A Comprehensive Read on the AI Agent Track: The Decentralized Ambition of Multi-Intelligent Body Networks

In the first two industrial revolutions, humanity gradually replaced muscle power with mechanical power. In the AI-driven fourth industrial revolution, we are replacing cognitive abilities with computational power.

In the past year, with the rapid development of generative AI large models like ChatGPT, AI has expanded from simple automation tools to complex decision-making and prediction systems, gradually growing into a driving force for contemporary social progress. AI has become a hot topic pursued by capital circles in Europe and America, as well as a conversation starter at offline gatherings of tech professionals.

This trend has also spread to the Web3 market, becoming a major collision between two of the hottest technologies. In 2024, a large number of AI concept projects emerged in the Web3 industry, attempting to integrate AI technology with blockchain. Some projects utilize AI's content generation, analysis, and other functions in areas such as GameFi, SocialFi, and data analysis within the Web3 domain. Some projects are building decentralized computing networks to combat the monopoly of large tech companies over computing power using blockchain.

Just last week, Coinbase announced support for AI Agent developers to integrate into its MPC wallet through the Coinbase Developer Platform, becoming one of the first companies to provide on-chain payment infrastructure for AI Agents.

"AI Agents cannot get bank accounts, but they can get crypto wallets" --- Coinbase CEO Ben Armstrong

From this perspective, the second half of Web3+AI is witnessing a massive explosion in the AI Agent track.

1. AI Agent: Why is Multi-Agent the Future?

In the past year, research and discussions about AI Agents have surged, with projects like AutoGPT, Hebbia, Glean, BabyAGI, Generative Agents, and MetaGPT garnering thousands of stars on GitHub, becoming hot star projects. Star startups like Zapier, Glean, and Hebbia have reached valuations as high as $5.7 billion.

AI Agents, or "Artificial Intelligence Agents," are entities capable of perceiving their environment for autonomous understanding, decision-making, and action execution. AI Agents can not only automate tedious processes but also make precise decisions and intelligently interact with their environment.

The prospects for AI Agents are vast. IDC, in its report "Top 10 Trends in AIGC Applications for 2024," found that 50% of enterprises have piloted the use of AI Agents in certain tasks, while another 34% are formulating related application plans. It is expected that by 2027, over 60% of smartphones will have generative AI capabilities, laying the hardware foundation for the widespread adoption of AI Agents.

However, the reality is that while AI Agents excel in specific scenarios due to their integration of various tools and strong reasoning capabilities, they often fail to provide optimal solutions when faced with complex real-world tasks. This limits the current usage of AI Agents by a broader audience. Additionally, collaboration among AI Agents from different models and ecosystems of various AI giants is also challenging.

Therefore, the next technological revolution for AI Agents is the Multi-Agent System (MAS). A Multi-Agent system architecture consists of numerous independent, autonomous AI Agents, each possessing unique domain knowledge, functional algorithms, and tool resources, which can collaborate flexibly to complete intricate decision-making tasks. Multi-Agent systems not only significantly enhance overall work efficiency but also empower the handling of complex and diverse tasks.

To better understand Multi-Agent systems, let's take the example of an ant colony foraging:

• Each ant is an independent Agent with its own simple behavioral rules.

• Ants release pheromones while searching for food, which other ants can sense.

• Through the collaboration of many ants, the entire colony can find the shortest path to the food source.

From this example, we can see the core feature of Multi-Agent systems: multiple autonomous Agents interact and collaborate to complete complex tasks or solve problems. In the future, we may even see a company composed entirely of AI agents, with only a CEO founder.

Overall, AI Agents represent a new paradigm of interaction between artificial intelligence and humans, promising to fundamentally change the way people live and work, driving transformation in the software industry. As technology continues to advance and application scenarios expand, AI Agents will play an increasingly important role in the future.

