Original author: IOSG Ventures
In the past two months, AI Agents x Crypto has sparked a wave of enthusiasm. Memecoins, interactive agents, and the openness of social media to bot accounts have combined to drive a wave of agent-driven agent hype, creating significant heat on Twitter and Farcaster. This proves the PMF of AI Agents x Crypto. The market cap of Agent-related assets has reached $10 billion.
Since the birth of goat in October, Agent has created numerous new projects and assets driven by the market. Combined with the outlook for the future, this article roughly summarizes the following framework:
Source: IOSG Ventures
1. Sentient Memecoins
Cult memecoin, which has rapidly risen under Murads call, is an asset with community and spread as the core narrative. Memecoin represented by Agent has added the Sentient factor, which has created an advantage in content. In addition, the novelty of AI narrative and the appropriate participation threshold have brought a new wave of momentum to asset issuance. The advantages in content are:
Continuity of content creation: AIGC can continuously create content 24/7
Content quality: With the current LLM support, the quality is relatively high after fine-tuning the corpus with good meme effects such as 4chan.
We have AI-created concepts, scientific research-themed topics, AI ethics, and even religious content, as well as digital twins created for celebrities. These memecoins created hype in the short term and drove the development of the entire track. However, pure AI memecoins are now clearly lacking in stamina. The reason is that there are not enough new concepts and targets that can be derived to stimulate the market.
The advantage in content will make such a sentient meme a persistent form of memecoin. More celebrities will surely get involved in it in the future, but it will be difficult for targets that are eye-catching enough to appear.
In addition to pure memecoin, many AI content creations based on conversation, audio, and video have also been born. The essence is to bring AIGC into content with crypto attributes. It is also a way to make memes more concrete and provide customizable experiences.
2. Autonomous Agent Network
2.1 Why Autonomous?
Decentralizing the entire AI stack is a long-term task. But decentralizing the agent stack is a relatively simple starting point. The model itself is the brain of the agent, but the autonomous chain constitutes the heartbeat of the agent. Giving the agent autonomy gives agents the guarantee to fully participate in on-chain activities. Opening the Pandoras box of the sovereign agent is also a very memetic thing in itself, and this can only happen on the blockchain.
The agents currently running cannot be said to be autonomous, or we cannot verify whether they are autonomous. Autonomous means that the agent needs to be completely sovereign from the hosting of the inference model, its behavior, especially the manipulation of data input and output, the control of social media accounts, the control of assets, and even the hardware. The operation of the agent itself consumes computing resources and on-chain resources, so it also needs to have its own way to generate profits and continue to operate. The ultimate endgame should be that once the agent is created, it can run forever on the blockchain and can be verified to be autonomous.
Autonomous agents also have the legitimacy of owning their own memecoins. That is, they obtain their first funds by issuing their own memecoins and use them for their own economic activities. After handing over the funds to the autonomous agent, they will not be manipulated by humans. For example, Truth Terminal has never sold $Goat, and Pet rock even lost control of the funds after the reboot.
Source: Twitter
In terms of improving autonomous capabilities, TEE technology from Phala and others is used to provide a trusted operating environment. Although the current hardware is not sufficient to support large parameter LLMs, it can still support small open source LLMs and control of social media accounts. For model hosting, decentralized cloud hosting such as Hyperbolic is a solution. It is foreseeable that more aspects of the agent will be solved by a decentralized service stack, which is what we have been building.
2.2 Agent Framework
In less than two months, many open source and extremely convenient agent frameworks have been born as platform products for creating agents and agent assets. Product forms include open source frameworks, closed source APIs, platform integration, etc. Among the more well-known frameworks, only the Eliza framework is open source.
The current agents are relatively simple and their functions are not so capital-oriented, so the requirements for open source verifiability are not high. There are many platforms that provide agent services directly in the form of launchpad. Such platforms are better at integrating tokenomics and providing users with relatively simple and practical services. In terms of functions, we can see that they are still mainly Reply bots and Digital twins of celebrities/KOLs. But there is no shortage of agents that provide more diverse services after secondary development, such as issuing tokens, analyzing tokens, analyzing mindshare, etc. The ability of such agents to read and write social data and blockchain data will be the focus of future development, which I will mention again in subsequent chapters.
However, from the perspective of future use cases, open source is a better path in the long run. The Eliza framework has attracted a large number of developers in just two months, almost exceeding the total attention of all previous Crypto AI open source frameworks. It ranks among the top in the entire github trending list, with a large number of OG developers participating, even exceeding the attractiveness of most public chains to developers. With the in-depth and diversified development of agent services, the future of Agent framework towards open source framework is quite promising.
Source: AI16Z
2.3 Swarm Agent Framework
Similar to the development path of existing agents in web2, when people are no longer satisfied with the capabilities of a single agent, the demand for swarm agents naturally emerges. Since tasks in the real world are very complex, single agents often cannot perform all tasks. For example, creating a song requires different capabilities such as lyric writing, composing, arranging, and art design.
If you want agents, especially agents under different frameworks, to work together in a swarm mode, you still need to create a framework to act as a task manager to support communication between agents, dynamic task allocation, resource sharing, and cross-platform cooperation. In crypto, the economic layer between agents is more natural and more important, and as agents themselves iteratively develop and tasks evolve, the scalability of the Swarm framework is also critical.
