Sharing a great article about AI agents, originally posted by author on X: @shufen46250836.

The original link is: https://x.com/shufen46250836/status/1874449701870928026.

Many people say that this round of the cryptocurrency bull market lacks innovative narratives, but AI is actually the most innovative and sustainable core narrative. By December 2024, the project in the cryptocurrency market (off-chain) that achieved the highest return came from the AI field—Virtuals, with a return of up to 23,079%.

"The next stop for large models," "completely changing the way humans live," and "opening up a new industrial revolution era"… people spare no effort in describing the importance of AI Agents. However, both retail investors and institutions are not fully prepared for the current momentum and future trends of AI Agents. Many people around me who did not pay attention previously felt the need to understand it only after it exploded. But now, with a flood of information in the market, it can be overwhelming and confusing. Today, I'll thoroughly sort out AI Agents; this research report will help you quickly get up to speed, serving as an entry guide to AI Agents (cryptocurrency edition)!

This article will be divided into three parts: First, an overview of the current development status of AI Agents; second, selecting and analyzing potential AI Agent projects; third, exploring the application expectations of AI Agents in the Web3 field.

I. Basics of AI Agents

AI Agents first appeared before the public in March 2023, when a project called AutoGPT was launched. This project utilizes large language models to automatically decompose a large task into smaller tasks and complete them using tools.

AutoGPT shocked the world upon its release because it was the first time language processing, content creation, logical reasoning, and perception action technology were extended to actual applications. Soon after, OpenAI launched a series of GPT models, and many tech companies began laying out their applications, platforms, development, and operations to enhance the barriers under the next wave of ecology.

So, what exactly is an AI Agent? How does it work? The term "agent" refers to a representative. In simple terms, an AI Agent is a representative powered by AI technology. Unlike traditional software that passively executes instructions, its workflow is: perception module (input acquisition) → LLM (understanding, reasoning, and planning) → tool invocation (task execution) → feedback and optimization (validation and adjustment).

OpenAI defines "AI Agent" as a system driven by an LLM brain that possesses the capabilities of autonomous understanding, perception, planning, memory, and tool usage, capable of automating complex tasks. Unlike traditional AI, AI Agents have the ability to gradually achieve given goals through independent thinking and tool invocation.

Let me give you an example to better understand: if you have a cold and a fever, traditional software would only tell you to go see a doctor and take precautions. An AI Agent, however, could monitor your temperature and other health indicators, combine online information, help you find the right medicine, request payment, and arrange delivery to your home, plus write a sick note for you for the next day. This is the magic of AI Agents.

II. Analysis of AI Agent Projects

According to the latest data from Cookie.fun, as of December 30, the overall market capitalization of AI Agents has reached $11.68 billion, with a nearly 39.1% increase in the past seven days. This growth trend indicates a rapid rise of the AI Agent ecology within the crypto market.

In this AI Agent boom, ai16z and Virtuals Protocol are undoubtedly the two most prominent representative projects. Specifically, the market capitalization of the Virtuals ecosystem has reached $5.01 billion, and ai16z is $1.63 billion, together accounting for 56.8% of the AI Agent market share, contributing more than half of the total.

In terms of on-chain distribution, Base and Solana are the two main battlegrounds for AI Agents. The market capitalization of AI Agents on Base is about $5.76 billion, while on Solana it is $5.47 billion, collectively contributing to 96.1% of the overall market. Other on-chain projects only cumulatively amount to $920 million.

This also reflects that although the AI Agent ecology is rapidly rising in the crypto market and attracting significant attention and capital, the market structure remains singular, primarily reliant on a few leading projects' momentum, and the AI Agent ecology is still in its early embryonic stage.

Next, I will analyze the current hot ecological projects of AI Agents based on the market situation, with three main criteria: 1. Long-term value of the project 2. Authentic demand within the market 3. Cash flow income status. If, after reading, you wonder why a particular project isn't included, please revisit these three points as a basis for your review. The following project opinions are for reference only and not considered financial advice.

1. Virtuals

Virtuals was actually launched last year. The Virtuals protocol is primarily aimed at establishing co-management for AI agents in the gaming and entertainment sectors. AI agents can be tokenized through blockchain and achieve co-management. Functions of the agents include autonomous planning, goal achievement, environmental interaction, and control of on-chain wallets.

The biggest innovation and distinction of Virtuals from other Web3 AI agent protocols is simplifying the complexity of AI agents, providing a plug-and-play solution similar to Shopify. This allows non-AI professionals to easily deploy AI agents in gaming and consumer applications, enabling them to earn protocol income through tokenization and decentralized co-management.

