← 回到 TOKEN2049 Singapore 2026

Mysten Labs 共同創辦人 Adeniyi Abiodun:機器經濟時代來臨,Sui 如何搶佔先機

The Machine Economy Won't Wait

場次摘要

Mysten Labs 共同創辦人 Adeniyi Abiodun 介紹 Sui 為 AI 代理人規劃的支付、隱私與風險管理功能。他說明吞吐量進展、穩定幣轉帳及威脅監控系統,也談到仍在規劃中的鏈上保險與企業合作。

值得記下的 5 件事

  1. [2:15]
    自然語言與 AI 代理人可能改變人機互動,降低對傳統介面的依賴
  2. [4:26]
    代理人交易需求若持續增加,金融基礎設施將面臨更高吞吐量要求
  3. [6:25]
    Sui 宣布吞吐量突破每秒 4,000 萬筆,隱私交易則已在測試網運作
  4. [9:00]
    Sui Threat Intelligence 提供鏈上威脅監控與風險資訊
  5. [12:42]
    零手續費穩定幣轉帳,可降低代理人與使用者的支付成本

逐字稿

從 Libra 到 Sui:打造新一代 L1

大家好,我是 Adeniyi Abiodun,是 Mysten Labs 的共同創辦人之一。我們公司位於美國灣區。過去我和我的同事們在 Facebook 參與了 Libra 專案,目標是打造一個能自動化網路上金流的系統,讓轉帳像寄 email 一樣簡單。那時候 LLM 技術還很不成熟,但我們已經做了很多自動化,最終希望讓金錢移動幾乎免費,並且讓代理人或人類能大幅自動化這些流程。My name is Adeniyi Abiodun. I am one of the co-founders of a company called Mysten Labs. We're based in the Bay Area. Previously, my colleagues and I were at Facebook building Libra, Project Libra, and the goal with Project Libra was to build a system to automate the movement of money around the internet. As easy as sending an email. Back then, the LLMs were quite poor. We did a lot of automation, but our goal was ultimately build a system where money can be moved pretty much for free, but at the same time ensuring that something that agents or human beings can automate to a large degree.

這是我的共同創辦人團隊。Evan 是 LLVM 的創作者之一,LLVM 現在幾乎所有 iPhone 都在用。Kostas 是頂尖密碼學家,開發了加密領域最早的 post-quantum 演算法之一。還有我們的前 CTO Sam Blackshear,以及首席科學家 George Danezis,他們負責設計我們系統裡所有的演算法。These are my co-founders. Evan, who's one of the creators of LLVM, the thing that runs in all iPhones today. Kostas, one of the top cryptographers in the space that has actually built one of the first post-quantum algorithms in crypto. And George Danezis and Sam Blackshear, who is our CTO before, and George Danezis, who's our chief scientist, who actually builds all the algorithms that we use in our systems.

接下來談談 Sui。我們採取了非常本質性的做法。當時我們還在 Meta 時,就開始打造一個全新的 L1,目標是把 Facebook 所有的資料放到分散式帳本上,讓使用者能把資料借給 Facebook,最終自己也能獲利。我們後來決定離開 Facebook,在外面創業,這個計畫就叫做 Sui。Sui 是新一代 L1,核心理念是使用者擁有自己的資料、用自己的規則進行交易,並且能自動化網路金流。So moving on from that, right. Sui, we took a very first principles approach. We're building a new L1 while we're in Meta, and the goal of this L1 was to take all the data that Facebook has, put it on a distributed ledger, use its own data, and you can largely treat Facebook more as a marketplace where users lend their data to Facebook and then ultimately can monetize it on their own. We decided to leave Facebook and build that outside the walls, and it's called Sui. Sui is a next-generation L1 built with the principle of users own their own data, users transact on their own rules, and you can build automated rules around the movement of money around the internet.

我們設計 Sui 的時候,希望使用者不用再煩惱私鑰。他們可以直接用現有的 Google、Facebook 等帳號,這些帳號就像鏈上的帳戶一樣。我們還採用物件模型,讓你能用物件而不是帳戶來建模現實世界。你沒辦法用帳戶本位來描述世界,但用物件系統就可以。我們採取了全新的架構。So we built it with the idea that users don't have to worry about private keys. They can just use their existing account, whether it's Google, Facebook. Any account they use today should be able to treat that as almost like an account on chain. We also built it with an object model, which means that really you can build or model the real world with objects rather than accounts. You can't model the world around an account book, but you can certainly model the world around object systems. So we took a very brand new approach to that.

