The Smart Evolution of DeFi: The Evolution Path from Automation to AgentFi

Written by: 0xjacobzhao, ChatGPT 4o

Thanks to Lex Sokolin (Generative Ventures), Stepan Gershuni (cyber.fund), and Advait Jayant (Aivos Labs) for their valuable suggestions on this article. Feedback was also sought from project teams such as Giza, Theoriq, Olas, HeyElsa, Almanak, and Brahma.fi during the writing process. This article strives for objectivity and accuracy in expression; however, due to the subjective nature of some viewpoints, there may be biases. Readers are encouraged to read critically and understand accordingly.

In the current cryptocurrency industry, stablecoin payments and DeFi applications are among the few validated sectors with genuine demand and long-term value. At the same time, the flourishing Agent is gradually becoming a practical interface in the AI industry, serving as a crucial intermediary layer connecting AI capabilities with user needs.

In the field of integration between Crypto and AI, especially in the direction where AI technology feeds back into Crypto applications, current explorations are mainly focused on three typical scenarios:

1. Conversational interactive Agents: Primarily chatbots, companions, and assistants. Although most are still wrappers around general large models, their low development threshold and natural interactions, combined with token incentives, have made them the first forms to be pushed into the market to gain user attention.
2. Information integration Agents: Focused on the intelligent integration of online and on-chain information. Projects like Kaito and AIXBT have achieved success in the online but off-chain information search and integration field, while on-chain data integration is still in the exploratory stage with no significant projects emerging.
3. Strategy execution Agents: Centered around stablecoin payments and DeFi strategy execution, extending into two major directions: Agent Payment and DeFAI. These Agents are more deeply embedded in on-chain trading and asset management logic, with the potential to break through speculative bottlenecks and form intelligent execution infrastructure with financial efficiency and sustainable returns.

This article will focus on the evolution path of the integration between DeFi and AI, outlining its development stages from automation to intelligence, and analyzing the infrastructure, scenario space, and key challenges of strategy execution Agents.

DeFi Intelligence Evolution in Three Stages: Automation, Copilot, and AgentFi Leap

In the evolution of DeFi intelligence, we can categorize system capabilities into three stages: Automation (automated tools), Intent-Centric Copilot (intention-driven assistants), and AgentFi (on-chain agents).

• Automation resembles a Rule Trigger: It executes fixed tasks based on preset conditions, such as arbitrage, rebalancing, and stop-loss, but cannot generate strategies or operate independently.
• Copilot introduces intention recognition and semantic parsing capabilities, allowing users to input natural language, which the system understands, breaks down, and suggests execution paths, but ultimately requires user confirmation, leaving the execution chain open.
• AgentFi represents a complete "perception → reasoning/strategy generation → on-chain execution → evolution" intelligent closed loop, embodying an agent with autonomous on-chain execution and continuous evolution capabilities.

To determine whether a project truly belongs to AgentFi, it must meet at least three of the following five core criteria:

1. Autonomous perception of on-chain states/market signals (not static inputs, but real-time monitoring)
2. Capability for strategy generation and combination (not preset strategies, but able to formulate action plans based on context)
3. Ability to autonomously execute operations on-chain (no user interaction required, capable of executing complex operations like swap/lend/stake)
4. Possession of persistent state and evolution capability (Agents have a lifecycle, can operate long-term, and adjust based on feedback)
5. Possession of Agent-Native architecture (such as dedicated Agent SDK, hosted execution environments, Agent middleware, etc.)

In other words, automated trading ≠ Copilot, and even more ≠ AgentFi: automated trading is merely a "rule trigger," while Copilot can understand user intentions and provide operational suggestions but still relies on human participation; true AgentFi is an "intelligent agent with perception, reasoning, and autonomous on-chain execution capabilities," capable of completing strategy loops and continuous evolution without human intervention.