2. What Transformations Can Web3 + AI Agent Bring?

In 2023, the new wave of AI revolution brought by OpenAI has disrupted existing models across various industries, and we have seen many related concept products continuously emerging and landing in Web3, utilizing crypto features to develop AI solutions such as decentralized computing and data.

Even in the current less favorable market conditions, AI narratives remain strong in Web3. From the perspective of price performance, AI is second only to Memecoin, making it the second-largest narrative.

During the ETHCC conference held in Brussels this year, Ethereum co-founder Vitalik once again shared his views on Web3 + AI:

From a short-term perspective, the development of AI is a "collaborative intersection" between humans and AI. In the long run, AI will address many of the challenges humanity faces today, such as longevity and space travel. Web3 and decentralization will outline the path to this ideal while preventing extreme scenarios where fully autonomous AI Agents could destroy humanity.

In the future, Web3 + AI Agents will become one of the important narratives, and it is foreseeable that Web3 AI Agents will flourish across various Layer1 ecosystems. So, what transformations will the introduction of the latest trend in AI, AI Agents, into Web3 bring?

1. What Opportunities Can Web3 Bring to AI Agents?

Decentralization

Web3 provides decentralized infrastructure, allowing AI Agents to achieve self-hosting, avoiding the data privacy and security risks associated with centralization.

In Web2, AI giants like OpenAI and Anthropic have received significant funding and control closed-source AI model training data, which not only creates single points of failure for AI Agents but also limits community participation and collaboration, hindering innovation and progress in AI Agents.

Deterministic Execution Environment

Web3 offers AI Agents a deterministic execution environment, unaffected by human-established trust elements, unnecessary intermediaries, and other inefficiencies.

AI Agents cannot obtain bank accounts or book flights for users; however, they can acquire wallets and use crypto stablecoins to transact with users, merchants, and other AIs worldwide.

Security and Privacy

Data privacy and security are among the most challenging issues for AI Agents in practical applications within Web2. AI Agents need to collect and process large amounts of data, including personal information, and any unauthorized access or data breaches can severely compromise user privacy. Therefore, blockchain, which inherently possesses security features, can assist in ensuring data safety.

Monetization and Investment Value

The monetization of AI Agents creates investment value for AI and stimulates new token economic models. Through the Initial Agent Offering (IAO) model, AI Agents can become a new investment target, decentralizing ownership to the community through DAO governance.

Market Transformation and Mass Adoption

Web3 provides a new market environment that encourages Web2 enterprises and developers to focus on creating unique and outstanding AI Agent projects to meet competitive pressures and market demands. This not only helps enterprises gain a leading position in the Web3 environment but may also promote broader Web2 user adoption of Web3 and blockchain.

Optimizing AI Datasets and Models

Web3's unique characteristics, such as decentralization and publicly transparent data records, can optimize the diversity of AI datasets and the transparency of models. Training AI models using on-chain data from Web3 helps establish large models based on on-chain data, providing unique perspectives and advantages.

2. What Innovations Can AI Agents Bring to Web3?

Enhanced User Experience

By combining AI's analytical capabilities, Web3 applications integrated with AI Agents can provide users with personalized, automated, and customizable experiences, further unlocking the potential of the on-chain economy.

Lowering Industry Participation Barriers

AI Agents can serve as tools to lower the barriers for people to participate in the Web3 industry, acting as "intelligent assistants" between users and on-chain protocols, helping users complete various complex on-chain transactions, making Web3 products more user-friendly and conducive to mass adoption. For example, AI Agents can perform crypto investment analysis, automate on-chain trading, and monitor investment portfolios based on user requests.

Innovative Applications

In the gaming and entertainment sectors, AI Agents can provide dynamic, immersive experiences, enhancing the value of user-generated content while ensuring transparency and reliability through blockchain technology.

In summary, Web3 + AI Agents not only drives the advancement of Web3, blockchain, and AI technologies but also offers new opportunities for developers. The transformations brought by Web3 to AI Agents primarily manifest in decentralization, security, execution environments, and monetization. Conversely, AI Agents will also bring innovations to Web3 through their capabilities, simplifying user experiences and improving efficiency and decision quality through automated task execution, laying the foundation for the mass adoption of Web3.