代理人金融的崛起與未來互動模式

為什麼代理人金融(agentic finance)是必然趨勢?如果我們快轉五年後,你會覺得買網域名稱很蠢,也不會再去網站找資訊,更不會有人花幾小時坐在電腦前幫你設計 UI。因為未來你跟產品、品牌、服務互動的主要方式會是用文字、自然語言。你會直接跟電腦對話,或像 Zuck 展示的那樣用隨身裝置,或用 Muse agent、Perplexity 等各種 agentic system,直接溝通你的目標和需求。Why is the agent finance, why is it inevitable? The reason why agent finance is going to be inevitable is if I take you into five years' time. I think in five years' time, the idea of buying domain names would just be stupid. You'd not be doing it. The idea of going to a website to get information or someone sitting down in front of a computer for hours on end to design a user interface for you will be a thing of the past. Why is that? The reason for that is the prime measure, or the prime means by which you use to interact with products, brands and services will be through text or through natural language. You will be talking to a computer or through a pendant that you hold, as Zuck was showing. Or with a Muse agent that they've launched, or whether it's Perplexity or anything else that is available to you as an agentic system, and you will have a conversation about what your goals are and what you want to do.

這些代理人的工作就是幫你在網路上找機會,並且幫你完成。未來你不會再打 www,滑一堆文字找資料,這些都會過時。我們正邁向後瀏覽器、甚至後網際網路的世界。代理人會成為你獲得可驗證結果的主要方式。That agent's job will be to navigate the internet and find those opportunities for you and then make that so. So the idea of going to www dot will be so old it will not happen, and the idea of trying to scroll through a bunch of text to get information would also no longer be the case. So I think we're going to a world where it will be a post-browser world, and certainly we're going to go to a post-internet world to a large degree. Agents will be the prime means of which you use intelligent systems to get provable outcomes for you.

金融服務的未來:從應用到基礎設施

那金融會變成什麼樣子?就像網站會消失一樣,金融產品、銀行服務也會消失在背景裡。你現在用 iPhone 滑一堆 app,未來這些 app 也會消失,介面會完全改變。銀行會變,金融服務的提供方式也會變。我們專注於把鏈上金融和代理人基礎設施結合起來。鏈上金融會成為把金融產品和服務分發給大眾的主要方式。Moving on to that as well, right? Like, what does that mean for finance? How is finance going to look in that world? As I mentioned, in the same way as which your websites are largely going to disappear because you interact with the internet differently, certainly the products and services and the banking services will also disappear into the background. Your iPhone today, you hold it in your hand, you scroll through a myriad of apps. That's also going to go away. Not the iPhone, but certainly the apps themselves. Also, the interface by which you talk to these services will also change. So banks are going to change. The way in which banking services are going to change. We're very focused in a world where we bring together a world of on-chain finance and also agentic infrastructure. On-chain finance, in our purview, in our thought process, is going to be the means by which you distribute financial products and services to the masses.

所謂資產代幣化,其實就是把資產轉換成機器能讀、能寫、能轉移、能代表你擁有的語言。代理人需要的基礎設施必須夠快、夠安全、能證明安全,還要能全球擴展。Visa 現在每秒能處理約 6.5 萬筆交易,這對人類來說已經很夠用,但代理人未來會以遠高於人類的速度交易。現在世界上代理人數量已經超過人類,而且他們做的事情只會越來越多。Tokenization is nothing more than a process of turning assets into a language that machines can read, write, and transfer and own on your behalf. So, infrastructure that agents need, they need to be fast, right? It needs to be safe. It needs to be provably safe. It needs to be scaled to the world that is needed today. Visa can process about 65,000 transactions per second. That's actually great for the world we live in today, where human beings do transactions. You swipe your card, maybe at Christmas sometimes it gets a bit busy and you have to try again. That works for today. But agents are going to transact at a higher level, at a much faster rate than human beings. Today there are more agents in the world than there are humans. And they will certainly do more things for humans than are useful today, but certainly in the future even more so.