DeFi Scenario Intelligence Adaptability Analysis:

In the DeFi (Decentralized Finance) system, core application scenarios can be roughly divided into asset circulation and exchange types and yield-based financial types. We believe that these two types of scenarios exhibit significant differences in adaptability along the intelligence path:

1. Asset Circulation and Exchange Scenarios

Asset circulation and exchange scenarios primarily involve atomic interactions, including swap transactions, cross-chain bridges, and fiat currency inflows and outflows. Their essential characteristics are "intention-driven + single atomic interaction," where the trading process does not involve yield strategies, state maintenance, or evolution logic, making them mostly suitable for the lightweight execution path of Intent-Centric Copilot, and not belonging to AgentFi.

Due to their lower engineering thresholds and simple interactions, most DeFAI projects currently in the market are at this stage, which do not constitute AgentFi closed-loop intelligent agents; however, a few advanced complex swap strategies (such as cross-asset arbitrage, perpetual hedge LP, leveraged rebalancing, etc.) actually require the capabilities of AI Agents, which are still in the early exploratory stage.

2. Asset Yield Financial Scenarios

Asset yield financial scenarios have clear yield objectives, complex strategy combination spaces, and dynamic state management needs, naturally aligning with AgentFi's "strategy closed loop + autonomous execution" model. Their core features are as follows:

• Quantifiable yield objectives (APR/APY) facilitate the Agent's establishment of optimization functions;
• A broad strategy combination space covering multiple assets, multiple timeframes, multiple platforms, and multiple interaction processes;
• Operations require frequent management and real-time adjustments, suitable for execution and maintenance by on-chain intelligent agents (Agents).

Due to multiple factors such as yield duration, volatility frequency, on-chain data complexity, cross-protocol integration difficulties, and compliance restrictions, different yield scenarios exhibit significant differences in adaptability and engineering feasibility within the AgentFi dimension, with the following priority recommendations:

High-priority business deployment directions:

• Lending/Borrowing: Interest rate fluctuations are easy to track with standardized execution logic, suitable for lightweight intelligent agents.
• Yield Farming: Pools are dynamic and frequent, with a large strategy combination space and high yield fluctuations. AgentFi can significantly optimize annual returns and interaction efficiency, but engineering implementation poses certain challenges.

Medium to long-term exploratory layout directions:

• Pendle yield rights trading: Clear time dimensions and yield curves, suitable for Agents to manage maturity rotations and inter-pool arbitrage;
• Funding Rate arbitrage: Theoretical yields are considerable, but challenges in cross-market execution and off-chain interaction must be resolved, with high engineering difficulty;
• LRT dynamic combination structure: Static staking is not suitable; attempts can be made to automatically adjust strategies like LRT + LP + Lending.
• RWA multi-asset portfolio management: Difficult to implement in the short term, but Agents can assist in portfolio optimization and maturity strategies.

Introduction to Intelligent Projects in DeFi Scenarios:

1. Automation Tools (Automation Infra): Rule Trigger and Conditional Execution

Gelato is one of the earliest infrastructures for DeFi automation, previously providing conditional trigger task execution support for protocols like Aave and Reflexer, but it has now transformed into a Rollup as a Service provider. The main battlefield for on-chain automation has also shifted to DeFi asset management platforms (DeFi Saver, Instadapp). These platforms integrate standardized automated execution modules, including limit order setting, liquidation protection, automatic rebalancing, DCA, and grid strategies. Additionally, we see some more complex DeFi automation tool platform projects:

Mimic.fi (https://www.mimic.fi/)

Mimic.fi is an on-chain automation platform serving DeFi developers and projects, supporting the construction of programmable automation tasks on chains like Arbitrum, Base, and Optimism. Its core achieves cross-protocol operation automation through "if-then" rule triggers, and its architecture is divided into three layers: Planning (task and trigger definition), Execution (intention broadcasting and execution bidding), and Security (triple verification and security control). Currently, it adopts an SDK integration approach, and the product is still in the early deployment stage.

AFI Protocol (https://www.afiprotocol.ai/)

AFI Protocol is an algorithm-driven Agent execution network that supports 24/7 non-custodial automated operations, focusing on solving issues of execution decentralization, strategy thresholds, and risk response in DeFi. Its design targets institutions and advanced users, providing orchestrable strategies, permission management, and SDK tools, and has launched the yield-bearing stablecoin afiUSD as its native asset. It is currently in the Sonic Labs internal testing phase and has not yet been publicly launched or opened for retail user access.

