Writing

DeFi Yield Optimization Evolution: From Auto-Compounding to AI Agents

DeFi Protocol Research Lab | Deep Dive | Data as of January 2026

In 2020, a programmer named Andre Cronje got tired of manually moving funds around every day chasing the highest yields, so he wrote an automation script. That script later became Yearn Finance, which managed $6 billion in assets at its peak.

Five years later, the DeFi yield optimization landscape has transformed completely. Yearn’s TVL has dropped to around $300 million, while a protocol called Pendle grew from obscurity to become a giant with $40 million in annual revenue, capturing over 50% market share in the entire yield category.

What happened in these five years?

The answer: The sources of yield changed, and so did the strategies.

First-generation protocols helped you auto-compound, second-generation protocols let you trade future yields, third-generation protocols let you earn multiple yields from the same ETH, and fourth-generation protocols let AI make decisions for you.

This article walks you through these four generations of evolution, understanding each generation’s core innovations, economic logic, and who truly emerged victorious in 2025.

Part 1: First Generation: Auto-Compounding Machines (2020-2021)

What Problem Did Yearn Solve?

DeFi in 2020 was chaotic. Liquidity mining had just exploded, with new “farms” appearing daily offering APYs of 100%, 200%, or more. The problems were:

If you had $100,000, spending an hour daily on manual operations might be worthwhile. But what if you only had $10,000? Gas fees might eat half your yields.

Yearn’s solution was straightforward: pool everyone’s money together and let the protocol operate automatically.

The Core Logic of the Vault Model

Yearn introduced the “Vault” concept. Deposit DAI into the DAI Vault, and the protocol does three things for you:

  1. Automatically finds optimal strategies: Monitors yield rates across Compound, Aave, Curve, and other protocols, moving funds to wherever yields are highest
  2. Auto-compounds: Periodically sells mined tokens, converts back to DAI, and reinvests
  3. Gas cost sharing: Operations for 1,000 users are combined into one transaction, each person bearing only 1/1000th of the Gas

At the peak of the 2021 bull market, Yearn managed $6 billion, with some Vaults reaching APYs over 50% (mainly from token incentives).

First Generation Limitations

But the Yearn model had fundamental problems:

Unsustainable yield sources. The high yields of 2021 mainly came from token incentives—protocols issuing their own tokens to “bribe” users into providing liquidity. Once token prices dropped or incentives stopped, yields collapsed. When the 2022 bear market arrived, many Vault APYs fell to 2-3%, barely higher than bank deposits.

Completely transparent strategies. Yearn’s strategy code is public, and fund flows are trackable. This means MEV bots could precisely predict when Yearn would sell mined tokens and front-run. Large Vault compounding operations were frequently “sandwiched,” with yields being skimmed. Yearn later introduced ySwaps service in V2 and Factory Vaults to address this—separating token swaps from harvests, executing through private channels like Flashbots rather than the public mempool, significantly reducing MEV losses. But this improvement came late and sacrificed timeliness (swaps might take days to complete).

No risk separation. Money deposited in Vaults had yields and principal bound together. You couldn’t say “I only want stable 5% yields without APY volatility risk,” or “I’m willing to take higher risk betting APY will rise.”

These problems drove the emergence of second-generation yield protocols.

Part 2: Second Generation: Yield Tokenization (2022-2023)

Pendle’s Core Innovation: Selling Yield Separately

Pendle was founded by TN Lee, launching V1 on Ethereum in June 2021, with V2 in November 2022 marking its maturation. TN Lee was previously a founding team member at Kyber Network as Head of Business, responsible for Kyber’s expansion in Korea, China, US, and Europe. Kyber was one of the earliest successful DEX aggregation protocols—this experience gave the Pendle team deep understanding of AMM design and tokenomics.

Pendle asked a simple question: Why must yield and principal be bound together?

Traditional bond markets have long had similar concepts—zero-coupon bonds and interest stripping. Pendle brought this concept on-chain.

Suppose you have a yield-bearing asset like wstETH (Lido’s wrapped stETH, with ~4% annual yield). Pendle splits it into two tokens:

These two tokens can be traded independently.

