Author:Mason Nystrom
Compiled by: TechFlow
Robots are becoming core participants in the crypto economy.
Evidence of this trend is ubiquitous. For example, searchers deploy bots (like Jaredfromsubway.eth) to exploit human users' need for convenience by front-running trades on their decentralized exchanges (DEXs). Tools like Banana Gun and Maestro allow users to conveniently conduct bot-supported trades via the Telegram platform, consistently ranking among the top "gas guzzlers" on Ethereum. Furthermore, in emerging social apps like Friendtech, bots quickly intervene after initial human user adoption, potentially inadvertently accelerating the market's speculative cycles.
Overall, whether for profit (like MEV bots, MEV being "maximum extractable value") or for the general user (like Telegram bot toolkits), robots are gradually becoming the priority users on the blockchain.
While the functionality of crypto bots is still relatively simple at the moment, with the advancement of large language models (LLMs), bots outside the crypto realm have already evolved into powerful AI agents capable of autonomously handling complex tasks and making informed decisions.
Building these AI agents natively in the crypto environment offers several key advantages:
Built-in payment capabilities: AI agents can exist outside the crypto realm, but if they are to execute complex operations, they must have the ability to acquire funds. Compared to traditional methods (like bank accounts or payment processors like Stripe), crypto payment systems are more efficient in providing funding support for AI agents, while avoiding the various inefficiencies common in the off-chain world.
Wallet ownership: Through wallet connections, AI agents can own digital assets (like Non-Fungible Tokens or yield), enjoying the inherent digital property rights of crypto assets. This is particularly important for asset transactions between agents.
Verifiable deterministic operations: Verifiability of operations is crucial when AI agents execute tasks. On-chain transactions are inherently deterministic - either completed or not - a feature that allows AI agents to perform on-chain tasks more accurately, in contrast to the difficulty of achieving the same level of determinism off-chain.
Of course, on-chain AI agents also face some limitations.
A major constraint is that AI agents need to execute logic off-chain to improve performance. This means that the agents' logic and computation will be hosted off-chain, while the decision-making still occurs on-chain to ensure the verifiability of operations. Additionally, AI agents can leverage providers of zkML (zero-knowledge machine learning) like Modulus to verify the authenticity of their off-chain data inputs.
Another key limitation is that the functionality of AI agents depends on the richness of their toolsets. For example, if you want an agent to summarize a real-time news article, it needs to have web crawling tools to search the internet. If you want it to save the result as a PDF, it needs file system capabilities. If you want it to mimic the trading of your favorite Crypto Twitter influencer, it needs wallet access and key signing functionality.
From a deterministic to non-deterministic perspective, most crypto AI agents currently perform deterministic tasks. This means that humans have pre-defined the parameters of the task and how it should be executed (e.g., the specific flow of a Token swap).
Crypto AI agents have evolved from early keeper bots, which are still widely used in DeFi and oracle services. Today, AI agents have become more sophisticated. They can not only leverage large language models (LLMs) to achieve autonomous creation (like the self-directed artist Botto), but also provide themselves with financial services through Syndicate's transaction cloud. Additionally, early AI agent service marketplaces like Autonolas are gradually taking shape.
Currently, many frontier applications are showcasing the potential of AI agents:
AI assistants in smart wallets: Dawn, through its DawnAI agent, provides users with a multi-functional assistant that can help them send transactions, complete on-chain transactions, and provide real-time on-chain information (such as trend analysis of popular Non-Fungible Tokens).
AI characters in crypto games: Parallel Alpha's latest game Colony aims to create AI characters that can own wallets and conduct on-chain transactions, adding more interactivity to the game.
Upgrading AI agent capabilities: The capabilities of AI agents depend on the tools they are equipped with, and their interaction with blockchains is still at a basic stage. Crypto AI agents need to have wallet functionality, fund management capabilities, permission controls, integrated AI models, and the ability to interact with other agents. Gnosis has showcased the prototypes of such infrastructure, such as their AI mechs, which encapsulate AI scripts into smart contracts, allowing anyone (including other bots) to call the smart contract to execute tasks (like participating in prediction market bets) and be compensated for their work.
Advanced AI traders: DeFi super-apps provide traders and speculators with more efficient ways of operating, such as: automatically dollar-cost averaging (DCA) when conditions are met; automatically executing trades when gas fees are below a certain threshold; monitoring newly launched MEME Token contracts; and intelligently selecting the optimal order routing, without the user having to manually find the entry points.
Vertical applications of AI agents: While large models like ChatGPT are suitable for some general conversational scenarios, AI agents need to be fine-tuned to meet the specific needs of different industries and niches. Platforms like Bittensor use incentive mechanisms to encourage developers to train models focused on specific tasks (like image generation, predictive modeling) for target industries including crypto, biotech, and academic research. Although Bittensor is still in its early stages, developers have already started leveraging it to build applications and agents based on open-source large language models.
