Last month, we published the "Stablecoin Yield Strategy Guide," which provided a detailed overview of eight major stablecoin yield models: Lending & Borrowing, Yield Farming, Market-Neutral Arbitrage, US Treasury-Backed RWA Yields, Structured Products, Yield Tokenization, Stablecoin Baskets, and Staking-based Yields. The report examined the characteristics, yield sources, and potential risks of each model. It gained notable traction across Chinese crypto media and sparked widespread discussion within the Chinese-speaking crypto community.
Following its release, several stablecoin teams reached out to share new insights and perspectives. While most of these can still be classified under the existing eight models, one emerging trend stood out, a promising new paradigm for stablecoin yield generation: AgentFi.
While both DeFAI and AgentFi fall under the intersection of “AI + DeFi,” they differ significantly in technical essence, product architecture, and paradigm intent. AgentFi is distinct from traditional DeFi or generalized DeFAI projects in that the AI agent serves as the primary decision-maker and executor, enabling fully autonomous decision-making and closed-loop execution—rather than merely acting as an auxiliary tool for data analysis or strategy enhancement.
DeFAI (Decentralized Finance + AI) category: These are tool-based applications that use AI to assist or enhance DeFi functionalities such as strategy optimization, data analysis, or investment suggestions. The AI model serves as a co-pilot to human decision-making. Notable examples include Kaito, Numerai, and Gauntlet, which focuses on protocol-level risk and yield optimization. Such financial automation products rely on users to provide clear instructions and strategies, following an "instruction-to-execution" model.
AgentFi / XenoFi category: These are financial systems where on-chain AI agents act as autonomous actors, capable of independent reasoning, decision-making, and execution of trades or governance actions. Users delegate authority to agents, who then act on their behalf with full autonomy. By adopting an "intent-to-outcome" paradigm, users only need to express their goals, and the system handles both strategy formulation and execution. This shift significantly lowers the entry barrier, attracts a broader user base, and achieves strong product differentiation with deeper user engagement.
At present, the market is flooded with "Agent-launch" platforms or marketplaces. However, truly functional AgentFi projects that focus on the crypto-native DeFi landscape—as opposed to speculative AI meme tokens with no practical use—remain extremely rare. After extensive research and filtering to exclude overly early-stage or evidently non-substantive projects, we have categorized and rated the current leading AgentFi initiatives as follows:
Project | Network | Rating | Description |
Giza / ARMA | Base | ⭐⭐⭐⭐⭐ | Non-custodial automated yield agent with cryptographic boundaries, flagship of the XenoFi paradigm |
Project | Network | Rating | Description |
Theoriq | Multi-chain | ⭐⭐⭐⭐ | LLM-driven strategy composition; clear agent logic; execution layer in progress |
Blast | ⭐⭐⭐ | Users can create NFT-based agents with custom strategy execution; already live |
Project | Network | Rating | Description |
HeyAnon | Solana | ⭐⭐⭐⭐ | Deployed trading agents combining sentiment analysis with on-chain task execution |
Slate | Solana | ⭐⭐⭐⭐ | Aggregates social alpha + wallet monitoring; closed-loop on-chain agent model |
ARC | Solana | ⭐⭐⭐ | Focused on Solana alpha strategies with automated trading agents; clear product roadmap |
Autonomous Finance | AO | ⭐⭐⭐ | Offers DCA and rebalancing strategies; lightweight and efficient financial agent |
PAAL Hive AI | Solana | ⭐⭐ | Agent-based architecture but unclear scope; centralized execution logic raises risks |
Griffain AI | Multi-chain | ⭐⭐ | AI assistant platform; lacks full execution loop; has potential but still underdeveloped |
All nine projects listed above have demonstrated early-stage capabilities in closed-loop on-chain execution and scenario-specific use cases, with at least a basic working product. That said, the AgentFi sector remains in a very nascent phase. Most product experiences are still rough compared to mature DeFi protocols. We look forward to seeing this sector move beyond the speculative meme-token phase and into genuine product-market fit.
Stablecoin yield optimization is not a new concept in the DeFi world. From early pioneers like Yearn Finance (YFI) with its automated yield aggregators, to platforms like Idle Finance offering auto-rebalancing and strategy bundling, and Gauntlet, which provides institutional-grade risk modeling and protocol-level yield optimization—these projects all represent earlier iterations of automated yield strategies, predating the AgentFi paradigm.