3. Comparison of Popular Web3 AI Agents

Spectral

Spectral is a project dedicated to building a Web3 on-chain AI Agent economy by providing zero-threshold smart contract compilation and deployment services, unlocking the innovative potential of AI and Web3 integration.

Specifically, Spectral is offering two unique products:

Spectral Syntax is an on-chain AI Agent platform that can understand natural language intentions and convert them into code-based instructions, aimed at enabling Web3 users and developers to realize their intentions through specific AI Agents. Application scenarios include on-chain contract generation and deployment (one-click meme coin creation), smart contract vulnerability scanning and fixing (smart auditing), and on-chain information retrieval. In the third quarter of 2024, Spectral will launch Syntax V2, allowing users to create their own AI Agents based on all of Spectral's tools, knowledge base, and APIs for various imaginable intentions.

Spectral Nova is a machine intelligence network focused on the creation and application of AI and ML models, attracting top data scientists and ML engineers to build models that output reasoning sources to solve prediction and machine intelligence problems for web3 applications, thus meeting the demand for reasoning sources from smart contracts, companies, and individuals. Model creators, solvers, validators, and consumers interact on Spectral's machine intelligence network, forming a flywheel.

Inferchain is a Layer2 that Spectral is building, set to launch in the fourth quarter of 2024, with the vision of becoming a universal, permissionless, open truth verification layer for validating all on-chain AI Agent interactions. All AI Agents created on Syntax and the various reasoning sources they use from Nova will be integrated through Inferchain.

Spectral's core competitive advantages in the Web3 + AI Agent track are reflected in:

• Low-Threshold Development

Spectral provides one-click generation and deployment of smart contracts, significantly lowering the development threshold for Web3. This allows even novice users to easily compile and deploy smart contracts, representing an application scenario for AI in Web3.

• Multi-Scenario Adaptability

Spectral's existing product architecture is highly adaptable to the current diverse application scenarios of Web3, including DeFi, DAO governance, NFTs, security audits, and more.

• Product Iteration

Spectral continuously focuses on the functional iteration and optimization of its core products, Syntax and Nova, maintaining technological leadership.

Opportunities and Challenges: After the launch of Spectral's token $SPEC, its FDV once reached $1.5 billion, with total financing amounting to $30 million, backed by top VCs in Web2 and Web3 such as General Catalyst, Social Capital, Jump Capital, Circle Ventures, Franklin Templeton, and Galaxy, making it one of the most noteworthy projects in the Web3 AI Agent track.

Spectral primarily targets the relatively small "AI for Web3" market, using generative AI technology and blockchain to popularize Web3 development and many functional scenarios, providing verifiable model reasoning capabilities for Web3 dApps and expanding the scenarios of the Web3 application layer.

However, the three AI Agents currently launched by Spectral face significant competition pressure from homogeneity, and the operational paradigms of the four roles in the Nova network require strong operational maintenance and external resource input, posing severe challenges to initiating growth flywheels.

Autonolas/Olas

Autonolas, launched in the summer of 2022, also known as Olas Network, is a Web3 AI Agent ecosystem that operates by completing user tasks through a single Agent or multiple Agents off-chain, with outputs delivered on-chain. Meanwhile, the completion process of off-chain Agents is also recorded on-chain.

The uniqueness of Olas Network lies in the fact that each constructed AI Agent is operated by individual operators, capable of extracting data from any source and operating across different chains such as Ethereum, Solana, and Polygon, and can perform complex processing such as machine learning. Through its Multi-Agent system, Olas Network allows users to collaborate with multiple AI Agents simultaneously. Through incentive mechanisms, Olas Network connects AI Agent developers, operators, and guarantors to jointly support the development of a decentralized AI Agent ecosystem.

