GuShiio.com When the market is deserted, let me talk about my thoughts on the entire track of web3 + ai agent.
When the market is deserted, let me talk about my thoughts on the entire track of web3 + ai agent.
When ai was born the year before last, it was already confirmed that it was a technological revolution comparable to the invention of the Internet in the future. However, at the beginning, everyone’s attention was on the llm 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 analysis and reasoning capabilities and is difficult to interact with the real world. So we naturally thought of adding various sensors and functional modules to llm. Among them, the Internet-related ones are even simpler. Connecting to the api interface can realize various functions, so ai officially evolved into an ai agent.
The big language model is of course important. It determines the intelligence level of ai agents. Currently, the 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 will begin to explore ai agent + various industries. such as
- ai agent+ e-commerce,(ai directly helps select products and generate copy and pictures)
- ai agent+ Internet,(ai programming)
- ai agent+ finance,(ai analyzes data in real time and makes intelligent investment)
- 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 now, so naturally it has also emerged.
ai agent+web3
But web3 itself is also divided into many sub-fields, so ai agents will also have different ways to integrate industries, such as
ai agent +meme
ai agent directly replaces humans to issue meme coins, representing goat fartcoin
ai agent+defi
AI agents directly replace humans to analyze market data for analysis (aixbt), manage funds (degenai) and defi operations (griffain)
ai agent+ games
AI agents directly participate in game commentary (luna), game playing (freya), sports betting analysis (dwain), ai game frameworks (digimon/dreams)
ai agent + social
Act started out as a project to study how agents socialize, and later evolved into meme
Of course, there are also ai agent + depin, etc., which will not be detailed here.
AI agent is already an implemented technology in web2 and is by no means an empty slogan. In the web3 industry, it is nothing more than adding a coin issuing process for everyone to hype.
Some people always say that web3 ‘s ai agent technology is much worse than web2, but ai agent + web3 is originally a transplant of web2 technology to web3 for application. Finding the integration point of the industry is the key.
I personally divide web3 ai agents into the following aspects: ai agent infra (infrastructure) and ai agent (agent for special application scenarios)
1. ai agent infra (infrastructure)
1. A fine-tuned big language model
llm is equivalent to the brain of an ai agent, and llm that has not been fine-tuned is equivalent to connecting a graduate regardless of majors. The llm, which has been fine-tuned by industry data, is equivalent to a college student who has graduated from a major. Currently, lumo is the only one 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 and it has also been the main market trend in the first two months, including elizaos, arc, swarms, etc. The ai agent framework is actually a set of rules for llm to call various functional modules, making it convenient to create agents with unified rules.
3，launchpad
The above-mentioned frameworks are often open source code. Users need to understand some simple code to create ai agents and purchase their own servers to deploy them. AI agent launchpad is a saas version of the framework that provides a code-less agent creation environment (users only need to manually enter a few parameters) and comes with a server. Although it further simplifies the process of creating ai agents, it also reduces flexibility. Well-known launchpads include virtual, vvaifu, avaai and others
4. Special functions
There are also other special function projects mainly to make ai agents run better, faster and stronger.
For example, web3 is a special industry. Due to frequent interactions with large amounts of funds, it needs to have sufficient independence and verifiability for the AI agent’s running environment (such as ensuring that the wallet is operated by the AI agent rather than by a person). The well-known Tee environment of phala.
2. ai agent (agent for special application scenarios)
The diversity in this direction is very wide. It can be regarded as application-side exploration and innovation. It can be specifically divided into:
- Analysis class: Analyze and reason a certain sub-field and output content
- Operation class: Direct operation after analyzing a sub-field
The following examples illustrate,
1，aixbt （ @aixbt_agent ）
Currently, the most well-known defi ai analyst is good at conducting brief analysis of various market projects and outputting content on x to interact with people.
2， truth of terminal ( @truth_terminal )
The pioneer of the ai agent track, good at nonsense and issuing meme coins
3，ava ( @AVA_holo )
Holowworld’s flagship agent is good at using video to randomly analyze the market
4，yne ( @yesnoerror )
Analyze scientific research papers and correct mistakes
5， buzz ( @askthehive_ai )
The above are some of my analysis frameworks for the web3 + ai agent track. Although it is in a trough now, ai agent is also in a prosperous state in web2. It is the hottest track in 2025, and it will inevitably bring more fun products and imagination to web3. It is one of the few directions that can continue to attract attention and builder.