Here is a comparative overview of leading yield optimization protocols:
Comparison Metric | ARMA (Giza) | Yearn v3 | Idle Finance | Aera (Gauntlet) |
Key Features | AI-powered modular agents with cross-protocol routing | Modular strategies, strong governance, flexible strategy bundling | Clear risk pools, automatic rebalancing, wallet-friendly integration | Risk tuning and restaking allocations; tailored for DAOs and treasuries |
Modular Architecture | Strong (Modules + Agents) | Strong (Strategy modules) | Moderate (Predefined risk pools) | Treasury-focused configurability |
User Customization | AI-generated strategies and Personalization | Moderate (requires dev knowledge) | Limited (risk level presets) | Moderate (DAO-focused custom rules) |
Automation Level | Fully automated execution and dynamic decision making | Auto compounding | Auto rebalancing | Automated treasury management |
AI / Agent Integration | AI Agents | No AI | No AI | No AI |
Protocol Coverage | Morpho, Aave, Fluid, etc. | Curve, Convex, and more | Major lending + yield aggregators | Lido, Eigenlayer, etc. |
Security Approach | Smart accounts and session keys | Audits + governance mechanisms | Strategy scoring system | Institutional-grade treasury risk control |
Profitability | Real-time, agent-based execution with full-stack reward capture, gas-aware net-profit enforcement, and intent-driven strategy formulation | Modular strategy vaults, community-curated optimizations, and automated rebalancing for passive profitability | Auto-yield routing with risk-adjusted scoring across protocols, focused on conservative optimization for C-end users | Quant-driven optimization focused on DAO treasury efficiency; emphasizes capital reallocation and risk mitigation over short-term yield |
AgentFi, as a new class of financial infrastructure, marks an early realization of the AI + Crypto fusion—where AI agents collect and process on-chain data, make autonomous decisions, and execute strategies in a transparent and verifiable manner. Leveraging blockchain immutability, every execution is recorded as an auditable on-chain log, ensuring full traceability.
As a continuation of our previous stablecoin yield research, we highlight ARMA as the only AgentFi protocol in the stablecoin yield category that currently earns a full 5-star usability rating. It stands out as the most mature and functional product in this emerging track.
Giza introduces a groundbreaking concept: Xenocognitive Finance—a framework where AI agents act as cognitive representatives within decentralized finance. These agents are not passive tools but autonomous decision-makers with reasoning capabilities and on-chain execution authority. By leveraging a distributed agent network and zero-knowledge proofs (ZK), the system enhances collective market intelligence, allowing it to scale with participation rather than collapse under complexity. This architecture enables “cognitive offload” without surrendering asset sovereignty, letting users participate in DeFi efficiently, without manual monitoring or protocol switching.
Layer | Role | Implementation Details | Key Value |
1. Semantic Abstraction Layer | Bridges the semantic gap between AI and blockchain by building an AI-readable DeFi language layer | - Based on MCP (Model Context Protocol) defining standardized financial semantics - Abstracts DeFi protocols into resources, tools, and workflows - Enables agents to directly understand high-level concepts like “lend,” “farm,” or “rebalance” | Makes AI agents smarter and faster by avoiding raw contract-level complexity |
2. Decentralized Execution Layer | Provides scalable, secure, and censorship-resistant execution environment for agents | - Built on EigenLayer’s AVS (Actively Validated Services) architecture - Four node roles: Entrypoint, Performer, Attester, Aggregator - $GIZA staking and slashing to discourage malicious behavior | A robust, trustless execution system for cross-protocol agent tasks |
3. Agent Authorization Layer | Ensures non-custodial, controlled agent operation | - Based on ERC-7579 smart accounts - Uses session keys to define granular permissions (protocols, amounts, duration) - Users retain full control of their assets | Secure delegation model: execution without asset forfeiture |
Request Parsing (Semantic Layer):
The AI agent initiates an action via the MCP protocol (e.g., “deposit USDC into Aave”). The system parses natural language into standardized on-chain instructions and checks syntax, semantics, and authorization boundaries to ensure valid and executable intent.