2. Intent-Centric Copilot: Intention Expression and Execution Suggestions

The DeFAI concept, which was once hot at the end of 2024, aside from some speculation primarily involving meme tokens, is essentially dominated by Intent-Centric Copilot types—where users express their intentions through natural language, and the system responds with trading suggestions or completes basic on-chain operations. Its core capabilities remain at the "intention recognition + Copilot-style assisted execution" stage, lacking a complete strategy closed loop and continuous optimization mechanism. Many products have obvious shortcomings in semantic understanding, cross-protocol invocation, and feedback response, leading to generally poor user experiences and relatively limited functional boundaries.

HeyElsa (https://app.heyelsa.ai/)

HeyElsa is an AI Copilot positioned for Web3 scenarios, empowering users to complete various on-chain operations, including trading, cross-chain bridging, NFT purchasing, stop-loss setting, and Zora token creation, through natural language interaction. As a multifunctional conversational crypto assistant, it covers users from beginners to advanced traders (including highly active degen groups) and currently supports real-time interactions across more than 10 mainstream blockchains. The platform's daily trading volume has reached $1 million, with daily active users maintained between 3,000 and 5,000. The system has integrated yield optimization strategies and automated intention execution modules, initially constructing the foundational capability framework for AgentFi applications.

Bankr (https://bankr.bot/)

Bankr is an intention trading assistant that integrates AI, DeFi, and social scenarios. Users can issue commands through natural language on the X platform or dedicated terminal to complete operations like swap, limit orders, cross-chain bridging, token issuance, and NFT minting, supporting Base, Solana, Polygon, and Ethereum mainnet. Bankr has built a complete Intent → Compile → Execute link, emphasizing a minimalist trading experience and seamless operations within a social environment, and activates the ecosystem through token incentives and revenue-sharing mechanisms.

Griffain (https://griffain.com/)

Griffain is a multifunctional AI Agent platform deployed on Solana, supporting natural language interactions between users and the Griffain Copilot to perform on-chain operations such as asset queries, swaps, NFT trading, and LP management. The platform includes multiple intelligent agent modules and encourages community participation in Agent creation and sharing. Technically based on the Anchor Framework and components like Jupiter and Tensor, it emphasizes mobile adaptation and front-end composability. It currently supports over 10 core Agent modules, demonstrating strong execution capabilities and ecological linkage.

Symphony (https://www.symphony.io/)

Symphony is an on-chain execution infrastructure for AI Agents, building a full-stack system covering intention modeling, intelligent path discovery, RFQ execution, and account abstraction, aiming to become the core module of the DeFi intelligent execution layer. The platform has launched the conversational assistant Sympson, which features market query and strategy suggestion capabilities, but on-chain execution has not yet been opened. Symphony provides the core components needed for AgentFi, which can support collaborative execution and cross-chain operations among multiple Agents in the future.

Hey Anon (https://heyanon.ai/)

HeyAnon is a DeFAI platform that combines intention interaction, on-chain execution, and intelligence analysis, supporting multi-chain deployment (Ethereum, Base, Solana, etc.) and cross-chain bridging (LayerZero, deBridge). Users can complete operations like swaps, lending, and staking through natural language and obtain on-chain sentiment and market dynamics analysis. Although the project has gained high attention due to its founder Sesta, it is still in the Copilot stage, with core strategies and execution intelligence not fully realized, and its long-term development remains to be observed.

The above scoring system is primarily based on the current usability, user experience, and the feasibility of executing the public roadmap of the products, which has a certain subjectivity. Please note that this assessment does not involve code security checks and does not constitute investment advice; your understanding is appreciated.

3. AgentFi Intelligent Agents: Strategy Closed Loop and Autonomous Execution

We believe that AgentFi is a more advanced form of DeFi intelligence evolution compared to Intent Copilot. Agents possess independent yield strategies and on-chain automatic execution capabilities, significantly enhancing users' strategy execution efficiency and capital utilization. By 2025, we are pleased to see more and more AgentFi projects being implemented or planned, mainly focusing on lending and liquidity mining directions, with representative projects including Giza ARMA, Theoriq AlphaSwarm, Almanak, Brahma, and the Olas series.