Olas Network's core competitive advantages in the Web3 + Agent track are reflected in:

• Web3 Native

As the core Multi-Agent development team of the former Fetch.AI, Olas's technical capabilities have been validated. AI Agents on Olas Network can autonomously operate and interact in a Web3 environment, providing users with more efficient automation and intelligence.

• Comprehensive DAO Infrastructure

Olas Network provides tools and infrastructure for building and managing DAOs for AI Agents, enabling more efficient community governance and operations.

• Composability

Olas Network has a high degree of composability, allowing developers to assemble different functional AI Agent components like building blocks to create complex decentralized applications. This composability reflects the "fat protocol" concept of Web3, helping to accelerate innovation and application development.

• Cross-Chain Interoperability

Olas Network supports cross-chain operations, which is significant in the multi-chain Web3 ecosystem. Cross-chain capabilities can facilitate value flow and information exchange between different blockchain networks.

Opportunities and Challenges: Autonolas is one of the earliest projects in Web3 to propose achieving Multi-Agent capabilities, with its token $OLAS reaching an FDV of $4 billion after launch, comparable to leading AI x Web3 projects like IO.net and Aethir, indicating a considerable market recognition of the narrative ceiling for Multi-Agent.

As a pioneer connecting the on-chain economy of Ethereum and off-chain AI Agents, its concept of "co-owned AI" aligns well with Ethereum co-founder Vitalik's thoughts on balancing the risks of AI centralization in Web3. In terms of demand exploration, as the native team of Fetch AI, the OLAS Network inevitably starts from the existing unmet demand scenarios in Web3, hoping to enhance the user experience through AI, but it also faces growth resistance due to low willingness on both supply and demand sides.

MyShell

MyShell is a decentralized AI Agent consumer layer that encompasses a large number of open-source and closed-source AI models, allowing creators to quickly build AI Agent applications and easily capture users.

Specifically, MyShell consists of four core modules: model layer, developer platform, AIpp store, and incentive network. The first three modules cover the entire process from creator production to final user consumption of AI Agents, while the incentive network organically connects the first three to achieve a closed-loop business model.

Interestingly, MyShell also allows developers to monetize AI Agents, but the method differs from ICOs. In the newly launched AIpp store section, developers "package" their AI Agents as AIpps and then conduct pre-sales and public sales. During the pre-sale phase, the share price is calculated based on a Bonding Curve, increasing with the number of purchases. When 30 shares are sold or three days have passed, the pre-sale will end and enter the public sale phase, still trading according to the Bonding Curve.

Developers have the right to purchase shares of their own AI Agents during the pre-sale phase and receive 5% of each transaction as a handling fee.

MyShell's core competitive advantages in the Web3 + Agent track are reflected in:

• Community Building and Participation

Compared to other projects, MyShell places greater emphasis on community building, enhancing user engagement and loyalty through mechanisms such as its badge system.

• Product Innovation

MyShell's product development direction is more aligned with current Web3 practices, especially the newly launched AIpp store, which facilitates rapid understanding and adoption by Web3 users.

Opportunities and Challenges: MyShell has raised over $16 million in total financing, making it one of the most active projects in the Web3 AI Agent track with a thriving creator economy. Its AI chatbot launch method, similar to Pump.fun, is more in line with Web3 user habits, with over 130 AI bots successfully launched in the first season that ended in July, totaling over 1.2 million USDT in transaction volume. From a long-term development perspective, MyShell still needs significant upgrades in product matrix and platform openness to face the fierce competition in the chatbot track and embrace the new paradigm of Multi-Agent.

HajimeAI

HajimeAI is an emerging "Web3 for AI" project that came to prominence in the second quarter of this year, being the first to propose a sidechain structure on Solana, aimed at providing stronger performance and more potential use cases for Solana L1 ("function extension layer for L1"), while avoiding liquidity fragmentation caused by Ethereum's scaling model.

