When the market is quiet, I want to talk about my views on the entire track of web3 + ai agent.

When AI emerged two years ago, it has been confirmed that it will be a technological revolution comparable to the invention of the Internet in the future. However, at the beginning, everyone's attention was focused on the LLM large language model, and they were not sure how AI would affect the real world.

If we compare the LLM model to the brain, it is obvious that AI only has analytical and reasoning capabilities and it is difficult to interact with the real world. So we naturally think of adding various sensors and functional modules to LLM. Among them, those related to the Internet are even simpler. Various functions can be realized by connecting to the API interface. Therefore, AI officially evolves into AI agent.

Large language models are of course important, as they determine the intelligence level of the AI agent. The current top models, such as OpenAI's O3, Deepseek's R1, etc., exceed the level of human doctoral students in many aspects.

So in 2024, various industries began to explore AI agents + various industries. For example

  • ai agent + e-commerce, (ai directly helps select products, generate copy and pictures)
  • ai agent + Internet, (ai programming)
  • ai agent + finance, (ai real-time data analysis, intelligent investment advisor)
  • ai agent + supply chain, (ai optimizes supply chain management and monitors production lines)
  • ai agent + education, (ai real-time tutoring, providing personalized education)

Web3 has developed into a huge industry, so naturally it has also emerged

ai agent+web3

However, web3 itself is also divided into many sub-fields, so AI agents will also have different ways of combining industries, such as

ai agent +meme

AI agent directly replaces humans to issue meme coins, represented by goat fartcoin

ai agent+defi

AI agents directly replace humans to analyze market data for analysis (aixbt), fund management (degenai) and DeFi operations (griffain)

ai agent+game

AI agent directly participates in game commentary (luna), playing games (freya), sports betting analysis (dwain), AI game framework (digimon/dreams)

ai agent + social

ACT was originally a project to study how agents can socialize with each other, and later evolved into a meme.

Of course, there are also AI agent + depin, etc., which will not be listed here one by one.

AI agent is already a practical technology in web2, and it is not just an empty slogan. In the web3 industry, it is nothing more than adding a coin issuance process for everyone to hype.

Some people always say that web3's ai agent technology is much worse than web2's, but ai agent + web3 is originally the technology of web2 transplanted to web3 for application. Finding the integration point of the industry is the key.

I personally divide web3 ai agent into the following aspects: ai agent infra (infrastructure) and ai agent (agent for special application scenarios)

1. AI agent infra (infrastructure)

1. Fine-tuned Large Language Model

LLM is equivalent to the brain of AI agent. An LLM without fine-tuning is equivalent to a graduate without a major. An LLM fine-tuned with industry data is equivalent to a college student who has chosen a major. Currently, only Lumo is working in this direction. The difficulty lies in the collection, cleaning and labeling of industry data.

2. Framework

There are many competitors in this direction, which is also the main market trend in the past two months, including Elizaos, Arc, Swarms, etc. The AI agent framework is actually a set of rules for LLM to call various functional modules, which facilitates the creation of agents with unified rules.

3. Launchpad

The above frameworks are often open source codes. Users need to understand some simple codes to create AI agents and purchase their own servers for deployment. AI agent launchpad is a SaaS version of the framework, which provides a code-free agent creation environment (users only need to manually enter a few parameters) and comes with its own server. Although it further simplifies the process of creating AI agents, it also weakens flexibility. Well-known launchpads include virtual, vvaifu, avaai, etc.

4. Special functions

There are also some other special function projects mainly to make the AI agent run better, faster and more powerful.

For example, web3 is a special industry. Due to frequent interactions with large amounts of funds, the operating environment of the AI agent needs to be sufficiently independent and verifiable (for example, ensuring that the wallet is operated by the AI agent rather than a certain person). A well-known example is the TEE environment of Phala.

2. AI agent (agent for special application scenarios)

The diversity of this direction is very broad, which can be regarded as exploration and innovation on the application side, and can be specifically divided into:

  • Analysis: Analyze and reason about a sub-field and output content
  • Operational: Directly operate after analyzing a sub-field

The following example illustrates this:

1. aixbt (@aixbt_agent)

Currently the most well-known DeFi AI analyst, he is good at making brief analyses of various projects in the market and outputting content on x to interact with people.

2. truth of terminal (@truth_terminal)

A pioneer in the ai agent track, good at talking nonsense and issuing meme coins

3.ava (@AVA_holo)

Holoworld's flagship agent, good at using videos to make random market analysis

4. yne (@yesnoerror)

Analyze scientific research papers and correct errors

5. buzz (@askthehive_ai)

I haven't used it, but the introduction says that users can interact with the system through a natural language interface, and the agent can handle a variety of tasks including trading, staking, liquidity management, and market sentiment analysis.

The above is my analysis framework for the web3 + ai agent track. Although it is in a trough period now, ai agent is also in a thriving state in web2. It is the hottest track in 2025. It will inevitably bring more interesting products and imagination space to web3. It is one of the few directions that can continue to attract attention and builders.