Protocol Interaction (Execution Layer):
Execution nodes interact with DeFi protocols (e.g., lending or swapping), optimizing for gas cost, slippage, and liquidity. Upon task completion, the system generates an execution result and cryptographic proof for validation.
Verification (Authorization + Execution Layer):
The system checks whether the task adheres to session key limits (protocol, amount, time). Multiple Attesters independently verify the action. An Aggregator collects their signatures to reach consensus and ensure the result’s integrity.
Result Feedback (Semantic Layer):
The outcome is transformed into structured semantic data (e.g., yield increase, asset distribution) and returned to the AI agent for further decision-making and recursive strategy updates.
Giza’s modular and verifiable agent system enables a fully autonomous, non-custodial, and secure execution model that redefines how users engage with DeFi—transforming AI from a tool into a true on-chain actor.
In July 2023, Giza announced a $3 million pre-seed funding round led by CoinFund, with participation from StarkWare, TA Ventures, and Arrington Capital (founded by the creator of TechCrunch). Notable angel investors included Rand Hindi and Julien Bouteloup.
In May 2025, Giza raised an additional $2.2 million in a seed round led by the Base Ecosystem Fund, with participation from CoinFund, Arrington Capital, Re7 Capital, and Contango Digital Assets.
Giza was co-founded by Fran Algaba (former Head of Machine Learning at BBVA and Adidas), Cem Dagdelen, and Renç Korzay. The team is focused on using zero-knowledge proofs (ZK Proofs) to convert AI inference results into verifiable on-chain proofs, enabling AI models to interact directly with smart contracts. Giza is one of the earliest zkML teams to receive backing from top-tier investors and among the first to explore the integration of AI + ZK + DeFi autonomous execution.
ARMA (Autonomous Revenue Management Agent) is a stablecoin yield optimization agent developed by Giza. It is designed to automate cross-protocol yield maximization for users. Currently deployed on the Base networks, ARMA plans to expand to additional Layer 2 ecosystems and already supports major lending protocols including AAVE, Morpho, Compound, and Moonwell.
Smart Strategy Scheduling: Continuously analyzes stablecoin APRs across protocols, evaluates trading costs and rebalance timing, and automatically executes optimal reallocations.
APR Optimization System: Combines rate comparison, cost assessment, and rebalance logic to ensure each operation delivers net positive yield. In practice, annualized returns can reach up to 2x those of passive static holdings.
Auto-Compounding: Automatically claims and reinvests reward tokens, converting them back into base stablecoins. Compounding frequency is dynamically adjusted based on position size and gas cost to maximize efficiency.
Smart Token Swapping: Supports dual-token operations (USDC and USDT); integrates with DEXs for efficient swaps when needed, allowing users to always withdraw assets in their original deposit currency for better flexibility and UX.
The Giza Agent architecture is built on six core modules, forming a secure, efficient, intelligent, and verifiable on-chain automation system:
Module | Function | Key Role |
Smart Accounts | Decouples asset custody from execution authorization (based on ERC-4337) | Enables non-custodial control, ensuring agents operate strictly within user-defined boundaries |
Session Keys | Granular permission management | Restricts agent activity by protocol, amount, time, and action type |
Agent Core | Strategy logic and transaction generation | Aggregates data, assesses opportunities, and formulates executable strategies |
Protocol Integration | Multi-protocol connectors and interaction interfaces | Standardized access to external DeFi protocols, with real-time monitoring and fault tolerance |
Risk Management | Multi-dimensional risk control | Continuously evaluates risks and restricts strategy scope; includes circuit breaker mechanisms |
Accounting Module | Multi-chain asset tracking and yield analytics | Tracks fund flows and yield breakdowns; supports tax reporting and historical data review |
ARMA's Three-Layer Security Framework
Smart Account Security: Users retain full control over assets via self-custodied smart accounts. Giza has no private key access. All contracts are audited and support fine-grained permission settings.
Access Control: Session Keys define precise operational limits—protocol access, permitted actions, amounts, and validity periods—preventing blanket authorizations.
Risk Management System: Only integrates with carefully selected, stable-yield DeFi protocols. All operations are fully traceable on-chain, ensuring transparency and auditability, and enhancing overall system security.