What’s the Use?

Sounds abstract, but it actually opens up many new strategies.

Lock in fixed yields. Suppose it’s January 2026, and you have 100 wstETH. There’s a wstETH pool on Pendle expiring June 2026 (each Pendle market has a fixed maturity, typically 3-6 months, with liquidity concentrated on a few popular dates). The current PT-wstETH price is 0.98 ETH.

You can split 100 wstETH into PT and YT, sell the YT, and hold only PT. At maturity, your PT can be redeemed for 100 wstETH. Calculate it out: you locked in 100 ETH of maturity value for about 98 ETH cost, roughly 4% annualized. No matter how wstETH yields fluctuate over these six months, your return is guaranteed.

Note: This is ETH-denominated fixed yield, locking in staking yield rate, not USD value. If ETH’s price drops, your fiat-denominated assets still shrink.

By the way, if you hold equal amounts of PT and YT (same maturity), it’s equivalent to directly holding wstETH—YT lets you collect staking yields, PT lets you reclaim principal at maturity. Pendle’s split is mathematically reversible.

Leveraged long on yield rates. Conversely, if you think wstETH staking yields will rise (say, expecting increased on-chain activity), you can buy just YT. YT prices are low (since they only represent future yields), essentially giving you leveraged exposure to yield rates with small capital.

Worth mentioning: while wstETH is used here for illustration, Pendle’s best-performing pools currently are mostly yield-bearing stablecoins (like Ethena’s sUSDe, Usual’s USD0, etc.), reflecting the trend shift in DeFi yield markets.

Points/airdrop speculation. In 2024-2025, Pendle’s hottest use case was trading points expectations. Many protocols (EigenLayer, Ethena, Blast, etc.) distributed “points” before token launch, which eventually convert to token airdrops. Pendle lets you trade these points’ value in advance—bullish? Buy YT betting on the airdrop. Don’t want risk? Sell YT to lock in yields.

Pendle’s 2025 Performance

Pendle’s growth has been stunning:

Metric Early 2023 Mid-2025 Growth
TVL $200M $4B+ 20x
Annual Revenue - $40M -
Yield Category Share <10% >50% -

Why did Pendle explode in 2024-2025? On the surface, it’s the “points speculation” narrative, but the deeper reason is Pendle caught a bigger trend: the rise of yield-bearing stablecoins.

In early 2024, yield-bearing stablecoins had only $1.5 billion in total issuance, just 1% of all stablecoins. By mid-2025, this grew to $11 billion, representing 4.5%. Of this $11 billion, over 30% (about $3 billion) was deposited in Pendle.

Pendle became the “liquidity hub” for the yield-bearing stablecoin sector. Whenever new yield-bearing assets appeared—whether Ethena’s sUSDe, EigenLayer’s LRT, or later Usual, Reserve—these protocols actively integrated with Pendle. The reason is simple: Pendle helps them acquire users and liquidity, while Pendle gets free asset sources. It’s a win-win flywheel effect.

In March 2025, Pendle launched permissionless asset listing, further amplifying this effect. Combined with Aave V3 integration (PT-USDe usable as lending collateral) and upcoming Solana and Hyperliquid support, Pendle is evolving from “Ethereum’s yield trading platform” to “cross-chain yield infrastructure.”

Traditional finance interest rate derivatives are a trillion-dollar market. What Pendle offers isn’t fake yields propped up by token incentives, but real tools for users to hedge or amplify interest rate risk.

Part 3: Third Generation: Modular Lending & Restaking (2023-2024)

Third-generation yield optimization has two main threads: one is the “modularization” of lending protocols, the other is “restaking.”

Modular Lending: Morpho and Euler v2

Traditional lending protocols (Aave, Compound) use “monolithic” design: all assets in the same pool, risk parameters set uniformly by governance. The problems:

Morpho proposed a “modular” solution:

It’s like going from “one big bank” to a “bank + asset management company” model. Curators can create high-yield strategies for institutions or conservative strategies for cautious users—each gets what they need.