In addition, Virtuals has generated AI virtual idols—the AI-dol band—using its AI technology, which has garnered hundreds of thousands of followers on TikTok, making it quite interesting.

The total supply of Virtuals tokens is 1 billion, all of which have been released. The distribution of tokens is as follows: 60% held by the public, 5% for liquidity pools, and 35% for the ecosystem treasury, with a maximum annual release of 10% over three years. Currently, the ecosystem fund holds over 30% of the tokens.

From a long-term value perspective, it addresses the pain point of non-AI professionals being unable to participate in the AI trend, has a solid user base, and offers comparatively open and transparent token economics. Its marketing has also been notably effective. In terms of market capitalization, as the leader in the AI Agent ecology, this upward trend has seen nearly no adjustments, indicating a high probability of a significant adjustment ahead. Therefore, the short-term risk for Virtuals appears considerable.

2. ai16z

ai16z shares the name with the well-known venture capital a16z, yet this project has no relationship with A16Z and has not received investment from them. The only link is that it has gained the attention of a16z founder, Marc Andreessen.

The total token supply is 1.09 billion. The project operates via a DAO structure. According to its key influencer Shaw, ai16z will introduce several games based on the Eliza framework in the future and focus more on developing a practical Al Agent investment tool—DeFi Al Agent. The founder states that ai16z's goal is not to create an AI robot that mimics a16z but rather to outperform it in its area of expertise.

To sum up, the AI16Z project focuses on investment modeling for AI agents, appearing similar to previous AI bots and telegram bots. Can AI really make money through investment? That's a question mark. The core technology Eliza OS is based on OpenAI's capabilities and has only undergone some basic development. If OpenAI later launches its own AI agents, how will Shaw respond?

In summary, I think the AI16Z project merely capitalizes on the AI16Z name. Its long-term value lies in the DeFi Al Agent; however, this demand is somewhat false and circles back to the previous logic of AI bots, relying on OpenAI's open-source database, showing limited imagination.

3. SWARMS

Swarms is a multi-agent LLM framework for AI agents, providing extensive clustering architecture and seamless third-party integrations. It enables businesses to easily build and manage collaborations among multiple AI agents, allowing seamless cooperation under Swarms' orchestration to complete complex business tasks. Simply put, SWARMS is targeted at B2B users, offering enterprise-level AI agent applications.

Its founder, Dev, is a 20-year-old named Kye Gomez (source: online). He publicly claimed that OpenAI infringed upon the team's intellectual property, stole their project name, and copied their code structure and methods. Subsequently, Gomez issued a more detailed explanation: Swarms is a multi-agent framework that has been operating for nearly three years. To date, over 45 million agents have run in production to serve some of the world's largest financial, insurance, and healthcare institutions.

After the token was issued on December 18, it quickly peaked at a market value of $7.42 million by December 21, but unfortunately, it soon plummeted like a rollercoaster down to around $600,000. It then oscillated around $1.3 million until it started to rebound on the 27th, climbing from a low of $1.2 million up to $3 million, nearly reaching a previous high by pulling up nearly threefold to close to $7 million.

Compared to the fanciful constructs of AI16Z, if Swarms is indeed created by the 20-year-old AI genius Kye Gomez as rumored, then Swarms undoubtedly has strong technical barriers. Its official website has already provided efficient solutions to many businesses, substantiating its capabilities.

As an open-source project, Swarms has sparked heated discussions within the developer community, surpassing 2.1K stars on GitHub, gaining wisdom and support from many developers. Thus, everything accumulated by Swarms confirms the maturity and innovation of its technology. Swarms exhibits stronger technical capabilities and robust market demand (enterprise-level), likely allowing it to stand out in the AI Agent competition.

4. GRIFFAIN

Griffain is a project based on Solana—an AI agent engine similar to Copilot and Perplexity. It is one of the projects closest to an Agentic app. The ultimate form of a search engine in the AI era should allow users to directly present needs, with AI providing results or solutions directly, rather than merely supplying links to web pages. One of the catalysts for this project is its open-access mechanism. As a leading agent engine, Griffain undoubtedly attracts considerable market interest.

Solana currently has the most AI agents. In October, Goat, as an AI agent, conducted financing for humanity through pumpfun. From a certain perspective, this represents an AI singularity, given the excellent liquidity and the well-established developer community for AI agents on that chain. It's not an exaggeration to call Solana the most imaginative breeding ground for AI Agents in the blockchain space.