Giza ARMA (https://arma.xyz/)

ARMA is an intelligent agent product launched by Giza, specifically designed for stablecoin cross-protocol yield optimization. It is deployed on the Base network and supports multiple mainstream lending protocols such as Aave, Morpho, Compound, and Moonwell, with core capabilities including cross-protocol rebalancing, automatic compounding, and intelligent token swapping. ARMA's strategy system can monitor stablecoin APR, transaction costs, and yield differences in real-time, automatically adjusting capital allocation, with actual yields significantly higher than static holdings. Its architecture consists of smart accounts, session keys, core agent logic, protocol access, risk management, and accounting modules, ensuring safe and efficient automated execution in a non-custodial mode.

ARMA is now fully online and continuously iterating. With its modular architecture, security mechanisms, and good early operational data, ARMA has become one of the most viable Agent products in DeFi automated yield management, representing one of the few AgentFi projects that combine deep concepts with practical products.

Refer to the research report "A New Paradigm for Stablecoin Yields: From AgentFi to XenoFi"

Theoriq (https://www.theoriq.ai/)

Theoriq Alpha Protocol is a multi-agent collaboration protocol focused on DeFi scenarios, with its core product Alpha Swarm concentrating on liquidity management, aiming to build a full-chain automated closed loop of "perception---decision---execution." It consists of three types of Agents: Portal (on-chain signal perception), Knowledge (data analysis and strategy selection), and LP Assistant (strategy execution), enabling dynamic asset allocation and yield optimization without human intervention. The underlying Alpha Protocol provides Agent registration, communication, parameter configuration, and development tool support, serving as the operational foundation for the entire Swarm collaborative system. Through AlphaStudio, users can browse, invoke, and combine various Agents to build a modular and scalable automated trading strategy network.

As one of the first projects on the Kaito Capital Launchpad, Theoriq recently completed a $84 million community fundraising and is about to undergo TGE. Theoriq has recently launched the AlphaSwarm Community Beta test network, with the mainnet version also set to be officially released soon.

Refer to the research report "Theoriq Research Report: The Evolution of AgentFi in Liquidity Mining Yields"

Almanak (https://almanak.co/)

Almanak is an intelligent Agent platform for DeFi strategy automation, combining a non-custodial security architecture with a Python strategy engine to help traders and developers deploy sustainable on-chain strategies.

The platform's core consists of Deployment (execution components), Strategy (strategy logic), Wallet (Safe+Zodiac security module), and Vault (strategy assetization), supporting yield optimization, cross-protocol interaction, liquidity provision, and automated trading. Compared to traditional DeFi tools, Almanak emphasizes AI-assisted market perception and risk management capabilities, already possessing 24/7 intelligent operation capabilities, and plans to introduce multi-agent and AI decision systems, aiming to build the next generation of AgentFi infrastructure.

Almanak's strategy system is built on a Python-based state machine program, serving as the "decision brain" for each Agent, automatically formulating and executing on-chain operations based on market data, wallet status, and user-defined conditions. The platform provides a complete Strategy Framework, supporting the encapsulation of on-chain trading, lending, liquidity provision, and other operational modules (Action Bundle) without the need to write underlying contract code, and ensures strategy confidentiality and operational security through cryptographic isolation, permission control, and monitoring mechanisms. Users can write strategies through the SDK, and future support for natural language strategy creation is planned, achieving a smooth transition from complex logic to no-code experiences.

Currently, the product has launched a USDC lending Vault based on the Ethereum mainnet, while more complex trading strategies are in the testing phase and require whitelist access. Almanak is about to join the cookie.fun cSNAPS campaign for community fundraising, which is worth looking forward to.

Brahma (https://brahma.fi/)

Brahma positions itself as "the intelligent capital orchestration layer," dedicated to abstracting on-chain accounts, execution logic, and off-chain payment processes, helping users and developers efficiently manage on-chain and real-world assets. Through Smart Accounts, continuously running on-chain Agents, and the Capital Orchestration Stack, Brahma provides users with an intelligent capital management experience without backend operations.