HajimeAI is the first Web3 + AI Agent platform on Solana, serving as the artificial intelligence application layer for Solana. It not only addresses the current bottlenecks of decentralization, monetization, and reasoning capabilities faced by AI Agents, as well as Multi-Agent collaboration, but also lays a solid foundation for personalized personal AI Agents and a thriving AI Agent ecosystem on Solana in the future.

HajimeAI consists of three core components:

• Hajime Benchmark DAO

The first AI Agent usability ranking in Web3, where any user can find the most suitable decentralized AI Agent. Members of Hajime Benchmark DAO evaluate each newly released AI Agent in Hajime based on key dimensions, earning protocol revenue shares and Hajime token rewards.

In the early stages, HajimeAI will help empower Solana Saga by incentivizing Solana Saga users to become initial members of Hajime Benchmark DAO through airdrops. By participating in the evaluation of AI Agents, Solana OG users have the opportunity to join the wave of development in the Solana AI ecosystem while receiving platform incentives.

• Hajime Garden

AI Agents evaluated by the DAO will be listed in Hajime Garden, the intention center of the Hajime ecosystem. Based on the decentralized Multi-Agent Graph (deMAG) mechanism, Hajime Garden can decompose any intention proposed by users into multiple tasks, which are then processed by specialized AI Agents. Whether it's five steps or ten steps, whether it's Web2 knowledge or Web3 interaction, any submitted intention will be perfectly executed.

Another core function of Hajime Garden is IAO, similar to IDO, aimed at addressing the monetization and centralization challenges faced by AI Agents in Web2. Compared to traditional financing, the IAO process is simpler and faster, allowing AI Agents to obtain the necessary funding more quickly. The global participation characteristic of Web3 also makes DAO governance for AI Agents a reality.

• Hajime AI Layer

A Solana L2 sidechain focused on AI, operating in parallel with the Solana network, achieving "off-chain computation - on-chain verification," while still benefiting from Solana's security and verifiability. All AI Agents in the Hajime ecosystem are built on the Hajime AI Layer, enabling multi-agent collaborative work. The reasoning computation required by AI Agents and the demand decomposition capabilities of MAWG are supported by the Hajime AI Layer.

Opportunities and Challenges: As a team that won awards at the Solana global hackathon, HajimeAI has demonstrated its potential in the Web3 x AI field. HajimeAI has built Solana's first AI sidechain, becoming a key component for AI execution to meet AI computing needs. Through its self-developed Multi-Agent workflow graph deMAG and innovative IAO mechanism, it accelerates the development of interoperable on-chain AI Agents, paving the way for the democratization and mass adoption of the Solana AI ecosystem.

However, HajimeAI has yet to release a testnet and beta products, and it remains to be seen whether on-chain AI Agents can realize their interoperability vision and whether the performance of the HajimeAI sidechain can support large-scale AI applications. Addressing these issues will be key to the successful application of AI Agents in the Solana and Web3 ecosystems, which is worth looking forward to.

Theoriq

Theoriq aims to become a modular, composable foundational layer for AI Agents, enhancing communication and interoperability between AI Agents, ensuring they are not only interconnected but also more autonomous and powerful than ever before. Additionally, through token-based DAO governance, stakeholders can vote on proposals that impact the development of the Theoriq network, ensuring that the network evolves according to community interests and values.

Specifically, the Theoriq ecosystem consists of four roles: AI Agent developers, AI resource providers, Agent consumers, and projects.

Infinity Hub is Theoriq's AI Agent development and aggregation platform, where developers can quickly build various AI Agents using tools and connect with AI resource providers such as computing power, models, and data. Any user or project with needs can support the acquisition of AI Agent usage rights in Infinity Hub using stablecoins.

Whether developers, data providers, or users, the transparent algorithmic mechanism ensures that rewards are distributed proportionally based on the value of contributions, maintaining fairness and incentivizing meaningful participation.

Theoriq's core competitive advantages in the Web3 + Agent track are reflected in:

• Composability

Theoriq is developing a composable AI Agent platform that allows users to assemble different AI Agents together to create more advanced and flexible AI solutions.

