Author: Shisi Jun
Review: 0xmiddle
Source: Content Guild - Investment Research
Introduction
Now, more than three years have passed since the explosion of DeFi Summer, and it has been more than half a year since the compliance benchmark ETF was approved. Has the situation changed?
Looking back, Ethereum's smart contracts have enhanced blockchain programmability, expanding blockchain from a single bookkeeping function to an infrastructure that supports various applications. Among many tracks, it is undoubtedly the most practical application scenario to decentralize traditional finance.
Let's take a look at the DeFi TVL data from DeFiLlama, where the total TVL of DeFi applications has exceeded 80 billion USD. In recent years, many public chains have emerged, and even Ethereum's L2s are siphoning off Ethereum L1's space, but currently, Ethereum still stably locks more than half of the total DeFi assets on a single platform.
Image source: defillama.com/chains
The initial ambition of DeFi was to disrupt the business models of lending, payments, insurance, etc., in the traditional financial system, allowing users to complete these operations without relying on banks or other traditional financial institutions. However, the TVL of DeFi has actually stagnated for a long time without breaking through the orders of magnitude.
Most opinions believe that DeFi is limited by the performance and cost issues of the Ethereum network, and therefore cannot achieve large-scale applications or complex financial scenarios. However, the DeFi ecosystems on various L2s and high-performance new public chains have not brought breakthroughs in DeFi scale; instead, they have led to liquidity fragmentation and decreased interoperability. Ethereum still retains the most complete DeFi ecosystem and the most sufficient interoperability, remaining the preferred platform for deploying DeFi projects.
Today, a new trend is emerging: a new type of DeFi paradigm based on AO—AgentFi. This innovation is breaking the limitations of traditional DeFi.
AO, based on Arweave's storage layer, has built a computing layer that supports parallel running processes, solving the scalability issue and achieving almost unlimited scalability. The combination of AO and Arweave is an implementation based on SCP (Storage-based Consensus Paradigm).
On AO, smart contracts exist as processes. By breaking free from performance limitations, anyone can run their own processes to proxy their financial actions, with consensus handled by Arweave's storage layer. This is exactly the foundation of AgentFi.
Will this new form of DeFi, namely AgentFi, replace traditional DeFi and become the new mainstream form of DeFi? Let me elaborate.
Limitations of traditional DeFi
In traditional blockchain architecture, block space is designed as a scarce resource, where users and applications must compete to acquire this resource. During network congestion, people need to pay more to compete for block space, which is the fundamental reason for performance limitations. Ethereum's performance limitations have become evident, with only about 30 TPS, making it quite restrictive. During peak periods, gas fees often skyrocket by dozens of times, which people have become accustomed to. In fact, L2s and most high-performance public chains also have performance ceilings; their ceilings may be higher, but it is still challenging to accommodate the scale of traditional financial businesses.
To save performance usage and reduce Gas costs for users, traditional DeFi was designed to use a single smart contract to host business assets and run financial operations. Since both funds and business logic are managed by a unified contract, it becomes difficult to achieve true diversification and personalized business operations. Although this design simplifies management processes and ensures consistency, it also deprives users of autonomy in business logic and financial operations, making it difficult to meet increasingly diverse user needs.
For developers, writing contracts must consider Gas calling fees and avoid writing complex contract code as much as possible. On Ethereum, the Gas Limit for an ETH transfer is 21,000 gwei, and for an ERC20 token transfer, it is 65,000 gwei. Slightly more complex scenarios, such as swaps, NFT transactions, and lending, require at least 300,000 gwei. If the business becomes even more complex, gas consumption will become unbearable for users. This greatly limits the creative space for developers and restricts the richness and innovation of DeFi.
To fundamentally solve the above problems, the market needs a more powerful infrastructure and accompanying financial systems.
Thus, AO was born, and AgentFi is a new exploration of the next generation of DeFi within the AO ecosystem.
AO: An infrastructure with almost unlimited scalability
AO is short for Actor Oriented, which, as the name suggests, is a decentralized computing protocol based on actor orientation.
In fact, compared to Ethereum, AO is closer to the concept of a world computer. I understand AO as a supercomputing layer, with the core goal of achieving trustlessness and collaborative computing services without scalability limits.
Let's take a look at the workflow diagram of the super-parallel computer built on AO:
Image source: AO white paper
Message Generation: Users or processes initiate requests by creating messages. These messages must comply with the specifications set by the AO protocol to be correctly transmitted and processed in the network.
Messenger Unit (MU) Relay: The Messenger Unit (MU) is responsible for receiving user-generated messages, acting as a router to direct messages to appropriate SU nodes in the network. During this process, the MU signs the messages to ensure data integrity.
Scheduling Unit (SU) Processing: When messages arrive at the SU node, the SU assigns a nonce to the message to ensure its order within the same process and uploads the message and nonce to Arweave's consensus layer for permanent storage.
Computing Unit (CU) Calculation: When the computing unit (CU) receives a message, it performs the corresponding computational tasks based on the message. After computing, the CU generates a signed result with the computation result and returns it to SU. This signature certifies the correctness and verifiability of the computation result.