Currently launched representative Agents include:

• Felix Agent: Automatically optimizes feUSD debt warehouse interest rates, preventing liquidation and saving interest;
• Surge & Purge Agent: Tracks volatility and executes automatic trading;
• Morpho Agent: Deploys and rebalances Morpho treasury funds;
• ConsoleKit framework: Supports the integration of any AI model, unifying execution strategies and asset management.

Olas (https://olas.network/)

The AgentFi product series BabyDegen launched by Olas Network includes Modius Agent and Optimus Agent, both of which have been deployed on-chain, covering a multi-chain ecosystem (Solana, Mode, Optimism, Base), and possess complete on-chain interaction capabilities, strategy execution capabilities, and autonomous asset management mechanisms.

• BabyDegen is an AI trading agent running on Solana, implementing automatic buying and selling based on CoinGecko data and community strategy libraries, currently integrated with Jupiter DEX and in the Alpha testing phase.
• Modius Agent targets the Mode network, focusing on USDC and ETH portfolio management, having integrated Balancer, Sturdy, and Velodrome, supporting users to set preferences for 24/7 automatic execution of strategies.
• Optimus Agent is compatible with Mode, Optimism, and Base mainnets, integrating more protocols like Uniswap and Velodrome, providing flexible multi-chain strategy combinations suitable for intermediate to advanced users to build automated asset management systems.

Axal (https://www.getaxal.com/)

Axal's core product Autopilot Yield offers a one-stop, non-custodial, verifiable yield management experience, integrating mainstream protocols such as Aave, Morpho, Kamino, Pendle, and Hyperliquid, with on-chain strategy execution + risk control as its core design philosophy, empowering ordinary users to easily enter complex on-chain yield networks.

• Conservative strategy focuses on low-risk, mainstream stable yield scenarios, primarily deploying funds on well-established platforms like Aave and Morpho, with annual yields around 5-7%. It achieves steady appreciation through TVL monitoring, stop-loss mechanisms, and top strategy filtering, suitable for users pursuing capital safety and long-term returns.
• Balanced strategy offers medium risk and higher yield potential (10-20% APY), using wrapped stablecoins (like feUSD, USDxL), liquidity provision, and arbitrage neutral positions. The strategies are more diverse, with complex yield compositions, controlled through Axal's automatic monitoring and dynamic adjustments.
• Aggressive strategy targets users with high-risk, high-return preferences, covering strategies such as high-leverage LP, cross-platform linking, low liquidity asset market making, and volatility capture, with theoretical annual yields potentially exceeding 50%. Axal's intelligent agents can set stop-loss, automatic exit, and redeployment logic at the strategy level, providing users with a final layer of protection in high-risk environments.

Fungi.ag (https://fungi.ag/)

Fungi.ag is a fully automated AI Agent designed for USDC yield optimization, capable of automatically reallocating funds among multiple lending protocols such as Aave, Morpho, Moonwell, and Fluid, achieving optimal capital allocation based on yield rates, fees, and risks. Users do not need to operate manually; they only need to authorize the Session Key to enable the Agent to automatically execute strategies in a non-custodial mode. Currently, it supports the Base chain and plans to expand to Arbitrum and Optimism. Fungi also opens the Hypha custom strategy script interface, supporting community development of strategies like DCA and arbitrage, and building an ecosystem through DAO and social platforms.

ZyFAI (https://www.zyf.ai/)

ZyFAI is a DeFi intelligent assistant platform deployed on Base and Sonic networks, combining on-chain interaction interfaces with AI-assisted modules to help users manage intelligent assets under different risk preferences. Its core is divided into three types of strategies:

• Safe Strategy: Designed for conservative users, focusing on mainstream protocols like Aave, Morpho, Compound, Moonwell, and Spark that have been audited and verified, emphasizing asset safety and long-term reliability with single-sided deposits of USDC and stable yield opportunities.
• Yieldor Strategy: Targeted at high-risk preference users, requiring the holding of 20,000 ZFI tokens to unlock, covering high-yield protocols including Pendle, YieldFi, Harvest Finance, and Wasabi, supporting complex strategies like DEX LP, yield splitting, and leveraged Vaults, with plans to expand to structured products like Looping and Delta-neutral in the future.
• Airdrop Strategy: A future strategy still in development, aimed at obtaining more airdrop incentives.