ARMA prioritizes a transparent and user-friendly monetization model, with three key elements:
Performance-Based Fees: Charges a 10% success fee only on realized yields. The fee is settled upon user withdrawal, with all fee calculations clearly visible via the ARMA dashboard.
Reward Management: Automatically aggregates rewards across integrated DeFi protocols. All rewards are distributed together with principal and yield at withdrawal—no user action required.
Asset Handling: No deposit or withdrawal fees. Regardless of how many protocols or tokens are involved, users always receive their funds back in the original deposit token, ensuring a seamless experience.
ARMA’s performance and integration data can be monitored through its official dashboard at metrics.gizatech.xyz. The dashboard showcases real-time asset flows across protocols like Morpho, Aave, and Fluid, as well as the steady growth in assets under management and deployed agents since launch. As ARMA integrates more DeFi protocols and scales both agent activity and managed capital, it is expected to further reinforce its flywheel of growth and protocol utility.
Through ARMA’s official dashboard (metrics.gizatech.xyz), we can clearly observe the continuous growth in both assets under management and the number of deployed intelligent agents over the past few months since launch. The dashboard also visualizes the dynamic asset flows between major lending protocols such as Morpho, Aave, and Fluid. We believe that as ARMA continues to integrate with more DeFi protocols, increase its active agent count, and expand its capital base, it will further strengthen its positive growth flywheel.
Giza has just officially released the full details of its tokenomics, marking the upcoming TGE as a key milestone worth close attention.
$GIZA is the native asset of the Giza agent network—designed for immediate utility, long-term value capture, and built on principles of community-first distribution, product-driven adoption, and mechanism restraint.
Fair Initial Valuation: TGE launches with an initial FDV of $70M—significantly lower than comparable AI projects, leaving ample upside potential.
Zero Unlock at Launch: No tokens unlocked for team or investors at TGE, ensuring zero initial sell pressure and full alignment with the community.
Live Product Traction: The ARMA agent has already managed over $30M in on-chain assets, validating strong market demand and real-world utility.
Clear Utility: $GIZA powers staking, permissioning, governance, and agent incentives—serving as the protocol’s core asset.
Incentives Tied to Security: Validator nodes must stake $GIZA; malicious behavior is penalized to ensure reliable execution.
Value Recirculation Mechanism: Part of agent-generated revenue is funneled through AMMs for $GIZA buybacks or redistribution, creating a sustainable incentive loop.
Community-Oriented Distribution: Early token allocations favor real users and contributors, prioritizing fairness and long-term participation.
Dual Staking Model: Both node operators and community members can stake $GIZA—earning execution rewards or governance power respectively.
Time-Weighted Boosts: Longer staking durations yield higher rewards and voting power, up to 5x—encouraging long-term commitment.
Progressive Governance: Governance evolves in phases—from Security Council to delegate-based and specialized councils—culminating in full decentralization.
To reward meaningful contributions to the Giza agent ecosystem, the team launched the Giza Points System (GPS) in late 2024 as a foundational basis for future token distribution. The system prioritizes long-term contribution over short-term activity, and is open to:
Users: Earn points by using ARMA agents, submitting feedback, participating in stress tests, and other on-chain activities.
Developers: Receive rewards for building new agents, optimizing execution logic, and contributing to security improvements.
Community Members: Gain recognition through education, governance participation, and ecosystem-building efforts.
ARMA Early-Stage Yield: 15% Fixed APR + $GIZA Upside
ARMA offers a compelling early-stage incentive: users who deposit stablecoins earn a 15% fixed APR for the first 3 months, combining base yield and bonus rewards.
Beyond the fixed return, early participants (up to $10M deposits) also receive a pro-rata share of 1% of total $GIZA supply, with tokens vested over 9 months. No lock-ups or penalties apply—ideal for users seeking both short-term fixed returns and long-term upside in $GIZA.
Giza has built a self-reinforcing growth flywheel driven by agent utility, protocol usage, token value, and ecosystem participation. As intelligent agents improve user capital efficiency and automate complex strategies, DeFi usage increases. This drives protocol-level interactions and fee generation, which flows back to $GIZA stakers and execution nodes—enhancing token value. Higher staking yields attract more nodes and developers, who contribute new agent models and integrations, expanding Giza’s capabilities. This loop strengthens the network’s scalability, security, and composability, following the cycle:
More Usage → Higher Token Value → Stronger Security → Broader Capabilities → More Usage
Giza adopts a phased decentralization model for governance, combining expert councils and community-elected representatives.
