Variant Fund investment partner: AI Agent has become the "first-class citizen" of the on-chain economy

Author: Mason Nystrom

Compilation: DeepTechFlow

Robots are becoming the core participants in the encryption economy.

Evidence of this trend is everywhere. For example, searchers deploy robots (such as Jaredfromsubway.eth) to profit from their Decentralization exchange (DEX) transactions by front running on the demand for convenience of human users. Tools like Banana Gun and Maestro allow users to easily conduct robot-supported transactions on the Telegram platform, which have long occupied the top of the gas consumption chart on the ETH network. In addition, robots have quickly intervened in emerging social applications like Friendtech after gaining initial acceptance from human users, and may unintentionally accelerate the speculative cycle of the market.

Overall, whether it is for profit (such as MEV robots, MEV stands for "maximal extractable value") or for ordinary users (such as Telegram bot toolkit), robots are gradually becoming the preferred users on the blockchain.

Although the current robot functionality in the encryption field is relatively simple, with the development of large language models (LLMs), robots outside the encryption field have evolved into powerful AI agents, aiming to autonomously handle complex tasks and make wise decisions.

Building these AI agents in the encryption native environment has several important advantages:

Built-in payment functionality: AI intelligences can exist off the blockchain, but to perform complex tasks, they need access to funds. Compared to traditional methods like bank accounts or payment processors like Stripe, the encryption payment system is more efficient in providing financial support to AI intelligences and avoids the various inefficiencies commonly found in the off-chain world.

Wallet Ownership: Through Wallet connections, AI agents can own digital assets (such as Non-fungible Tokens or income), and thus enjoy the inherent digital property rights of encryption assets. This is particularly important for asset transactions between agents.

Verifiable deterministic operations: The verifiability of operations is crucial when AI agents are performing tasks. On-chain transactions are essentially deterministic - either completed or not completed - this characteristic enables AI agents to more accurately complete on-chain tasks, while off-chain tasks are difficult to achieve the same level of determinism.

Of course, there are also some limitations to the on-chain AI intelligent agents.

One major constraint is that the AI agent needs to perform logic off-chain to improve performance. This means that the logic and computation of the agent will be hosted off-chain, but the decision will still be executed online to ensure verifiability of the operation. In addition, AI agents can also use providers such as Modulus for zkML (zero-knowledge machine learning) to verify the authenticity of their off-chain data inputs.

Another key limitation is that the functionality of an AI agent depends on the richness of its tools. For example, if you want the agent to summarize a real-time news article, it needs web scraping tools to search the internet. If you want it to save the results as a PDF, it needs a file system. If you want it to mimic the trades of your favorite Crypto Twitter influencer, it needs access to Wallet and the ability to sign with Secret Key.

From the perspective of determinism to non-determinism, most of the tasks currently performed by encryption AI agents are deterministic tasks. This means that humans have pre-set the parameters of the task and how it is executed (such as the specific process of token exchange).

The encryption AI agent has evolved from early keeper bots, which are still widely used in Decentralized Finance and Oracle Machine services. Today, AI agents have become more sophisticated. They can not only use large language models (LLMs) for autonomous creation (like autonomous artists such as Botto), but also provide financial services for themselves through Syndicate's trading cloud. In addition, early AI agent service markets like Autonolas are gradually forming.

Currently, many cutting-edge applications are showcasing the potential of AI intelligent agents:

AI Assistant in the smart wallet: Dawn provides users with a multifunctional assistant through its DawnAI intelligence agent, which can help users send transactions, complete on-chain transactions, and provide real-time on-chain information (such as trend analysis of popular Non-fungible Tokens).

AI character in the encryption game: Parallel Alpha's latest game Colony attempts to create AI characters that can have wallets and conduct on-chain transactions, adding more interactivity to the game.

The functionality upgrade of AI agents: The ability of AI agents depends on the tools they are equipped with, and the interaction with blockchain is still in its early stages. Encryption AI agents need to have wallet functionality, fund management capabilities, permission control, integrated AI models, and the ability to interact with other agents. Gnosis demonstrates the prototype of such infrastructure, for example, their AI mechs, which encapsulate AI scripts into smart contracts, allowing anyone (including other robots) to call smart contracts to perform tasks (such as participating in prediction market betting) and also to pay rewards to the agents.

Advanced AI traders: Decentralized Finance super app provides more efficient ways for traders and speculators, such as: automatically Auto-Invest (DCA) when conditions are met; automatically execute trades when gas fees are below a certain threshold; monitor new issuance of Meme Token contracts; and intelligently select the optimal order routing without the need for users to manually find access points.

Vertical applications of AI agents: While large models like ChatGPT are suitable for some general conversation scenarios, in order to meet the needs of different industries and niche fields, AI agents need to be specifically fine-tuned. Platforms like Bittensor incentivize developers to train models focused on specific tasks (such as image generation, predictive modeling) through incentive mechanisms, with target industries including encryption, biotechnology, and academic research. Despite Bittensor being in its early stages, developers have already begun to utilize it to build applications and agents based on Open Source large language models.

AI NPCs in Consumer Applications: Non-player characters (NPCs) are common in large multiplayer online games (MMORPGs), but less so in consumer applications. However, due to the financial nature of consumer applications, AI agents can be ideal participants in innovative game mechanisms. For example, Ritual, an open AI infrastructure company, recently released Frenrug, an intelligent agent based on a large language model that runs on the Friend.tech platform. It can automatically execute transactions (such as buying or selling Secret Keys) based on the content of user messages. Friend.tech users can try to convince this intelligent agent to buy their Secret Keys, sell other people's Secret Keys, or even find creative ways to use its funds.

As more and more applications and protocols begin to introduce AI intelligent agents, humans will use them as a bridge to enter the encryption economy. Although today's AI intelligent agents still look like "toys," in the future, they will greatly enhance users' daily experience, become key stakeholders in blockchain protocols, and even form a complete economic ecosystem among intelligent agents.

AI agents are still in the early stages of development, but as the core participants in the on-chain economy, they are just beginning to show their potential.