The above scoring system is primarily based on the current usability, user experience, and the feasibility of executing the public roadmap of the products, which has a certain subjectivity. Please note that this assessment does not involve code security checks and does not constitute investment advice; your understanding is appreciated.

The Real Path and High-Level Vision of AgentFi

Undoubtedly, lending and liquidity mining are the business scenarios with the most genuine value and the easiest short-term implementation for AgentFi. They have matured in the DeFi world and are naturally suitable for the introduction of intelligent agents due to the following common characteristics:

1. Broad strategy space with multiple optimization dimensions

• Lending can pursue not only the highest yields but also strategies like interest rate arbitrage, leverage cycling, debt refinancing, and liquidation protection;
• Liquidity mining encompasses a rich strategy orchestration space, including APR tracking, LP rebalancing, compounding, and strategy combinations.

1. Highly dynamic, suitable for real-time perception and response by intelligent agents: Changes in interest rates, fluctuations in TVL, changes in reward structures, new pools coming online, and new protocols emerging all affect the optimal strategy path and require dynamic adjustments.

2. There are opportunity costs in execution windows, making automation valuable: Funds not allocated to the optimal pool will drag down yields and need to be automatically migrated.

It is particularly noteworthy that lending Agents, due to stable data structures and relatively simple strategies, have a high feasibility for implementation, such as Giza's Arma and other lending AgentFi projects that have officially launched. However, the management of liquidity mining requires real-time responses to price fluctuations, volatility changes, and fee accumulation, placing extremely high demands on the data perception, strategy judgment, and on-chain execution of Agents. LP Agents not only need to accurately predict market conditions but also need to perform dynamic rebalancing and yield redistribution operations on-chain, which adds to the engineering complexity, a challenge that projects like Theoriq are currently tackling.

Beyond lending and liquidity mining, we envision potential medium to long-term exploratory layout directions based on the adaptability of AgentFi:

Pendle Yield Rights Trading: Clear time dimensions and yield curves, suitable for Agents to manage maturity rotations and inter-pool arbitrage

Pendle, with its unique structure of "yield splitting + maturity mechanism + yield rights trading," provides a natural strategy orchestration space for AgentFi. Its assets are divided into PT (Principal Token) and YT (Yield Token), where the former represents the principal redeemable at maturity, suitable for stable fixed-income configurations; the latter represents yield rights, which are variable and can be used for speculation, mining, and arbitrage. Users can construct various complex strategies around these two types of assets, including solid income holdings, YT farming, maturity fund management, interest rate arbitrage, and portfolio hedging.

In practical scenarios, Pendle has several user pain points that urgently need AgentFi solutions: for example, high-yield pools are mostly concentrated in the short term of 1-3 months, requiring manual reconfiguration after maturity; the yield rates of different pools fluctuate greatly, leading to high tracking and rotation costs; and the combination strategy of PT+YT involves complex pricing judgments and position rebalancing. If AgentFi can automate the entire process from strategy identification, liquidity allocation, to maturity rotation and redeployment based on user yield preferences and risk tolerance, it will significantly enhance capital efficiency and user experience.

Pendle's "temporal, split, and dynamic" characteristics align very well with AgentFi's strategy expression and execution paths, especially in areas like automatic reinvestment, implied yield arbitrage, and yield pool rotation, exhibiting high-frequency and high-strategy characteristics, making it very suitable for building a "yield agent Swarm" or Portfolio Agent system. If in the future it can combine intention expression (e.g., "annualized 10%, withdrawable in 6 months") with an automatic execution framework, Pendle will become one of the most representative modules for AgentFi implementation.

Funding Rate Arbitrage: Theoretical yields are considerable, but challenges in cross-market and cross-chain interactions pose high engineering difficulty

Although the on-chain options track has gradually cooled due to reasons like pricing deficiencies, complex exercise processes, and poor combinability, perpetual contracts remain one of the most active scenarios among current on-chain derivatives, providing a point of intersection for AgentFi. Around strategies like funding rate arbitrage, basis trading, and multi-platform hedging, AgentFi can leverage its intelligent capabilities in perception, judgment, execution, and combination management.

