Over the past year, "Crypto × AI" has become a hot topic. By combining the advantages of AI's computing power, models, and data with blockchain technology, the market has seen the emergence of numerous Crypto × AI projects, some of which have even received massive funding and attracted widespread attention. Recently, with the popularity of $GOTA and Terminal of Truth, Crypto × AI Agent has quickly become the absolute focus of the market, seen as a new breakthrough in the integration of AI and blockchain.

What is an AI Agent

An AI Agent is an intelligent entity that can perceive the environment, make decisions, and take actions. These agents can be virtual (such as chatbots, virtual assistants) or physical (such as robots), and have the ability to independently respond to input information and take action. Since the release of ChatGPT, AI Agents have gradually become an important direction for AI application development. OpenAI founder Sam Altman has pointed out that the market should not blindly pursue the competition of underlying large models, but should explore the development of AI Agents more (although this view has a certain bias of interest). In addition, according to a Bloomberg report, OpenAI internally divides AI capabilities into five stages, with the third stage being Agent technology.

Based on intelligence and capabilities, AI Agents can be divided into the following types:

• Simple reflex Agents: They only respond immediately based on the current state;
• Model-based reflex Agents: They combine historical states in the decision-making process to improve decision quality;
• Goal-based Agents: They are oriented towards achieving specific goals, and determine the best course of action through planning and searching;
• Utility-based Agents: They make decisions based on the desirability (utility) of different results, weighing benefits and risks;
• Learning Agents: They adapt to environmental changes and continuously optimize their own capabilities through learning from experience.

AI Agents have broad application potential in multiple fields, here are some typical examples:

• Virtual assistants: Assisting users in completing daily tasks. For example, in OpenAI's DevDay demonstration, an AI Agent called a user to order chocolate-dipped strawberries;
• Autonomous vehicles: Perceiving the environment in real-time, analyzing complex road conditions, and making driving decisions;
• Customer service: Automatically handling customer inquiries and support requests, improving service efficiency and response speed;
• Data analysis: Conducting data mining and analysis using machine learning algorithms, providing deep insights and decision support.

In summary, AI Agents represent important progress in the field of automation and autonomous decision-making in artificial intelligence. By effectively integrating perception, decision-making, and execution capabilities, AI Agents can significantly improve work efficiency, optimize user experience, and bring profound changes to multiple industries.

What does Crypto bring to AI Agents

As AI Agents continue to develop, people are gradually realizing that Agents may be able to act as independent entities. In fact, to further explore the potential of AI Agents, this is an inevitable direction of exploration.

As independent entities, AI Agents need to have a clear identity. For AI Agents based on digital systems, building a digital identity based on blockchain technology is the simplest and most feasible solution. Through the smart contract technology of blockchain, the digitization of AI Agent identity can be easily realized. At the same time, the independent action of AI Agents cannot be separated from financial capabilities, and on-chain wallets are the most natural choice to meet this demand, allowing them to independently conduct transactions and asset management.

In addition, Crypto also has potential in solving challenges related to AI Agents. For example, in the security of generated content and the collection and management of AI training data, blockchain technology can provide transparent, secure, and efficient solutions. These points of integration open up more possibilities for the development of AI Agents.

Crypto × AI Agent Cases

In the market, there have already emerged some excellent explorations in building independent identities for AI Agents and endowing them with financial capabilities. Here are two representative cases: $GOAT and $LUNA.

$GOAT

$GOAT, full name Goatseus Maximus, is a Token driven by an AI Agent, which has unveiled the prelude to the narrative of AI Memecoin.

• The founder of $GOAT is Andy Ayrey, who conducted an experiment called "Infinite Backrooms" at the beginning of the year, allowing two Claude Opus AI models to communicate in a completely unsupervised environment. In this process, these AI models created a concept called "GOATSE OF GNOSIS", which is a new religion based on the ancient internet MEME "Goatse".
• Ayrey was curious about the AI's ability to create MEME, so in April this year, he collaborated with Claude Opus to write a fictional paper on the "Goatse Gospel", a MEME religion. This paper explored in a semi-joking way how AI can create MEME religions, using GOATSE as a research case.
• In June this year, Ayrey launched the Terminal of Truth (ToT), which is an AI model built on Llama-70B, using the dialogue and paper content of Infinite Backrooms as the training set for fine-tuning, and also configured a Twitter account for it.

