AgentFi: A New Paradigm for Stablecoin Yield

In the context of increasingly diverse stablecoin yield strategies, but with users' attention and execution efficiency limited, AgentFi serves as the next paradigm of "AI + DeFi" integration, providing DeFi users with an intelligent agent solution that requires no monitoring and automatically optimizes.

1. Classification and Comparison of DeFAI and AgentFi

DeFAI and AgentFi, although both are in the "AI + DeFi" crossover field, have fundamentally different technological essences, product forms, and paradigm missions. AgentFi distinguishes itself from traditional DeFi or generalized DeFAI projects; its core feature is that the AI Agent is the main executor and decision-making entity, fully automated in decision-making and closed-loop execution, rather than simply using AI as a decision-support tool for data analysis and strategy enhancement.

DeFAI Category: Tools that use AI to assist or enhance DeFi-related functions, such as strategy optimization, data analysis, investment advice, etc., with AI models assisting human decision-making. These financial automation products rely on clear user instructions and strategies, executing a process of "from instruction to operation."

AgentFi Category: Building a financial system where on-chain AI entities serve as the primary executors. Agents can autonomously reason, make decisions, and complete transactions or governance operations. Users authorize the agent to act on their behalf, and the intelligent entity makes decisions and executes them proactively. Using the "from intent to result" paradigm, users only need to express their goals, and the system can formulate and execute strategies.

Currently, the market is flooded with a large number of Agent launch platforms or trading markets. Compared with AI Meme tokens that are purely speculative and have no practical use, there are very few AgentFi projects that truly focus on the Crypto Native Defi world. Through extensive online research and screening, after eliminating projects that are too early or obviously air, we have made the following simple classification and rating of the AgentFi projects currently on the market:

Stablecoin yield optimization type:

• ARMA (Giza): ★★★★★
• Orb (Symmetry): ★★★

Strategy Combination/Liquidity Mining Class:

• Odos: ★★★★
• DEXAI: ★★★
• Perennial: ★★★

On-chain transaction execution type:

• Instadapp: ★★★★
• Shapeshift: ★★★
• DefiLlama: ★★★
• Frantec: ★★

The above 9 AgentFi projects are all at an initial stage of having on-chain execution loops, clear scenarios, and basic products that are somewhat usable. However, to be honest, the AgentFi space is still too early, and the experience of most products is hard to describe. Compared to mature Defi protocols, there is still a significant gap in product quality. We look forward to the future of the AgentFi space being able to move beyond the meme token speculation phase and enter a true PMF stage.

2. Stablecoin Yield Optimization in AgentFi

The track of stablecoin yield optimization is not new in the Defi world. From the earliest automated yield aggregators and automated asset rebalancing and strategy combination platforms, to professional DeFi risk modeling and protocol-level yield optimization platforms aimed at institutions, they all belong to relatively representative automated yield optimization projects before AgentFi.

AgentFi has initially realized the vision of mutual integration between Crypto and AI: that is, AI empowers users to better participate in the native DeFi world of Crypto through on-chain information collection and decision-making capabilities, while utilizing the immutability of blockchain to retain transaction details in an on-chain log format, ensuring logical verifiability and traceable behavior. As a continuation of the stablecoin yield article, we focus on discussing the product usability aspect of the AgentFi track, which is uniquely rated five stars and is the only asset in the "stablecoin yield optimization" category------ARMA.

III. Giza Project Vision and Paradigm Concept

Giza is a blockchain infrastructure project focused on building "Verifiable AI Agents," dedicated to promoting the emerging paradigm of Agent-native Finance. As one of the few project teams in the ZKML track, Giza differs from teams that lean towards theoretical research or infrastructure development by placing greater emphasis on engineering and practicality. It is the first project in the ZKML track to achieve a complete closed loop of "on-chain smart agents + verifiable execution + AI scheduling."

Giza has proposed a groundbreaking concept—Xenocognitive Finance, which introduces AI agents as a new interface in decentralized finance to address the cognitive bottlenecks of human participants and reconstruct market intelligence and structure. The agents are not passive tools but "cognitive representatives" with independent reasoning capabilities and on-chain execution permissions. By enhancing the information processing capacity of the entire market through a distributed network of agents, market intelligence expands with the scale of participation rather than collapsing, and provides credible results through zero-knowledge proofs (ZK). In short, Xenocognitive Finance is a new paradigm for efficiently participating in decentralized finance through autonomous intelligent agents that extend human cognitive abilities, achieving "cognitive offload" without sacrificing sovereignty in a non-custodial model, and eliminating the need for constant monitoring and automatic protocol switching.

The architecture of the Giza Protocol includes the following three layers:

• Semantic Abstraction Layer: Bridging the semantic gap between AI and blockchain, helping AI understand the blockchain world.
• Decentralized Execution Layer: Provides a scalable, secure, and censorship-resistant agent execution environment.
• Agent Authorization Layer: Ensures that agents operate in a non-custodial, secure, and controllable manner.