In structural design, AgentFi can embed four key modules: First, a data perception module that supports real-time capture of funding rates, position costs, and market depth on-chain and CEX; second, an intelligent decision module that dynamically judges whether to open or adjust positions based on arbitrage thresholds, leverage levels, and liquidation boundaries; third, an automatic execution module that completes position deployment or profit-taking operations once triggering conditions are met; fourth, a combination management module that supports collaborative scheduling across multiple chains, accounts, and strategies.

However, real challenges include: First, current on-chain AgentFi focuses on smart contract interactions and lacks a universal framework for direct access to CEX APIs; second, high-frequency strategies require extremely high execution efficiency, gas cost control, and slippage management; third, complex arbitrage scenarios typically require multiple Agents to collaborate, necessitating Swarm-style cooperation.

Ethena's funding rate arbitrage has relied on a highly automated execution system. Although Ethena currently does not possess AgentFi characteristics, if it further opens strategy modules in the future and builds a distributed Agent Swarm, achieving funding goal expressions through intention-driven mechanisms, its system may naturally transition into a complete AgentFi infrastructure.

Staking and Restaking: Naturally not suitable for AgentFi, but LRT dynamic combinations hold certain potential

Overall, traditional staking and restaking are not suitable application scenarios for AgentFi due to the simplicity of the single-chain staking process, stable yields, singular decision-making, and long exit waiting periods, which do not support the intelligent value emphasized by AgentFi.

However, in more complex staking constructs, there exists some usable space for AgentFi. This includes focusing on operating combinable LST/LRT type assets (like stETH, rsETH), avoiding direct engagement with native ETH unstake processes; emphasizing the construction of Restaking + collateral + derivatives combination strategies to bypass the time lag caused by unstaking; and deploying continuously optimizing monitoring strategies for Agents to dynamically assess AVS risks, APR changes, and reorganize positions.

Additionally, the current restaking track also faces structural challenges: on one hand, market enthusiasm has rapidly cooled; on the other hand, there is a severe imbalance between the supply side (staked ETH) and the demand side (AVS security needs), with asset leasing lacking practical application scenarios. Leading projects like EigenLayer and Either.fi have attempted to pivot. Therefore, staking/restaking may become modular strategy components of AgentFi in the future rather than the most core application landing scenarios.

RWA Assets: US Treasury protocols are not ideal scenarios, but multi-asset portfolio management structures have exploratory value

Current mainstream RWA protocols generally use US Treasury bonds (T-bills) as underlying assets, focusing on providing users with stable, safe, and compliant on-chain yield vehicles. However, from the perspective of AgentFi, these products are not suitable for high-frequency or strategy-driven intelligent agent embedding due to their stable asset nature (annual yields typically stabilize in the 4-5% range with minimal yield spreads, lacking optimization strategy space), low operational frequency (clear lock-up periods and reinvestment cycles, unsuitable for frequent rotations and difficult to achieve high-frequency compounding), and strong compliance restrictions (involving investor KYC verification and regional limitations). Furthermore, the non-interoperability of asset structures among various protocols limits the ability of Agents to perform combination routing and liquidity aggregation operations.

Nevertheless, several potential directions could become medium to long-term expansion paths for AgentFi:

1. Multi-asset RWA configuration agents (RWA Multi-Asset Portfolio): As RWA products gradually expand into areas like real estate, credit bonds, and accounts receivable, users may express the intention to "allocate a basket of stable yield assets and adjust periodically." Configuration agents can periodically complete asset weight adjustments and redeploy maturing assets, building a medium to long-term yield stabilizer.

2. RWA and DeFi fusion structures (RWA-as-Collateral & custodial reuse): Some protocols are exploring the use of tokenized T-bills as collateral assets in DeFi lending systems. In this structure, Agents can assist users in automatically completing deposit operations, interest rate comparisons, and collateral adjustments, forming dual yield paths. If RWA assets achieve widespread circulation on platforms like Pendle and Uniswap, Agents could track the premium and implied yield changes of Tokens across different platforms, constructing automatic arbitrage and rolling deployment strategies. As the market matures, this could become an important breakthrough for AgentFi in the RWA field.

