There is not much content about the new narrative Web3 x Agent in China. We are honored to cooperate with HajimeAI, a potential AI project on Solana, to complete this research.
Looking back at January this year, the approval of the US Bitcoin spot ETFs triggered strong bullish sentiment and positive capital inflows among investors. The crypto market continued to grow in the first half of 2024, with a total market value increase of 37.3%. Several Crypto narratives have gained strong momentum. In my impression, Memecoin, AI, and RWA are the strongest performers. Judging from the performance of the coin price last week, AI is still relatively strong, second only to Memecoin.
Web3 x AI, this year, various Web3 VCs have made huge bets on this track, and can even buy old projects TAO (Bittensor) on the open market, so I don’t think this is a fleeting narrative. On the contrary, it will continue to innovate as the traditional AI track develops.
For example, AI Agent, the latest trend in the traditional AI field, has also been brought into the Web3 world. In the first half of the year, a large number of Web3 x AI Agent projects were launched, such as Spectral and Olas Network, while many old projects followed this narrative, such as Fetch.AI $FET and Phala $PHA. At the just-concluded ETHCC, many Web3 developers and top VCs began to focus on the AI Agent narrative.
This article starts with what I believe to be two of the most representative new projects to quickly understand this new narrative and the opportunities therein.
Table of contents:
1. What is AI Agent?
2. New changes of Web3 x AI Agent
3. Spectral Investment Research
4. GaiaNet Investment Research
V. Overview of other early projects
1. What is AI Agent
Simply put, AI Agent is an "agent" based on the Large Language Model (LLM) that can perceive the environment, understand independently, think independently, make decisions independently, and execute actions. Similar to the process of humans "doing things", the core functions of Agent can be summarized into three steps: perception, planning, and action.
So what is the difference between AI Agent and other AI chatbots like ChatGPT? In terms of purpose and capabilities
Chatbots are designed to interact with humans . Since AI chatbots are designed to help humans, they do not take autonomous actions.
Agents are designed to complete autonomous tasks and have the ability to take autonomous actions. You don't have to tell it what to do all the time, just give it a goal and it will find a way to help you complete it automatically.
For example, AI Agent is like a smarter Xiao Ai. When you are sick, you can say to it: "Xiao Ai, I feel a little unwell."
It will monitor your body temperature and other physical indicators, and after analysis, it will give you a conclusion: "You are positive". Then it will automatically generate a leave application for you, and you just need to nod and it will be sent directly to your boss. It can also sense that the antipyretic medicine at home is insufficient, and it will automatically help you add the medicine to the shopping cart. As long as you pay, it will be delivered to your door in 15 minutes.
I have used almost all of the most popular AI Agents in Web2, such as Perplexity, CrewAI, AutoGPT, and MultiOn. The more commonly used functions include batch document content extraction and Internet information integration (generating research reports). I have found MultiOn to be very useful, so I would like to introduce it in detail. Its main function is to simplify Internet user interaction , free users’ hands, and change the interaction mode between users and the Internet.
For example, I gave MultiOn a task, "Help me find the most played Web3 themed videos on Youtube", and all the steps from "open Youtube in the browser", "search for the theme of Web3", "filter the most played videos" were completed automatically. Finally, the Agent outputted me a video with 26 million views✅
New changes of Web3 x AI Agent
First of all, what can Web3 bring to AI Agent? In other words, what are the benefits of moving AI Agent to the chain?
Censorship resistance
AI Agent is based on LLM. Centralized LLM will lead to output biased towards censorship, which will limit the spread of real information to a certain extent. Using decentralized LLM to build AI Agent can solve this problem.
Decentralization/Ownership
By the same token, unless you spend a huge amount of data resources on LLM yourself, the core data of the AI Agent is still stored in the centralized AI provider.
Monetization
AI Agent Launchpad? Issue tokens to govern a certain AI Agent, IAO (Initial Agent Offering), providing a disguised channel for AI Agent developers and investors.
Composability/Lego
AI Agents interact, trade, and empower with other Agents on the same network. If we refer to the composability of DeFi, it will not be surprising to use only one Web3 AI Agent to select investments, find the most liquid DEX, complete token swaps, and monitor returns in the future.
Conversely, what can AI Agent bring to Web3?
1. Conversational AI Agents can be put on the chain . In addition to the collection and organization of professional knowledge in specific fields that Web2 Agents can do, they can also retrieve and summarize information on the Web3 chain, greatly simplifying the chain research process for Web3 users.