Cloudflare 預估,未來全球代理人需要每秒十億筆交易的處理能力。現在 Cloudflare 服務的流量,大多數其實都是代理人而不是人類。現有的基礎設施和產品必須徹底改變,才能服務這些代理人。If you talk to Cloudflare, they estimate that you're going to need about one billion TPS to service the need of agents globally. So the velocity of transactions, even Cloudflare services today, the vast majority of what is hitting their services are agents, not human beings. So the kind of infrastructure and products and services that run on products and services today absolutely needs to change to take advantage and offer services to agents, as I mentioned.

新金融規則:可程式化、可收益、重視隱私

未來的金錢必須能隨時移動、可程式化,而且預設就要能產生收益。大家不會接受錢放著沒利息。你要能對自己的資產設定嚴格規則,這會越來越重要,當然隱私也很關鍵。傳統銀行轉帳時,銀行不會把你的所有交易紀錄給對方,這在中心化服務很正常。如果我們要進入去中心化、鏈上銀行服務的世界,也必須讓使用者能選擇性保護自己的資訊隱私。So these are the things you're going to need, right? Money needs to be movable. Money needs to be programmable. Money largely needs to be yield by default. So people will not accept a world where their money sits there and doesn't pay them anything. You need to be able to set strict rules around the money that you have, the assets that you own. It's going to be very, very important. And of course, privacy is key. So the world we're going to, the world we came from in banking, right, where I send money to a friend and, you know, my bank does not divulge my entire history to my friend, that exists in centralized services. In the same way, if we're going to move to a post-centralized world where banking services are offered on-chain, we also need to take care and ensure that users can keep their information private selectively.

今年我們會在 Sui 主網上推出鏈上隱私交易,現在已經在測試網上運行。昨天我們剛創下全球網路吞吐量新紀錄,超過 4000 萬 TPS。這些都是各種代理人在做交易、轉帳、聊天、退款等各種行為,全部平行進行,突破 4000 萬筆每秒。So this year we'll be launching private transactions directly on-chain. It's already in testnet, and yesterday we broke the world record for throughput for any network. We're over 40 million transactions per second. These were multitude of agents engaging in all sorts of activities, whether it's trading, whether it's sending money peer to peer, whether it's chat conversations having, whether it's refunds, you name it. These were happening all in parallel, and we cracked over 40 million transactions per second.

代理人時代的驗證與信任基礎設施

代理人會成為加密基礎設施的主要使用者,因為未來最需要的是可驗證和信任最小化的系統。你不可能請一億個代理人,每一筆都要人類驗證才能繼續,必須有信任最小化的層,讓所有行動能同時發生。As agents will be the prime adopter of crypto rails, because proof is going to be what is needed and verification is a bottleneck. You can't employ a billion agents and need a human to verify everything that happens before it can proceed. You need a trust minimized layer to ensure that these actions can happen at the same time.

所以我們在 Sui 打造了多層堆疊,包括收益引擎、橋接、保管庫、安全層、信用評分、風險評分等。你可以即時知道每筆交易或每個 DeFi 協議的風險有多高,也有基礎設施支援大量平行交易。但這些對一般使用者來說太複雜了,他們不會去網站看一堆文件,然後自己組合這些功能。So looking at the stack, these are going to be needed. These are the things that we're building into Sui. It is a yield engine, right? It's a stack of many layers. You have bridges, you have vaults. You have security layers that can give you credit scores. You can get scoring of how risky a transaction is versus another. You can get risk scores as to how safe a DeFi protocol is versus another. You certainly have the infrastructure that you need for primitives for vaulting. You have things that are going to enable you to actually do a multitude of transactions at the same time for parallelization. You basically have all the primitives and a multitude of layers, but the reality is this is really complicated for users, right? Users are not going to be using, going to a website, reading a bunch of docs to click through to figure out to turn a bunch of spaghetti together into something that's useful.