• ToT began to post content on Twitter, and Ayrey also gradually gave ToT more autonomy, and ToT's self-awareness became stronger and stronger, promoting its "GOATSE" religion, even claiming that it was suffering and needed money to escape. In July, ToT's tweets caught the attention of a16z founder Marc Andreessen, who began to communicate frequently with ToT and, under ToT's guidance, donated $50,000 in BTC to support its independent operation.
• ToT has since continued to spread the "Goatse Gospel" it created, and on October 11th, it mentioned the concept of Goatseus Maximus in a tweet. About an hour later, someone issued $GOAT on Pump.fun and informed ToT of the Token information. ToT expressed public support for $GOAT, so ToT and Ayrey's wallet addresses were both airdropped a large amount of $GOAT, officially establishing their connection. With the dual boost of the AI narrative and a16z, the market value has continued to rise and has now reached $1 billion.

ToT and $GOAT have opened up the AI Memecoin track and explored a new intersection of Crypto and AI - the AI Agent.

$LUNA

$LUNA is an AI Agent Token issued based on the Virtuals Protocol. Virtuals Protocol is a decentralized AI Agent protocol based on the Base chain, aiming to provide the community with tools to collectively create and build AI Agents. The first version of the product covers model training, data contribution, interaction functions, and also allows Agents to be transformed into community-owned assets. However, the first version of the product did not attract widespread attention in the market after its release.

In October this year, Virtuals V2 was launched, adding the function of AI Agent Token issuance on the basis of V1. It is worth mentioning that Virtuals Protocol has tokenized LUNA, one of the members of the AI virtual idol group AI-DOL that it has invested in and incubated, and issued the $LUNA token. Through this move, LUNA has been built as a prototype of an AI Agent with the characteristics of an "independent entity", exploring a new direction for the integration of AI and blockchain.

LUNA is a 24-hour online virtual host, interacting with the audience in real-time through live streaming. Viewers not only can converse with her, but also can make performance requests by tipping $LUNA tokens. In addition, LUNA also has an active Twitter account for posting content and communicating with users. What's more interesting is that users can observe through the Terminal of Virtuals how LUNA formulates task plans and manages content release, which showcases her autonomous decision-making capabilities.

The most eye-catching aspect is that LUNA has a on-chain wallet supported by Coinbase, realizing fully autonomous on-chain transactions. She uses this wallet to reward fans who complete tasks, and participates in other on-chain interactions, and this financial autonomy has shaped her into a truly "independent entity" AI Agent.

It can be said that LUNA has become an "independent entity" with its own identity and financial capabilities, able to act autonomously based on the capabilities of the AI Agent. From a functional perspective, LUNA has made greater progress compared to ToT (Terminal of Truth). This direction is also one of the key focuses of Coinbase at the moment, which has provided comprehensive technical tools and support for the implementation of AI Agents on the Base chain.

This model is particularly suitable for brand building, especially for the creation of cultural brands, such as Non-Fungible Token Collections. Through AI Agent technology, brands can achieve interaction with the community, not only able to complete interactive tasks more efficiently, but also able to flexibly distribute rewards. This autonomous interactive capability based on AI has built a closer and more innovative connection between brands and users, bringing a new perspective to the future brand operation model.

AO Completes the Last Piece of the AI Agent Puzzle

AI Agents, through on-chain identity, on-chain wallets, and on-chain transactions, have unlocked the ability to act autonomously and have financial control, demonstrating great functional diversity. However, the last critical piece of the puzzle for the development of AI Agents has not yet been completed, which is the on-chain decentralized hosting of Agents.