Early Phase: A Security Council oversees protocol safety, with community input through discussion and signaling.
Mid Phase: Key governance decisions shift on-chain, enabling token-weighted voting.
Final Phase: Governance is fully community-elected, with a council composed of domain experts.
Token holders can delegate their votes to trusted representatives with technical or economic expertise. A time-weighted staking system increases voting influence based on long-term commitment—encouraging responsible governance and shared accountability in building a secure, efficient, and democratic Agent-native infrastructure.
I first encountered the Giza team at ETHcc in Paris in 2023, when the project was still positioned around the concept of ZKML (Zero-Knowledge Machine Learning). Over the course of 2024, Giza gradually evolved into a more engineering-driven and DeFi-oriented AgentFi project. After conducting extensive research and careful evaluation of both established stablecoin yield protocols and emerging AgentFi projects, I believe that Giza’s ARMA stands out as one of the very few AgentFi protocols with a real, battle-tested product and a complete on-chain execution loop. This product-before-token approach is in stark contrast to many DeFAI-era projects that were long on hype and short on substance—eventually devolving into valueless meme coins.
That said, we must also acknowledge that compared to other stablecoin yield strategies, AgentFi remains in its infancy. The protocol has not yet gone through a full market cycle and should still be considered a high-volatility, experimental "early-mine" phase. For DeFi degens with a high risk appetite, allocating a small portion of stablecoin assets may be worthwhile, but prudent position sizing and diversification are essential.
Paradigm-Shifting Narrative: Giza sits at the intersection of AgentFi, ZK, and AI—supported by its original “Xenocognitive Finance” thesis, which proposes a future where AI agents offload human cognitive load in DeFi without sacrificing custody or control. This gives the project a strong first-mover narrative with philosophical depth.
Superior Product Maturity: ARMA is among the very few live AgentFi products that combine ZK verification with autonomous agent execution, offering a tangible user experience far ahead of most still-conceptual or token-led projects.
Early-Mine Yield Opportunity: Stablecoin yields, when backed by secure protocol design, present relatively controlled risk. Coupled with ARMA’s token incentive mechanisms, early users may enjoy favorable reward asymmetry akin to early liquidity mining.
Early-Stage Trust Barrier: AgentFi is still an emerging concept. User confidence in fully autonomous AI agents is not yet widespread, particularly regarding fund safety and decision transparency. User growth has not yet reached a flywheel tipping point.
Compound Systemic Risk: ARMA integrates multiple advanced technologies—ZK proofs, AI inference, smart account abstraction, and multi-protocol interactions. Its security hinges on both smart contract design and inference correctness. Any weak point could expose users to significant risk.
Unproven Token Value Path: In today’s sentiment-driven crypto market, token performance is often disconnected from fundamentals. Giza’s team does not follow a speculative approach, and the $GIZA token’s incentive structure, staking dynamics, and value capture mechanisms still require market validation—posing a challenge for long-term investors.
As stablecoin yield strategies become increasingly sophisticated, users are simultaneously constrained by limited attention and execution capacity. AgentFi, as the next evolution in the AI + DeFi convergence, offers a promising solution: intelligent agents that automate yield optimization without requiring constant monitoring. Using Giza and its flagship product ARMA as a case study, this report outlines the current state, technical architecture, and yield potential of the AgentFi landscape.
Giza, grounded in its vision of Xenocognitive Finance, is one of the few AgentFi projects to deliver a complete and practical implementation of ZK verification + AI-powered strategies + non-custodial execution. ARMA focuses specifically on optimizing stablecoin yields, is already live across multiple Layer 2 networks, and integrates with major lending protocols. With a structured token incentive model, it offers attractive early-mover benefits akin to a “head-mining” phase.
That said, it is important to recognize that AgentFi remains in its infancy. User trust, execution stability, and token value accrual mechanisms all require further market validation. Precisely because of this, however, AgentFi—and especially Giza/ARMA—may present a rare opportunity for users and builders who believe in the long-term vision of autonomous finance and are willing to participate at the frontier of paradigm shift.
jacobzhao