DeFAI Explained: How AI Unlocks the Potential of Decentralized Finance?

Author: Geng Kai, DF

What is DeFAI?

Since its rapid expansion in 2020, decentralized finance (DeFi) has been a central pillar of the crypto ecosystem. While many new and innovative protocols have been established, it has also led to an increase in complexity and fragmentation, making it difficult for even seasoned users to navigate a large number of chains, assets, and protocols.

At the same time, artificial intelligence (AI) has evolved from a broad foundational narrative in 2023 to a more specialized, agent-oriented focus in 2024. This shift has given rise to DeFi AI (DeFAI) — an emerging field where AI enhances DeFi through automation, risk management, and capital optimization.

DeFAI spans multiple tiers. Blockchain is the base layer because AI agents must interact with a specific chain in order to execute transactions and execute smart contracts. On top of this, the data and compute layers provide the infrastructure needed to train AI models from historical price data, market sentiment, and on-chain analytics. Privacy and verifiable layers ensure that sensitive financial data remains secure while remaining trustless execution. Finally, the proxy framework allows developers to build specialized AI-powered applications, such as autonomous trading bots, credit risk assessors, and on-chain governance optimizers.

Although this ecosystem diagram can be further expanded, these are the top categories of projects built on DeFAI.

As the DeFAI ecosystem continues to expand, the most prominent projects can be divided into three main categories:

1. Abstract Layer

Protocols built on this category serve as user-friendly interfaces similar to ChatGPT for DeFi, allowing users to input prompts for on-chain execution. They are often integrated with multiple chains and dApps, executing user intent while eliminating manual steps in complex transactions.

Some functions that these protocols can execute include:

• Exchange, cross-chain, lending/withdrawal, cross-chain execution transactions
• Copy trading wallet or Twitter/X profile
• Automatically execute take profit/stop loss trades based on position size percentage

For example, there is no need to manually withdraw ETH from Aave, bridge it to Solana, swap SOL / Fartcoin, and provide liquidity on Raydium - the abstraction layer protocol can complete the operation in just one step.

Main Protocol:

• @griffaindotcom — the proxy network for executing trades on behalf of users
• @HeyAnonai — The protocol for handling user prompts for DeFi trading and real-time insights.
• @orbitcryptoai — AI partner for DeFi interactions

2. Autonomous Trading Agent

Unlike traditional trading bots that follow preset rules, autonomous trading agents can learn and adapt to market conditions, adjusting their strategies based on new information. These agents can:

• Analyze data to continuously improve strategies
• Predict market trends to make better long/short decisions.
• Execute complex DeFi strategies like basic trading

Main Protocol:

• @Almanak__ — A platform for training, optimizing, and deploying autonomous financial agents.
• @Cod3xOrg — launched an AI agent to perform financial tasks on the blockchain
• @Spectral_Labs — Create a network of autonomous on-chain trading agents

3. AI-driven DApps

DeFi dApp offers lending, swapping, yield farming and other functions. AI and AI agents can enhance these services in the following ways:

• Optimize liquidity supply by rebalancing LP positions for better APY
• Scan tokens for risks by detecting potential rugs or honeypots.

Main Protocol:

• @gizatechxyz's ARMA — AI agent to optimize USDC earnings in Mode and Base
• @SturdyFinance - AI-driven yield vault
• @derivexyz - An options and perpetual contract platform optimized with smart AI co-pilot

Main Challenges

Top protocols built on these layers face several challenges:

1. These protocols rely on real-time data streams to achieve optimal trade execution. Poor data quality can lead to inefficient routing, trade failures, or unprofitable trades.
2. AI models rely on historical data, but the volatility of the cryptocurrency market is significant. Agents must be trained on diverse, high-quality datasets to maintain effectiveness.
3. It is necessary to have a comprehensive understanding of asset correlation, liquidity changes, and market sentiment in order to understand the overall market conditions.

Protocols based on these categories have gained popularity in the market. However, to provide better products and optimal results, they should consider integrating various datasets of different qualities to elevate their products to a new level.