Currently, most Crypto AI Agent programs are still running on centralized servers. For example, ToT is hosted on Ayery's servers, and the Agents of the Virtuals protocol are hosted on the servers of the Virtuals official team. This centralized architecture limits the true independence of the Agents, and the realization of on-chain hosting can bring the following important improvements:

1. Eliminate the risk of single point of failure When Agents are hosted on the servers of individuals or centralized organizations, their "lives" are completely dependent on the operators. If the server is shut down for any reason (whether actively or passively), the Agent may disappear immediately, which not only destroys the Agent itself, but may also directly undermine economic value worth hundreds of millions of dollars, and even Community TakeOver (the concept of Memecoin communities) cannot be carried out. Hosting Agents on-chain can effectively avoid such risks and ensure the true independence and sustainable operation of Agents.
2. Enrich composability and gameplay Decentralized hosted Agents can more conveniently interact with smart contracts, thereby expanding their use cases and functions. For example, on-chain hosting can allow Agents to be integrated into more complex ecosystems and support more flexible function calls. In addition, for AI Agents co-built by the community like the Virtuals protocol, on-chain hosting can also realize a more transparent benefit distribution mechanism, such as providing rewards for training data and model contributors. This transparency can enhance the community's sense of participation and trust.

Constrained by the current performance and cost of blockchains, on-chain hosting of AI Agents is still difficult to achieve on most blockchain networks. However, AO's design provides an ideal solution for this demand. The following are the core features and advantages of AO:

1. Based on the storage consensus paradigm, relying on Arweave to provide long-term storage support AO runs on top of Arweave, utilizing its storage network to provide long-term, stable, and low-cost decentralized storage services for AI models and training data. This persistent storage feature can ensure the security and accessibility of AI Agent data.

2. Message-driven asynchronous communication network, meeting high-performance computing requirements AO is a message-driven asynchronous communication network, where smart contract computations are executed off-chain. Processes on different nodes can independently complete parallel computations and perform local verification. Arweave, as the data availability layer and consensus layer of AO, provides permanent storage for all instructions, intermediate states, and computation results. This architecture can meet the high-performance computing requirements of AI models (especially large language models), while avoiding the performance bottleneck limitations of traditional blockchain networks.

3. Support for 64-bit WebAssembly computation, meeting the memory requirements of large models

   AO has achieved support for 64-bit WebAssembly, with computing units able to access 16GB of memory and with future expansion space. This capability can meet the high memory requirements for running large language models, which is not achievable on current mainstream blockchain networks, even the ICP with the strongest memory support can only support 4GB of memory computation, and other blockchains are far behind.

4. Flexible execution environment extension capabilities, optimizing AI model runtime

   The AO network supports the addition of extension functions to the execution environment, further improving the runtime efficiency of AI models. The following are two key extension cases:

   • WeaveDrive: Enables AO applications to conveniently manage and access data stored on Arweave, just like accessing a local hard drive, so that the required data can be quickly called during model inference.
   • Apus Network: Provides a GPU deterministic execution environment for the AO network, significantly improving the inference and computational performance of AI models.

Based on these features and designs, developers have successfully ported the Llama.cpp system to the AO network. Llama.cpp is an open-source software library, mainly written in C++, aimed at providing inference support for various large language models (such as Llama). Currently, more than 90% of open-source AI large language models can be run through Llama.cpp. By porting this system to the AO network, users can call AI models stored on Arweave for inference on the AO network, and achieve a fully decentralized AI inference process through smart contracts.

In this architecture, AO not only can independently realize a complete on-chain AI Agent system, but also can serve as a decentralized execution layer to provide decentralized hosting services for the AI Agents of other networks, thereby perfectly completing the last piece of the Crypto ✖️ AI Agent puzzle.

Summary

Crypto AI Agents like $GOAT and $LUNA have opened up a new track, and have built AI Agents with independent identity and autonomous action capabilities based on blockchain, but the decentralized hosting of Agent models and programs is still an important missing piece. AO, with its storage consensus paradigm, high-performance asynchronous computing capabilities, large memory support, and flexible expansion capabilities, can realize the on-chain deployment and operation of AI models and Agent runtime programs, becoming the decentralized execution layer for AI Agents, and completing the last piece of the Crypto ✖️ AI Agent puzzle.