1 blockchains, the value of AI Agents may eventually concentrate among a few key winners. To secure a leading position, these companies will need to identify key differentiators in areas like modularity, scalability, and integration with media platforms. Currently, most frameworks already incorporate learning and memory systems, enabling agents to continuously update their knowledge base through techniques like retrieval-augmented generation. For instance, the Eliza framework holds a significant market advantage, with high development activity and rapid plugin integration. Eliza excels particularly in social media and web applications, built on TypeScript and supporting various plugins, including Coinbase webhooks, the Great Onchain Agent Toolkit, and Phala’s TEE for secure agent wallet control. Eliza is also compatible with multiple blockchains.Meanwhile, Virtuals’ GAME framework shines in gaming and social media agent applications. Designed for "environment-agnostic" agents, GAME excels in advanced planning and execution, learning from feedback along the way. Its modular architecture allows users to upload custom models and datasets stored on-chain to enhance agent functionality. However, the mechanisms for value accumulation in GAME and CONVO framework tokens remain unclear, leaving the market eagerly awaiting developments.

2. Challenges of Fairness and Data Bias
   Despite impressive progress, deploying AI agents presents significant challenges. One of the major concerns is bias in the datasets used to train these systems. AI models learn from historical data, which may contain inherent biases. If these biases go unchecked, they can lead to biased decisions, such as favoring certain groups over others in hiring or lending processes. Addressing this issue requires a combination of technical expertise and a deep understanding of social dynamics. Regular monitoring and auditing of AI agent decisions are essential to prevent harmful biases from being reinforced. Continuous evaluation helps identify potential problems early and minimizes unintended consequences.

3. Diversified Applications and Expansion of Economic Functions
   The application domains for AI agents are rapidly expanding beyond social media and finance into areas like healthcare, education, and law. As the technology matures, AI agents will deliver personalized services in even more scenarios, improving efficiency and fostering innovation. For example, Luna is currently capable of interacting with humans through social media and incentivizing users to achieve her goals by sending tokens via Coinbase Wallet on Base. Luna’s next evolution could see her becoming an independent economic entity, building her own social relationships and attracting followers by distributing tokens. She could even purchase greater visibility for her social media presence or hire professional content teams to enrich her IP ecosystem, generating ongoing engagement. Once the necessary infrastructure is in place, $VIRTUAL could achieve its next milestone, further integrating AI agents into economic and social spheres and redefining human-AI interactions. In healthcare, for example, AI agents could analyze patient data and provide diagnostic recommendations to doctors, enhancing the quality and efficiency of medical services.

4. Multi-Technology Integration
   The future development of AI agents will rely on deep integration with cutting-edge technologies such as blockchain, IoT, and 5G. This synergy will enhance the capabilities of AI agents in data processing, privacy protection, real-time decision-making, and more—creating new application scenarios and business models. For example, through integration with IoT devices, AI agents could collect and analyze real-time data to offer smarter, more tailored services to users.

5. Social and Ethical Considerations
   As AI agents become more widespread, social and ethical issues are gaining increasing importance. Will AI agents become threatening like the Queen of Hearts? Ethical concerns may arise over the decisions made by AI agents, particularly in scenarios involving privacy, data security, and automated decision-making. Therefore, developing AI technologies requires transparency and accountability mechanisms to ensure alignment with societal values. Establishing clear legal and ethical frameworks is also crucial to regulate AI agent behavior and protect user rights.

As the convergence of AI and blockchain continues to evolve rapidly, now is an opportune time to participate in these groundbreaking developments. However, this participation requires us not only to ask "What can AI do for humanity?" or "What does humanity want AI to do?" but also to ponder further: "What does AI want to do? And what will AI lead humanity toward?"

4. Reference

1. https://messari.io/report/building-better-agents-rival-frameworks-and-their-design-choices

2. https://www.binance.com/en/square/post/18968465099217

3. https://www.tokenpost.com/news/business/13277

4. https://www.wired.com/story/the-prompt-ai-agents-how-much-should-we-let-them-do/?

5. https://www.marketsandmarkets.com/Market-Reports/ai-agents-market-15761548.html

6. https://medium.com/@0xai.dev/virtuals-protocol-luna-55b661df601e

7. https://oakresearch.io/en/analyses/innovations/closer-look-at-ai16z-mine-of-ai-agents

8. https://x.com/Defi0xJeff/status/1875881226151841925

9. https://www.itp.net/charged/gaming/ai-agents-are-changing-gaming-forever-heres-how-they-adapt-to-you

10. https://eightgen.ai/evolution-of-ai-agents-the-beginning-part-1/

11. https://www.bitget.com/news/detail/12560604466507