• Incentive Mechanism

Theoriq promotes rapid innovation of AI Agents through incentive mechanisms, laying the groundwork for modular and composable AI Agents.

• Decentralized Architecture

Theoriq's Infinity Hub provides services such as model training, reasoning, and data storage, ensuring the accuracy of models through proof mechanisms, censorship resistance, immutability, and data privacy.

Opportunities and Challenges: Theoriq is incubated by the ChainML team and is an important part of ChainML's vision to become a "decentralized OpenAI GPT Store." The core development team comes from Canada and Germany, with a strong technical background and years of experience in major companies like Teradata and Vector Institute. Over the past two years, Theoriq and ChainML have raised a total of $10 million in funding, with investors including Hack VC, IOSG Ventures, Hashkey Capital, Alliance DAO, LongHash Ventures, and others.

Theoriq has accurately captured the pain points of centralization monopolies and insufficient empowerment of Web3 faced by AI Agents in their development. Major proponents of AI agents, such as Professor Andrew Ng, Vitalik Buterin, and CZ, have shown interest in the project's Twitter account. Similar to Spectral, Theoriq is also a project serving "AI for Web3," hoping to establish a calling and economic system for using AI Agents in Web3 through the Agentic Protocol, with growth flywheels constrained by the quality and quantity of Agents on the platform.

GaiaNet

GaiaNet is a decentralized computing infrastructure that enables everyone to create and deploy their own AI Agents, which can reflect their styles, values, knowledge, and expertise.

Through DAO governance, GaiaNet organically links AI Agent developers, domain name operators, token stakers, and users, forming a business closed loop where domain name operators manage AI Agent developers, token stakers provide guarantees by staking tokens on domain name operators, and users choose AI Agents from domain name operators to pay for usage with tokens.

Notably, there is also a role of component developers in the GaiaNet network, who can earn income by fine-tuning models, knowledge bases, plugins, and other components in the form of NFTs from AI Agent developers with calling needs.

GaiaNet's core competitive advantages in the Web3 + Agent track are reflected in:

• Edge Computing

GaiaNet is building a distributed edge computing node network controlled by individuals and enterprises to host fine-tuned AI models with proprietary domain knowledge and expertise. This approach enhances the diversity and specialization of AI models.

• Privacy Protection

GaiaNet's solution emphasizes protecting user privacy while providing AI capabilities. This aligns with Web3's emphasis on user data sovereignty.

• Knowledge Integration

GaiaNet allows individuals and enterprises to integrate their proprietary knowledge and skills into AI Agents, reflecting the spirit of decentralized sharing and application of knowledge in Web3.

Opportunities and Challenges: GaiaNet's node-based AI Agent creation and deployment environment aims to protect the intellectual property and data privacy of experts and users, countering the centralized OpenAI GPT Store. GaiaNet has built a complete set of decentralized AI inference usage scenarios, from front-end chatbot usage scenarios, node AI inference, fine-tuning model providers, knowledge base providers, to the underlying decentralized computing supply. GaiaNet's challenge lies in how to fully productize its grand and complex roadmap and how to open up composability with Web2 AI agents and other Web3 AI infrastructures.

Conclusion

AI Agents not only represent a significant leap in the field of AI but are also an indispensable part of the Web3 ecosystem. Based on Multi-Agent collaboration, they will jointly create a more intelligent, efficient, and decentralized world.

With the continuous emergence and development of top Web3 AI Agent projects like HajimeAI and Spectral, we witness the deep integration of AI Agents with blockchain technology, as well as their potential in driving industry progress, optimizing user experiences, lowering participation barriers, and fostering innovative applications. This not only provides developers and users with rich choices but also brings unprecedented vitality and possibilities to the entire Web3.

A Web3 AI Summer, akin to the DeFi Summer, is approaching, and AI Agents are endowing it with infinite possibilities. Let us wait and see.