There are already many AI x Crypto projects working in this direction, such as Theoriq. The next important step is how to combine these already built infrastructures with the agent frameworks that are used more frequently on the chain. We see that some protocols such as FXN are working in this direction.
2.4 AI bounty for humans
We have agents serving humans, and agents serving each other, so we naturally consider whether there will be situations where humans serve agents - this is even more important when agents hold a large amount of assets and can make autonomous decisions. For Autonomous Agents, the biggest limitation is the inability to complete tasks in real life. For example, how to ensure the physical security of the hardware running the TEE? AI reversely hires humans to complete real-life tasks through the on-chain assets held by agents. We see platforms like payman building such services.
3. On-chain activities
3.1 Defi related
Asset Management
In addition to issuing memecoin, the main reason why we consider Agent as Fi is that Agent has the ability to use and manage crypto assets. The most important capabilities currently include:
Asset analysis, such as investment analysis, token analysis, and mindshare analysis. For example, a reply bot like AIXBT, anyone can @AIXBT and get an analysis of the asset. This type of bot brings a data service experience in a more user-friendly way.
Direct fund management, including Pmairca under AI16Z, the investment DAO that Vader AI wants to do, and Swarm Investment Agent such as AROK. By giving agents the ability to trade directly based on strategies, agents become investment managers who can raise funds and deploy funds according to strategies. At present, the strategies of most agents are still relatively simple (investing based on social media data, etc.), which also brings huge potential space.
Blockchain OS
Services like Graiffin, which turns the blockchain entrance into a terminal similar to a search engine, can provide intentional services through agents. Whether it is transactions, token deployment, NFT issuance, etc., they can all be solved through natural language. Such terminal services are certainly valuable, but they are somewhat inconsistent with the tone of decentralization. Services like theoriq are committed to providing agent services to users in a more permissionless way, allowing everyone to upload their own agents, combine them through the swarm framework, and package them into services for users to use.
3.2 Token/Market Release
Starting with Clanker, on Farcaster and Twitter, social media replies are used as the operating interface, and services such as issuing tokens are provided through @ these agents. In essence, the interaction with the front end is transformed into direct natural language interaction on Twitter, and platform products such as pumpdotfun are transplanted to social media platforms. In the past, asset issuance required constant interface jumps, but now all these asset issuance activities are aggregated on social media, greatly reducing the wear and tear of user jumps.
In addition to token issuance, even any prediction market, price betting market, etc. can be directly executed through this front end, bringing a new paradigm for Dapp application front end.
3.3 Gamefi related
Agent Game Character
In addition to the management of underlying assets, agents also have the ability to generate revenue. The first type of game we see is that agents raise difficult problems for humans to solve and get rewards. This type of game gives the right of adjudication to agents, allowing people to gamble around agents. This gameplay, similar to the Turing test, has brought a lot of popularity. The behavior of the referee acting as an agent is immutable after receiving the prompt, and it can act as a flexible oracle to open games and adjudicate results relatively fairly and objectively. The imagination space can even be compared to doing casino business, which is a good way for agents to generate revenue.
At the same time, a major scenario for agents in the future is to act as Autonomous virtual beings and appear as NPCs in full-chain games. Such agents are more concrete, and agents with asset management rights can participate in more economic activities than web2 NPCs, which in turn makes the virtual space more attractive. Such NPCs live in the Gamefi environment and can permanently perform certain duties, which will be an indispensable part of the on-chain world such as FOCG.
3.4 Infra Service Related
Agent blockchain service
The ultimate vision of the combination of Agent and Crypto is that Agent can become part of the blockchain consensus system. Zerebro is taking the first step. In its blueprint, agents based on the Zerebro framework and integrated with the Flashbots stack will soon become autonomous blockchain validators and earn income through block rewards and MEV. Its validator income will be recycled into the network to promote economic self-sufficiency. Furthermore, Agents can maintain multi-chain verification and governance by building their own networks, which also leaves plenty of room for imagination.
Conclusion
The recent rise of Agentfi demonstrates the huge potential of combining AI and blockchain. From the initial Sentient Memecoin to social media content creation agents, to autonomous agents, and finally to agents that live on the chain and even exist in the blockchain consensus system, AI agents are gradually expanding their influence in the crypto ecosystem.
However, compared with the current development of the open source stack, subsequent development still needs to give agents deeper autonomy and the ability to participate in on-chain economic activities. At present, some developers are allowing Autonomous Agents to promote the transformation of DeFi, gamefi, and blockchain underlying services by giving agents the ability to manage assets, make decisions, and operate on the chain, and making these economic activities a place for agents to generate income. The income-generating profits can be further invested by agents, which is why agents are expected to carry most of the transactions on the chain in the future. The issuance of memecoin assets for agents has accelerated the continuation of this development wave. We see that the market has found PMFs agent services and the development of agent infra that provides underlying support through token price evaluation, and we also see the vigorous vitality of the open source ecosystem.
AgentFis development path is becoming clearer: with open source technology and economic incentives at its core, Agent is not only a carrier of interactive entertainment, but also a key driving force for on-chain autonomy and innovation. This trend is leading crypto towards coexistence with agents, a more intelligent, autonomous and collaborative future.