The agent execution process of Giza is collaboratively completed by three core modules and is divided into four key steps:

1. Request Processing (Semantic Abstraction Layer): The AI agent initiates an operation request through the MCP protocol (e.g., "Deposit USDC into a certain platform"), the system parses the natural language strategy into standardized on-chain operation instructions, while verifying the format, semantics, and permission boundaries to ensure the request is legal and executable;

2. Protocol Interaction (Decentralized Execution Layer): Execution nodes perform actual operations by calling DeFi protocols (such as lending and exchanging) based on instructions. The system will dynamically optimize the trading path, considering gas costs, slippage tolerance, and protocol liquidity. After task execution, results and task proofs are generated for subsequent verification.

3. Execution Verification (Authorization Layer + Execution Layer): The system checks whether the operation complies with the Session Key authorization scope set by the user (protocol type, amount, time). Multiple Attester nodes independently verify the operation, and the Aggregator node collects signatures to reach consensus, ensuring that the result is secure and trustworthy.

4. Result feedback (semantic abstraction layer): The execution result is transformed into structured semantic information (such as changes in revenue, asset distribution) and returned to the AI agent system for subsequent strategy adjustments and automated decision-making.

In July 2023, Giza announced the completion of a $3 million Pre-Seed round of financing, led by an investment institution with participation from several other institutions. In May 2025, Giza completed a $2.2 million seed round of financing, led by a certain fund with participation from several other institutions. These two rounds of financing demonstrate the capital market's ongoing optimism about Giza's vision in the "AI + DeFi" sector.

IV. Representative Product ARMA: Stablecoin Yield Optimization Agent

ARMA (Autonomous Revenue Management Agent) is a stablecoin yield optimization smart agent launched by Giza, designed to achieve automated yield maximization for users across protocols. It is currently deployed on the Base network and plans to expand to more Layer 2 ecosystems, supporting multiple mainstream lending protocols.

( 4.1 Overview of ARMA Working Mechanism

• Intelligent Strategy Scheduling: Real-time analysis of the stablecoin yield (APR), trading costs, and rebalancing timing of various protocols, automatically executing optimal rebalancing operations.
• APR Optimization System: Through comprehensive optimization of interest rate comparison, cost assessment, and rebalancing logic, it ensures that each operation brings net profit; the actual annualized return can reach up to 2 times that of traditional static holdings.
• Automatic compound interest mechanism: Automatically receive and reinvest incentive tokens, converting rewards into the original stablecoin; the compounding frequency is dynamically optimized based on position size and transaction costs to enhance efficiency.
• Smart Coin Exchange Logic: Supports dual currency operations with USDC and USDT, automatically and efficiently exchanging coins through integrated DEX when necessary, ensuring users can always withdraw assets in the original currency, enhancing flexibility and experience.

![AgentFi to XenoFi, a new paradigm for stablecoin yields])https://img-cdn.gateio.im/webp-social/moments-86b52246444d196d16553ad20f01c05c.webp###

( Overview of 4.2 ARMA Component Module and Security Architecture

The Giza Agent architecture consists of six modules: Smart Accounts, Session Keys, Agent Core, Protocol Integration, Risk Module, and Accounting Module. These modules work together to build a secure, efficient, intelligent, and verifiable on-chain automation execution system.

At the same time, ARMA ensures the safety of user funds and operations through a three-layer protection mechanism:

• Smart Account Security: Based on self-custody smart accounts, users always maintain control over their assets, and Giza cannot access private keys; contracts are audited and support flexible permission management.
• Access control mechanism: Fine-tune proxy permissions through Session Key, limited to specific protocols, operation types, and validity periods, avoiding the risk of full authorization.
• Risk management system: Only connect to prudently selected protocols, focusing on stable income scenarios; the entire trading process is traceable, ensuring operational transparency and auditability, enhancing overall system security.

![AgentFi to XenoFi, a new paradigm for stablecoin yields])https://img-cdn.gateio.im/webp-social/moments-de27123568f85c35bb7e8af287749188.webp###

( 4.3 ARMA Fee Mechanism and Revenue Management

The fee and revenue mechanism of ARMA is centered around user-friendliness, transparency, and efficiency, covering three main aspects:

• Fee Structure: Charged based on earnings, transparent and verifiable, only a 10% success fee is charged on the actual earnings generated. This fee is settled in one lump sum when the user withdraws, and all calculation processes for the fees can be viewed in real time through the dashboard.
• Reward Management: Automatic collection, unified distribution. The ARMA intelligent agent automatically collects and aggregates all rewards from various integrated DeFi protocols. All rewards will be returned to users along with the principal and profits upon withdrawal without manual operation by the user.
• Asset Management: ARMA does not charge any deposit or withdrawal fees. Regardless of how many protocols are involved in the process and what type of tokens are used, users will always withdraw their funds in the original tokens deposited, ensuring the convenience of fund recovery.

![AgentFi to XenoFi, a new paradigm of stablecoin yields])https://img-cdn.gateio.im/webp-social/moments-edf4d625712f07f4ac7cd0551bea721e.webp###

( 4.4 ARMA's integration with DeFi protocols and data performance

Through ARMA's official data dashboard, we can visually observe the continuous growth of its asset management scale and the number of smart agents in the months since its launch. At the same time, the dashboard also displays the dynamic flow of assets among multiple mainstream lending protocols. We believe that as ARMA connects to more DeFi protocols, the number of active agents increases, and the scale of funds continues to expand, it will further strengthen its positive growth flywheel effect.

![Agent