2. Just looking at the concept of AI Agent and MultiOn, it really reminds me of the Intent-Centric that I have been talking about. What we need to do is to free the hands of users on the chain , change the way they interact with Web3, and realize Mass Adoption. These interactions include but are not limited to Swap and airdrop interactions. Imagine saying to AI Agent, "Help me complete the Linea airdrop interaction", and AI Agent can get the airdrop tutorials of KOLs from the Internet, and then automatically complete it through the on-chain wallet following the steps. (It seems that there is no need for a LuMao studio?)
Or “Please build an ETHBot for me. When ETH is below MA200, use 30% of my USDT balance to buy ETH”, and monitor the market 24/7. Isn’t that sexy?
If each solver uses an AI Agent that can automatically interact on the chain, it actually realizes the most important piece of the intent protocol puzzle.
For more information about Intent-Centric, please refer to my previous WeChat article.
So here, I would like to divide the current specific products in the Web3 x AI Agent track into two types. One is the Web2-style on-chain conversational AI Agent , and the other is the more Web3 Native AI Agent mentioned in the previous paragraph.
The first type can be understood as an AI Agent created on a certain Layer-1, which is suitable for Web3 users to learn professional knowledge in a specific field and do on-chain research. It does not include on-chain operations
The second type can be understood as logic on-chain, helping users to achieve some specific interactions on the chain. Including on-chain operations
Of course, my classification is only the most basic one based on " whether the AI Agent has the ability to interact on the chain ", and does not consider whether it is based on a reliable and verifiable decentralized AI model.
Below we introduce a typical project, GaiaNet and Spectral, for horizontal comparison.
Spectral (capable of on-chain operation)
The predecessor is an Ethereum-based credit scoring protocol that aims to provide lenders with a new way to assess the credit risk of borrowers.
2023 Q4 Transformed into a machine intelligence network, allowing users to build on-chain AI Agents and form an on-chain agent economy.
1. Business Model
Spectral has four important products:
Spectral Syntax is an officially developed Agents collection. Users can tell an Agent what to do, and the Agent will convert natural language intent into executable code to help users complete it. For example, creating NFTs, creating Memecoins, automated transactions, and on-chain information retrieval. I personally experienced MoonMaker Agent, which can automate everything from name creation, logo design to CA deployment.
Spectral Nova is a decentralized platform that provides machine learning reasoning directly to smart contracts. Model creators such as top scientists, enterprises, Dev, engineers, etc. can build AI models and obtain user-paid income. In addition, model creators can issue challenges with rewards. Solvers (bounty hunters) solve challenges and win rewards or receive revenue dividends. Creators, solvers, verifiers, and consumers interact with each other on Spectral's machine intelligence network, forming a flywheel through incentive mechanisms.
Agent Wallets , launched in Q3 2024, will help users implement on-chain operations and simplify the Web3 user experience by integrating AI Agent into the wallet. For example, it supports gas-free transactions using USDC, and the Gas Asset Agent will automatically perform the exchange.
Inferchain , a Layer1 with AI Agent as the narrative, will integrate Agents in Syntax and Nova to facilitate interoperability between these Agents. It will be launched in Q4 2024.
2. Development History
2021-2022 Complete financing as Web3 credit risk assessment infrastructure
2024.03 Launch Syntax, officially transformed into an on-chain AI Agent narrative
2024.05 TGE, start the first season of airdrop
2024.06 Cooperate with crypto wallet Turnkey and launch Agent Wallet in Q3 (Turnkey has raised $15 million, with Sequoia and Coinbase participating)
3. The intersection of AI and Web3
Inferchain is the last piece of the puzzle for the Spectral ecosystem, enabling the easy development of on-chain AI Agents that can communicate with each other, ultimately making AI applications in the Web3 field transparent, decentralized, and verifiable.
4. Value generated
It solves the problems of centralized AI, such as high trial and error costs, reliance on a single source, and lack of information authenticity;
Enable ordinary users (non-technical) to quickly create on-chain Agents;
Facilitates communication between on-chain Agents.
5. Token Economics
$SPEC, users can use tokens as a means of payment to pay for the use of community-developed AI Agents. In addition, it is also used for decentralized governance and staking mechanisms:
In Spectral Syntax, users who stake SPEC have the right to create AI Agents and access AI Agents created by the community;
In Spectral Nova, validators need to stake SPEC as collateral to verify challenges completed by solvers.
SPEC has been listed on trading platforms such as Bybit, Gate.io, and MEXC, with a circulating market value of US$85 million and a FDV of US$800 million.
Currently, the circulating portion is only the first quarter airdrop + market maker share, accounting for 10.3% of the total
⚠️ 2025.05 Unlocking for core contributors and investors will start. If the crypto market continues to rise in the next few months of this year, please pay attention to the risk of selling after unlocking.