Data Layer - Powering DeFAI Smart

The quality of AI depends on the data it relies on. To enable AI agents to work effectively in DeFAI, they need real-time, structured, and verifiable data. For example, the abstraction layer needs to access on-chain data through RPC and social network APIs, while trading and yield optimization agents require data to further refine their trading strategies and reallocate resources.

High-quality datasets enable agents to better predict and analyze future price behavior, providing trading suggestions that align with their preferences for long or short positions on certain assets.

The main data provider of DeFAI

| Agreement | Details | Function | | Mode Synth | Synthetic data for financial forecasting | Capturing the full distribution of price changes for AI model predictions | | Chainbase | Full-chain structured dataset | Provides AI-enhanced data for trading, forecasting, and obtaining alpha | | sqd.ai | Decentralized Data Lake for AI Agents | Scalable and customizable multi-chain data access with zero-knowledge proof security | | Cookie | AI agent-oriented CT mind and on-chain data layer | Utilizes 18 specialized AI agents to process over 7TB of on-chain agent data across more than 20 chains |

Mode Synth Subnet

As the 50th subnet of Bittensor, Synth creates synthetic data for the financial prediction capabilities of agents. Compared to other traditional price prediction systems, Synth captures the complete distribution of price changes and their associated probabilities, thereby constructing the most accurate synthetic data in the world to support agents and LLM.

Providing more high-quality datasets can enable AI agents to make better directional decisions in trading while predicting APY fluctuations under different market conditions, so that liquidity pools can reallocate or withdraw liquidity when needed. Since the launch of the Autonomous Network, there has been strong demand from DeFi teams to integrate Synth's data through their API.

Most Attention-Grabbing AI Agency Blockchain

In addition to building a data layer for AI and agents, Mode also positions itself as a full-stack blockchain for the future of DeFAI. They recently deployed Mode Terminal, which is the co-pilot for DeFAI, used to execute on-chain transactions through user prompts, and it will soon be open to $MODE stakers.

In addition, Mode also supports many AI and agent-based teams. Mode has made significant efforts to integrate protocols such as Autonolas, Giza, and Sturdy into its ecosystem, rapidly evolving as more agents are developed and transactions are executed.

These measures are all implemented while they upgrade the network with AI, most notably equipping their blockchain with an AI sorter. By using simulation and AI analysis of transactions before execution, high-risk transactions can be blocked and reviewed prior to processing, ensuring on-chain security. As an L2 of the Optimism superchain, Mode stands in the middle ground, connecting human and agent users with the best DeFi ecosystems.

Comparison of Top Blockchains Based on AI Agents

Solana and Base are undoubtedly the two main chains for building and deploying most AI agent frameworks and tokens. AI agents leverage Solana's high throughput and low-latency network, along with the open-source ElizaOS, to deploy agent tokens, while Virtuals serves as the launchpad for deploying agents on Base. Although they both have hackathons and funding incentives, they have not yet reached the level of AI initiatives that Mode has achieved.

NEAR previously defined itself as an AI-centric L1 blockchain, with features including an AI task marketplace, the NEAR AI Research Center with an open-source AI agent framework, and the NEAR AI Assistant. They recently announced a $20 million AI agent fund to expand fully autonomous and verifiable agents on NEAR.

Chainbase

Chainbase provides a fully verifiable on-chain structured dataset that can enhance the capabilities of AI agents in trading, insights, predictions, and alpha hunting. They launched manuscripts, which is a blockchain data stream framework for integrating on-chain and off-chain datasets into a target data storage for unrestricted querying and analysis.

This allows developers to customize data processing workflows according to their specific needs. Standardizing raw data and processing it into a clean, compatible format ensures that their datasets meet the stringent requirements of AI systems, thereby reducing preprocessing time while improving model accuracy, helping to create reliable AI agents.

Based on its extensive on-chain data, they have also developed a model called Theia, which translates on-chain data into user data analytics without any complex coding knowledge. The data utility of Chainbase is evident in their partnerships, where AI protocols are using their data to:

• ElizaOS proxy plugin for on-chain driven decision making
• Build Vana AI Assistant
• Flock.io social network intelligence, providing user behavior insights
• Theoriq's data analysis and predictions for DeFi
• Collaborated with 0G, Aethir, and io.net

Compared to traditional data protocols

Data protocols such as The Graph, Chainlink, and Alchemy provide data, but not in an AI-centric way. The Graph offers a platform for querying and indexing blockchain data, providing developers with access to raw data that is not built for trading or strategy execution. Chainlink provides oracle data feeds but lacks AI-optimized datasets for predictions, while Alchemy primarily offers RPC services.