In 2025, Morpho’s TVL exceeded $3 billion and was listed by Grayscale as a “Q3 2025 Priority Asset.”

Restaking: EigenLayer Creates a New Paradigm

Restaking is one of the biggest narratives of 2024-2025.

The principle is simple: you’ve already staked ETH with Lido earning ~4% yield—why not “reuse” that stake’s security to protect other protocols and earn additional yields?

EigenLayer realized this idea. You deposit stETH into EigenLayer, and it “rents out” that security to protocols that need it (called AVS, Actively Validated Services). These might be oracles, cross-chain bridges, data availability layers, etc.—they need staked assets for security but building their own staking network from scratch is too hard.

The result: Same ETH, earning two or even three yields.

EigenLayer growth data:

Time TVL
Early 2024 $1.1B
Mid-2025 $18B+

EigenLayer is now the second-largest DeFi protocol after Lido.

But here’s a key question: Where does the money come from?

Theoretically, restaking yields come from “security rental fees” paid by AVS. But as of late 2025, most AVS are still in early stages, generating far less revenue than expected. What users mainly earn is EIGEN token incentives and various points, not real money from AVS.

In the second half of 2025, as airdrop incentive effects weakened, Liquid Restaking Token (LRT) TVL showed notable declines. This reminds us: Restaking’s high yields currently rely heavily on token incentives—the AVS ecosystem’s business model hasn’t been fully validated.

Meanwhile, competition is intensifying. In early 2025, Paradigm-backed Symbiotic launched its mainnet, directly competing with EigenLayer. First-mover advantage is being diluted, and EigenLayer needs to hold its ground before the AVS ecosystem matures.

Liquid Restaking Tokens (LRT)

EigenLayer’s problem: deposited assets are “locked,” unusable in other DeFi protocols.

ether.fi, Renzo, KelpDAO and others solved this. They deposit into EigenLayer for you and give you a liquid token (like eETH, ezETH) that can:

This is “nested yields”—layer upon layer, pushing capital efficiency to the limit.

ether.fi’s TVL in 2025 exceeded 2.5 million ETH (about $10 billion), the absolute leader in the LRT space.

However, the ether.fi team clearly realized that restaking yields alone can’t sustain long-term development. In April 2025, ether.fi launched a Visa “Cash” credit card, letting users use eETH and other assets as collateral for daily spending, repositioning from “liquid restaking protocol” to “crypto-native neobank.” This strategic pivot somewhat reflects the LRT sector’s anxiety about single yield sources.

Part 4: Fourth Generation: Intent-Driven & AI Agents (2024-2025)

The fourth generation is just beginning, but the direction is clear: Let users just say “what I want,” without worrying about “how to do it”.

Intent-Driven Trading

Traditional DeFi trading is “imperative”: you tell the protocol “swap 1 ETH for USDC on Uniswap,” and it executes. The problems:

Intent-based trading is “declarative”: you just say “I want to swap 1 ETH for as much USDC as possible,” regardless of which chain it’s on. Then, a professional Solver network competes to execute your order, finding optimal paths and bearing MEV risk.

CoW Protocol pioneers this direction. Its Solver network:

In 2025, CoW Protocol’s daily trading volume frequently exceeded $100 million.

AI Agents: DeFAI

The more radical direction is having AI manage your positions.

DeFAI (DeFi + AI) is a 2025 narrative. The core idea:

Still early, but several projects are working on this:

The challenge: What if AI makes mistakes? Who’s responsible for losses? How do you reconcile smart contract determinism with AI uncertainty?

No standard answers yet, but this is yield optimization’s endgame—users don’t need to understand underlying protocols at all, just choose their risk preference.

Part 5: Who Emerged in 2025?