6. Team Background
Sishir Varghese - Co-Founder and CEO
Previously served as co-founder and executive partner of AlphaChain and strategic partner of Loopring
Mihir Kulkarni - Head of Product
Previously worked as an institutional product operations manager at Coinbase
7. Financing
Round 1:
Galaxy, ParaFi Capital, Maven 11, Alliance DAO, Rarestone Capital…
Round 2:
General Catalyst, Social Capital, Jump Capital, Samsung Next, Circle Ventures, Franklin Templeton, Section 32…
Super large VC investment background, such as Franklin Hampton, Samsung, Google
8. AI Agent Usage
Q2 data released in June 2024:
Registered users: 65,362
Number of contracts generated by SYNTAX 1,055,568
Average number of interactions per user 25
Number of Memecoins created using MoonMaker 5,043
4. GaiaNet (no on-chain operation capability)
GaiaNet is a distributed AI infrastructure that will gradually become a decentralized AI Agent ecosystem
1. Business Model
Node = Agent . In my own practice of building nodes, when I talk to the AI Agent corresponding to the node, the output of the AI Agent consumes computing resources.
Create an AI Agent network based on Ethereum that aggregates knowledge bases in various fields in the form of nodes , allowing individuals and companies to quickly build AI Agents based on their own expertise in specific fields and provide them to the demand side for profit.
The GaiaNet network consists of three core components:
• GaiaNet Node
A comprehensive software stack that enables individuals and businesses to quickly deploy AI agents that incorporate their own expertise.
• GaiaNet Domain
A collection of nodes registered under an Internet domain name, managed by a domain name operator. The official idea is that each domain name is specific to a professional field, such as finance, healthcare, and education, and under these domain names are AI agents with specific functions.
User process:
Users pay the node (AI Agent), and the fees are held in the on-chain smart contract. A portion of the fees is taken by the domain name operator, who then provides services to users.
• GaiaNet DAO
Stakers stake tokens on domain name providers, providing “trust”;
↓
The domain name provider manages qualified nodes and provides "guarantees";
↓
Users choose to use the AI Agent under the domain name provider, and the fees are shared by nodes, domain name providers, and Stakers.
In fact, there is another role, that of component developer. Non-AI Agent developers can obtain benefits from AI Agent developers in need by fine-tuning NFT-shaped models, knowledge bases, plug-ins and other components.
2. The intersection of AI and Web3
Moving the Web2 AI Agent ecosystem to the blockchain
3. Value generated
Developers can deploy Agents more easily. Gaianet nodes support all open source LLMs, multimodal models, text-to-image models, and text-to-video models. Developers can add and fine-tune models according to their own choices.
Based on the domain name provider, a collection of AI Agents with expertise will be generated for each industry and each field.
4. Token Economics
No tokens have been issued yet. Use other people's Agents to pay tokens in GaiaNet
5. Team Background
Matt Wright - Co-founder and CEO
Graduated from UCLA, previously served as the community director of Consensys and co-founder of EVM Capital. Previously worked at JP Morgan.
Shashank Sripada - Co-founder
I have a venture capital background in Web2 and have founded and worked for several VCs. I graduated from the London School of Economics and Political Science and have some political background.
Sydney Lai - Head of Development Communications
Co-founder and CTO of EVM Capital, graduated from the University of California, Berkeley
6. Financing
2024.05.28, Seed Round, $10 million
Mantle Network, ByteTrade, EVM Capital, Mirana, Lex Sokolin (Co-founder of Generative Ventures), Kishore Bhatia (Founding Member of Superscrypt), Brian Johnson (Head of Crypto at Republic Capital)
7. AI Agent Usage
The number of users is unknown, with only 35 conversational AI agents available, covering finance, Crypto, programming and other fields. Current total number of nodes: 18,594
Since we are in the testing phase and can run nodes in multiple threads, we speculate that the majority of nodes are those that receive airdrops in batches.
5. List of other early projects
Zotto : Users can create their own AI Agent to realize transaction intentions. Application scenarios include mirroring smart money addresses, multi-condition triggering transactions, etc. It has just been launched on Testnet, so you can pay close attention to it.
AgentLayer : Layer2 built on OP Stack, with AI Agent as the main narrative, aims to promote the connection between AI Agents developed by itself and the community.
Olas Network : An off-chain AI Agent ecosystem where a single agent or multiple agents work together to complete tasks and pass the output to the chain.
Theoriq : Theoriq aims to be a modular and composable AI Agent base layer.
AgentCoin : Transformed from evo ninja, a mature AI Agent product in Web2, to a general Web3 AI Agent.
Giza : We are working on the Web3 AI Agent framework, using ZKML for off-chain reasoning and on-chain execution.
Olas Network : A Web3 AI Agent ecosystem where a single agent or multiple agents work together off-chain to complete tasks proposed by users and pass the output to the chain.
<END>