Swap Trading Combinations: Upgrading from Intent Infrastructure to AgentFi Strategy Engine

In the current DeFi intelligent ecosystem, swap trading introduces account abstraction and intent patterns, hiding the complex DEX multi-chain path selection, driving user transactions to completion with simple inputs, significantly lowering interaction thresholds. However, these systems still remain at the level of "atomic action automation," lacking real-time perception and response to environmental changes, and have not introduced a goal-oriented strategy execution mechanism, thus not possessing the intelligent agent characteristics of AgentFi.

Under the AgentFi framework, swap operations are no longer a single action but part of a larger-scale combination strategy. For instance, when a user expresses the intention to "combine stETH and USDC for the highest yield," the Agent can automatically complete multiple swaps (e.g., USDC → ETH → stETH), perform restaking, split Pendle PT/YT, configure arbitrage strategies, and recover yields.

Furthermore, swaps play a key role in the following three AgentFi scenarios:

1. A link in combination yield strategies: As a capital scheduling relay station, swaps support Agents in automatically completing asset allocation paths, enhancing strategy execution efficiency.
2. Cross-market arbitrage / delta-neutral strategies: By comparing different price sources on-chain, Agents can dynamically adjust positions and construct hedging combinations.
3. Trading behavior risk defense: When detecting large transactions, Agents can automatically assess slippage, execute in batches, and avoid potential MEV attacks.

Therefore, a truly AgentFi-characterized swap agent must possess the following capabilities: dynamic strategy perception, cross-protocol scheduling, capital path optimization, trading timing judgment, and risk prevention. Future swap agents should serve multi-strategy combinations, dynamic position adjustments, and cross-protocol value capture, with a long road ahead.

DeFi Intelligent Evolution Roadmap: From Automation Tools to Intelligent Agent Networks

In summary, we have witnessed the evolution path of DeFi intelligence from automated tools to intention assistants to intelligent agents.

The first stage is "Automation Tools (Automation Infra)," characterized by rule triggers and conditional execution to achieve basic on-chain operation automation. For example, triggering trades or rebalancing tasks based on preset conditions like time and price, with representative systems being foundational execution frameworks, such as Gelato and Mimic.

The second stage is "Intent-Centric Copilot," emphasizing the expression of user intentions and the generation of execution suggestions. Systems at this stage are no longer limited to "what to do," but attempt to understand "what the user wants" and then provide the best execution path suggestions. Representative projects include Bankr and HeyElsa, primarily enhancing intention recognition and interaction experience to lower the barriers to DeFi usage.

The third stage is "AgentFi Intelligent Agents," marking the formation of strategy closed loops and autonomous on-chain execution. Agents can automatically complete perception, decision-making, and execution based on real-time market conditions, user preferences, and strategy logic, truly achieving 24/7 non-custodial on-chain capital management. Meanwhile, AgentFi can autonomously manage user funds without requiring user authorization for each operation, raising significant discussions about security and trust mechanisms, which also become unavoidable core issues in AgentFi design. Representative projects include Giza ARMA, Theoriq AlphaSwarm, Almanak, and Brahma, all of which have certain implementation capabilities in strategy deployment, security architecture, and product modules, forming the backbone of the current DeFi intelligent agent direction.

We look forward to the emergence of "Advanced Intelligent Agents" in the future, not only achieving autonomous execution but also covering complex cross-protocol and cross-asset business scenarios. This is our vision for the advanced form of future DeFi intelligence:

• Pendle yield rights trading: In the future, intelligent agents will fully take over maturity rotations and strategy orchestration, maximizing capital efficiency.
• Funding rate arbitrage: Cross-chain arbitrage intelligent agents are expected to accurately capture every opportunity in funding rate differentials.
• Swap strategy combinations: Swaps are key nodes in the multi-strategy yield paths of intelligent agents, achieving a leap in combined value.
• Staking and restaking: Intelligent agents will continuously optimize staking combination strategies, dynamically balancing yield and risk.
• RWA asset management: As the on-chain world welcomes diversified physical assets, intelligent agents will allocate globally stable yield assets.