Looking back at these four generations, the 2025 market landscape is clear:

Protocol Category TVL (2025) Core Advantage
Pendle Yield Tokenization $4B+ $40M annual revenue, absolute yield category leader
EigenLayer Restaking $18B+ Second-largest DeFi protocol after Lido
ether.fi Liquid Restaking 2.5M+ ETH LRT market leader
Morpho Modular Lending $3B+ Institutional-grade risk management, Grayscale priority
Ethena (sUSDe) Yield-Bearing Stablecoin Peaked at $6B (declined after Nov 2025) Delta-neutral strategy, yields fluctuate with funding rates

Ethena: A Special Case

Ethena’s sUSDe deserves special mention. It’s a “yield-bearing stablecoin” that earns about 11% annual yield just by holding.

Where does the yield come from? Ethena uses user funds for a delta-neutral strategy:

  1. Hold stETH (earning staking yield)
  2. Simultaneously short ETH perpetual contracts on centralized exchanges (earning funding rates)

Because ETH perpetual contracts have positive funding rates most of the time (more longs than shorts means longs pay shorts), Ethena can earn consistently.

In 2025, sUSDe became Pendle’s highest-volume traded asset—users trade sUSDe’s future yields on Pendle, forming a massive fixed income market.

But Ethena has risks. In November 2025, due to funding rate compression, sUSDe’s yield dropped from 11% to 5%, and TVL fell 50% in one week. High yields come from complex strategies, and complex strategies have tail risks.

Part 6: Insights & Risks

Trend Assessment

DeFi yield optimization is moving from “wild growth” to “refined operations”:

  1. Yield sources changed: From “token incentives” to “real economic activities” (staking yields, lending spreads, funding rates)
  2. User needs diverged: Some want fixed yields (buy PT), some want leveraged speculation (buy YT)—modular design satisfies different needs
  3. Capital efficiency pushed to limits: Restaking + LRT + yield tokenization—same asset “nested” multiple layers
  4. Professional specialization emerged: Solver networks, risk curators, AI agents—regular users don’t need to make all decisions themselves

Risk Warnings

But high yields always come with high risks:

Stacked smart contract risk. Your assets might be simultaneously exposed to Lido, EigenLayer, ether.fi, and Pendle—four layers of contracts. Any layer having problems collapses the entire chain. The March 2023 Euler Finance hack for $197 million is a cautionary tale.

Yield tokenization interest rate risk. YT value depends on future yield rates. If yields suddenly drop (like Ethena in November 2025), YT can go to zero.

Restaking slashing risk. If EigenLayer’s AVS has problems, your staked assets might be slashed. The slashing mechanism hasn’t been triggered at scale yet, but the risk exists.

Liquidity risk. LRTs (like eETH, ezETH) can depeg from ETH during extreme market conditions. In 2024, ezETH briefly depegged, causing leveraged positions to be liquidated.

Practical Advice

If you want to participate in these yield strategies:

  1. Start with PT: Pendle’s PT is the closest thing to “fixed income,” with relatively controllable risk. Choose PT of mainstream assets (stETH, sUSDe) with shorter durations (3-6 months).

  2. Be conservative with restaking: Don’t put all your ETH into EigenLayer. LRTs are usable, but don’t exceed 2x leverage.

  3. Understand what you’re earning: Staking yields, lending spreads, funding rates, token incentives—each yield source has different risk characteristics. Token incentives are least sustainable.

  4. Monitor health factors: If using any leverage (E-Mode, looping), you must monitor health factors and collateral prices. No one waits for you to top up during extreme markets.

Back to the Beginning

In 2020, Yearn helped users auto-compound with APYs often exceeding 100%, but most was bubble from token incentives.

In 2025, Pendle lets users trade future yields, EigenLayer lets the same ETH earn multiple yields—yield sources are more real, strategies more complex.

The essence of yield optimization hasn’t changed—trading risk for yield.

What’s changed: Risk can be more precisely priced, divided, and transferred. You can choose to bear only principal risk (hold PT), or only yield volatility risk (hold YT), or outsource risk to professional curators or AI agents.

This is DeFi’s mark of maturity: not making everyone a trader, but letting everyone find products matching their risk preference.

But always remember that old saying: If you don’t know where the yield comes from, you are the source of yield.

References:

Updated: January 2026

Disclaimer: For research only, not investment advice

❦ ❦ ❦