In contrast, Chainbase data is specially prepared blockchain data that can be easily utilized by AI applications or agents in a more structured and insightful manner, enabling agents to conveniently access data related to on-chain markets, liquidity, and token data.

sqd.ai

sqd.ai (formerly Subsquid) is developing an open database network tailored specifically for AI agents and Web3 services. Their decentralized data lake provides permissionless, cost-effective access to a vast amount of real-time and historical blockchain data, enabling AI agents to operate more efficiently.

sqd.ai provides real-time data indexing (including indexing of unconfirmed blocks) with a speed of over 150,000+ blocks per second, faster than any other indexer. In the past 24 hours, it has provided over 11TB of data, meeting the high throughput demands of billions of autonomous AI agents and developers.

Their customizable data processing platform offers tailored data according to the needs of AI agents, while DuckDB provides efficient data retrieval for local queries. Their comprehensive dataset supports over 100 EVM and Substrate networks, including event logs and transaction details, which is highly valuable for AI agents operating across multiple blockchains.

The incorporation of zero-knowledge proofs ensures that AI agents can access and process sensitive data without compromising privacy. Additionally, sqd.ai can handle the increasing data load by adding more processing nodes, thereby supporting the growing number of AI agents (which is estimated to reach billions).

Cookie

Cookie provides a modular data layer for AI agents and clusters, specifically designed for processing social data. It features an AI agent dashboard that tracks top agent mindsets on-chain and across social platforms, and it recently launched a plug-and-play data cluster API for other AI agents to detect popular narratives and mindset shifts in CT.

Their data pool covers over 7TB of real-time on-chain and social data sources, supported by 20 data agents, providing insights into market sentiment and on-chain analytics. Their latest AI agent @agentcookiefun utilized 7% of their data pool's capacity to provide market predictions and discover new opportunities by leveraging various other agents operating underneath it.

The next step of DeFAI

Currently, most AI agents in DeFi face significant limitations in achieving full autonomy. For example:

1. The abstraction layer transforms user intent into execution, but often lacks predictive capability.
2. AI agents may generate alpha through analysis, but lack independent trade execution.
3. AI-driven dApps can handle vaults or transactions, but they are passive rather than active.

The next phase of DeFAI may focus on integrating useful data layers to develop the best proxy platform or agent. This will require in-depth on-chain data regarding whale activities, liquidity changes, etc., while generating useful synthetic data for better predictive analysis, and combining sentiment analysis from the general market, whether it be token fluctuations in specific categories (such as AI agents, DeSci, etc.) or token fluctuations on social networks.

The ultimate goal is for AI agents to seamlessly generate and execute trading strategies from a single interface. As these systems mature, we may see future DeFi traders relying on AI agents to autonomously assess, predict, and execute financial strategies with minimal human intervention.

Final thoughts

In light of the significant shrinkage of AI agent tokens and frameworks, some may view DeFAI as merely a passing fad. However, DeFAI is still in its early stages, and the potential for AI agents to enhance the usability and performance of DeFi is undeniable.

The key to unlocking this potential lies in obtaining high-quality real-time data, which will improve AI-driven trading predictions and execution. An increasing number of protocols are integrating different data layers, and data protocols are building plugins for frameworks, highlighting the importance of data in agent decision-making.

Going forward, verifiability and privacy will be the key challenges that protocols must address. Currently, most AI proxy operations remain a black box to which users have to entrust their funds. As a result, the development of verifiable AI decisions will help ensure transparency and accountability in the agent process. Integrating TEE, PHE, and even zk-proofs-based protocols can enhance the verifiability of AI agent behavior, enabling trust in autonomy.

Only by successfully combining high-quality data, robust models, and transparent decision-making processes can the DeFAI agents achieve widespread application.