So, where does consensus come from?
On AO, storage is equivalent to consensus. During the process of running, processes generate the transmission of messages, which are written to Arweave, thus creating a 'holographic state.' This means that the running state of the process can be verified. In other words, Arweave's immutable storage guarantees verifiability. This is somewhat counterintuitive, but if you fully understand the SCP paradigm, you can instantly understand this. If you still can't understand, you can think of it in terms of inscriptions.
In addition to verifiability, we also need to solve the question of who validates. With verifiability, anyone can provide validation services. On AO, applications can choose their own validation services based on their business nature, flexibly deciding their security. Coupled with the economic game of optimistic challenges, the reliability of validation can be guaranteed.
On the computer built on AO, applications are constructed from an arbitrary number of communication processes.
On AO, processes are not allowed to share memory, but they are allowed to communicate through the native messaging standards.
Since message transmission is asynchronous, by focusing on message transmission, AO has achieved a scalability mechanism similar to that of traditional Web2 distributed system environments.
This means that theoretically, AO does not have performance limitations.
For developers, public nodes can be chosen, but it is also entirely possible to run their own services using their own nodes. In this case, if performance bottlenecks occur, they can simply scale their own nodes, just like running Web2 services.
Moreover, this work model also brings additional benefits—computing nodes can provide computational power support for AI scenarios. We will have a chance to discuss this later.
What is different about AgentFi?
Unlike traditional DeFi, which hosts funds and runs financial businesses based on unified smart contracts, the concept of AgentFi allows everyone to run processes on AO computers and host their own funds, proxying their financial actions. What does this form look like? Let's take the leading DEX Permaswap on AO as an example for explanation.
In traditional DeFi, if Alice wants to exchange Token A for Token B, a liquidity pool needs to be created on the DEX, where a smart contract hosts the funds to provide the A/B token exchange functionality. The exchange rate for the transaction is determined by the market-making curve adopted by the smart contract (e.g., x*y=k). In Permaswap, it takes this form, where each LP uses their own agent process to host their market-making funds and customize the market-making curve and strategies. Of course, LPs can also adopt an 'extreme market-making strategy'—simply placing a limit order.
In fact, we find that Permaswap can integrate both AMM and order book trading forms. For users, when TA initiates a trade, what matches them to help complete the trade may be AMM, a limit order, or even both.
Overall, AgentFi has three characteristics:
1. Self-custody: Users host their own funds and execute their own trading strategies through their controlled agent processes, rather than entrusting them to a unified contract.
2. Personalization: Users can flexibly set their own financial business parameters through their controlled agent processes. This means it's akin to users running their own exchanges, customizing trading strategies and rates. If expanded to lending services, it can be understood as users running their own banks, customizing interest rates. Furthermore, users can completely run custom financial strategy programs using self-custodial processes, which can even be intelligent strategy programs integrated with AI.
3. Peer-to-peer: Matching between supply and demand parties, no longer following the traditional DeFi point-to-pool model, but returning to a peer-to-peer model.
On Ethereum, there is a distinction between contract accounts (CA) and external accounts (EOA). Different financial scenarios are implemented through different contract codes, and financial actions require human intervention. On AO, however, it is a concept oriented towards Agents, where different Agents can implement different functions, and financial actions can be proxied by Agents. I believe the concept of AgentFi is more like building blocks, allowing for a richer decentralized financial ecosystem.
When there are a large number of self-custodial processes, how can they communicate with each other and have composability? This leads to the mention of the FusionFi Protocol, which is a development standard and communication specification for agents on AO. Almost all financial services can be abstracted as the circulation and processing of bills, and the FusionFi Protocol defines a standard format for these bills. With such a standard, complex and diverse financial forms can achieve integration. Developers can implement various financial services such as exchanges, lending, futures, and even stablecoins based on FusionFi standards. In the future, the FusionFi Protocol can reference industry standard proposals such as BIP, EIP, and NIP to involve more people in jointly formulating protocol standards and promote the sustainable development of the ecosystem.
I will write a separate article detailing the FusionFi Protocol.
Summary
The performance and cost issues of Ethereum have constrained the current pace of DeFi development. Although the scalability of L2s and new public chains has been effective, there still exists an invisible ceiling limiting the development of financial services.
To completely break this ceiling, a network different from the traditional blockchain paradigm—the AO super-parallel computer—has emerged. Due to AO's infinitely scalable performance, AgentFi has become possible. Users can run their own processes, manage their funds themselves, and customize financial services.
The agent-oriented financial model has a broader range of applications compared to traditional DeFi.
Data source:
1. Interpretation of Ethereum TPS
https://www.chaincatcher.com/zh-tw/article/2102262
2. Ethereum transaction Gas usage statistics
https://etherscan.io/gastracker
References:
1. Technical details of ao super-parallel computers
2. AO Protocol: Decentralized, permissionless supercomputer
https://x.com/kylewmi/status/1802131298724811108
3. Smart Finance: From AgentFi to FusionFi
https://www.notion.so/permadao/AgentFi-FusionFi-6461feb8915c4ea5a1252eca80aa6a4a