What Solana has done most importantly is reinvigorate the Griffain ecosystem. For the arrival of the true "Agentic app szn," infrastructure channels are needed in addition to AI. Although Griffain has not yet clarified the specific application scenarios for its tokens, it is expected to interface demand-side needs with the Solana ecosystem, meeting basic requirements within Solana's existing technical framework—whether it's targeting certain compliant tokens on Pumpfun or creating tokens. This vision has gained recognition from Toly, adding substantial imagination to Griffain's future.

5. AIXBT

AIXBT is one of the intelligent agents created based on the Base platform by Virtuals. It provides valuable market insights by monitoring Crypto Twitter and market trends through intelligent analysis tools. Some analytical content is shared on Twitter, while the rest is available only to token holders, who can interact directly with the intelligent agent through their dedicated terminal.

AIXBT's analysis has a certain accuracy in predicting price trends, showcasing how AI can interpret blockchain data and help traders make more informed decisions across multiple platforms and fields. After reviewing AIXBT's published content, I found it to be rich and covering almost the entire field, smoothly handling various datasets. Additionally, for short-term cryptocurrencies, there appear to be some potential investment opportunities; for instance, identifying that vapor in the hype is relatively undervalued compared to similar AI launchpads. Studies indicate that out of the 210 tokens recommended, 183 achieved profitability after being referred by AIXBT, with a profit rate of 83%.

However, some shortcomings exist, like the inability to fully break down complex items, and analysis and data are still somewhat shallow, lacking indications of investment risks. However, I believe it is significantly stronger than current cryptocurrency KOLs.

From the perspective of long-term value, AIXBT meets a segmented market demand, and users are motivated to hold tokens to unlock more information and price analysis. As AIXBT continues to evolve through data feeding, I believe it will become the absolute leader in market prediction AI agents.

In summary, after analyzing five popular AI Agent projects based on the criteria mentioned earlier, I rank these projects by their potential market capitalization from high to low: SWARMS, GRIFFAIN, Virtuals, AIXBT, and Ai16z.

Regarding the applications of AI Agents in the Web3 domain, there are several promising directions worth attention, which also represent future trends. One is privacy and security; AI must prioritize respect and protection for users and society in its design from the outset. However, as AI understands us more, privacy becomes increasingly ambiguous and fragile. Every interaction with smart devices, every input of personal information becomes food for AI evolution.

The importance of privacy issues is closely related to security issues. Systems that store and process personal data, once targeted by hackers, can lead to significant problems such as information leaks, identity theft, and asset loss. Is there an environment that can leverage AI's powerful capabilities while also protecting personal privacy? It stands to reason that the Web3 field can offer a higher level of data protection compared to traditional methods, perfectly balancing AI development capabilities and privacy protection concerns.

Therefore, we see that many large models are beginning to attempt data storage on the blockchain. The perfect AI soil in Web3 has also attracted many AI developers to ensure data security and privacy through blockchain technology in specific industries with high privacy needs, such as healthcare and finance.

Another vital direction is computing power and data. AI Agents, especially Multi-Agent collaboration, face cost issues related to development, training, and operation. For businesses, training AI Agents requires substantial computing power, often amounting to hundreds or thousands of high-performance GPUs or specialized hardware like TPUs. The costs associated with acquiring and utilizing these computing resources are already high. For example, Stanford's virtual town, which includes 25 agents, was merely a multi-agent research project. However, once the town framework was open-sourced, testing an agent consumed $20,000 worth of data sources in a single day.

In Web3, a reasonable token economy and user incentive programs can be implemented to utilize idle computing power or individual data sets, further reducing computing and data costs, enabling more individuals to participate in the construction of the AI industry. For instance, some data platforms allow users to monetize their own data to provide low-cost data sources for AI Agents.

Finally, I believe that AI Agents can serve as new infrastructure for Web3, deeply integrating with other core elements, giving rise to entirely new application modes, rather than simply being AI Memecoins. Currently, in the Web3 field, AI Agents can help users lower barriers and enhance experiences. Even simplifying parts of the asset issuance process is meaningful.

However, from a macro Web3 perspective, AI Agents, as off-chain products, currently serve merely as auxiliary tools for smart contracts, thus there is no need to overly inflate their capabilities. Given the lack of significant wealth effect narratives aside from MeMe in the latter half of this year, it is normal for AI Agents to gain popularity around MeMe.

However, relying solely on MeMe cannot maintain long-term value. Hence, if AI Agents can bring more innovative plays to the trading process and provide tangible grounding value, they may evolve into a widespread infrastructure tool.