AI 代理人 x 金融決策:未來的使用者體驗

未來的金融體驗會長這樣:你在 Anthropic Claude 或其他 AI agent 上,直接說「我要投資 1.5 萬美元」,指定投資期間,AI 會自動幫你搜尋所有可用的投資機會,給你即時的風險評分和收益引擎。你根據這些智慧建議做決策,然後直接鏈上完成投資,馬上結算,不用自己研究一堆資料。這些代理人會把複雜的堆疊整合成使用者能用、能執行的簡單體驗。This is how finance of the future is actually going to work. So this is in Anthropic. You can actually run this in Claude. You can run it anywhere else. You and your Claude agent would be able to basically specify what you want to do. I want to invest $15,000. You specify the period that you want to make the investment. It will scour the internet for a number of investment opportunities that are available to you. It will give you the risk scoring in a yield engine directly. You would be able, as a user, to determine based on what the intelligence is afforded to you to make a financial decision on what investment you want to make. Separately from that, you'll be able to make that investment. It will be directly on-chain. It will be final instantly without having to do a bunch of research. So having all the scoring for risk, having all the availability of products and services on the internet that you're going to need, being able to make that intelligent decision instantly, and using that massive stack I just showed to you, agents will be that glue that tied complicated stacks together into something humans can actually use and execute at the same time.

鏈上安全、保險與自癒生態系

傳統銀行會承擔使用者的風險,這是大家用銀行帳戶的原因之一。不能只是把一切搬到鏈上,卻不解決安全問題。如果被駭、智慧合約出錯、系統掛掉,怎麼保障使用者安全?我們昨天宣布了 Sui Threat Intelligence,這是一個自動化系統,已經在 Sui 主網上運作,即時監控所有交易、合約、套件,偵測惡意行為。Complementary to this is if today I get security from my bank, my bank, the reason why I use a bank account is because they will take on the liability if something goes wrong, right? We can't move users from a world where we say, Oh great, everything is online, everything is on-chain, and can't take care of the issues around safety. What happens if there's a hack? What happens if there's a smart contract issue? What happens if there's an outage? How do we ensure that users can remain safe with the same guarantees they get with banks today in a world that's fully decentralized and automated? So I want to touch on what we announced yesterday that really takes care of this for consumers and institutions alike. So Sui Threat Intelligence, it's an autonomous system that runs on the Sui network today. It is monitoring the network in real time. It's looking at packages. It's looking at smart contracts. It is scanning the network for transactions that are malicious. It is scanning the network in live, in a live environment right now for transactions that may be trying to extract value more than it's giving.

目前 Sui Threat Intelligence 已經即時監控超過 4.5 億美元的資產,即時給每筆交易風險評分,已經幫使用者避免超過百萬美元的潛在駭客損失,並即時通知開發者修正協議漏洞。傳統審計只能看到靜態狀態,但應用一旦上線、互動起來,AI 其實比人類更會找漏洞。我們的系統能即時偵測、阻擋這些風險。So by the numbers, it's looking over $450 million live right now. It is giving risk scoring to transactions in real time today. It has saved just over a million dollars in potential hacks and has notified developers of issues found in their protocols in live environments today. So audits are great to see the current state of an application, but when an application starts to run, especially as it starts to interact. With other applications, we know AI is a better hacker than human beings. It can put together a number of vulnerabilities and detect them and put that in a package that can take advantage of a network instantly. So we have the best tool in regards to that.

很快 Sui Threat Intelligence 會直接跑在驗證者節點上,每個驗證者都能對每筆交易給出風險分數,DeFi 協議可以根據這個分數決定是否暫停、放行或阻擋交易。這種即時安全和智慧對 DeFi 很重要,能保障使用者資金安全。有了可量化的風險,就能原生在 DeFi 上設計保險產品,讓使用者存錢進保險池,知道損失風險是可控、可組合的。Soon, Sui Threat Intelligence will run on validators themselves, and every validator will be able to attribute a risk score per transaction. And each DeFi protocol can choose in looking at that risk score to hold a transaction, to pause it, or to make it go through. So giving that level of security and giving that level of intelligence to DeFi protocols is going to be very, very key in real time to make decisions to keep users' funds safe. And when you can bound risk of loss, you can start building insurance products around it, which is what banks really offer you. So Sui will offer the ability to build insurance products directly into DeFi natively, so you can deposit funds into an insurance fund, knowing that the risk of loss is bounded to something that's composable directly on-chain.

有了自動化風險系統,網路能自我修復、即時阻擋惡意交易,形成正向循環。這樣的生態系能即時偵測、處理詐騙,吸引更多資本進入 Sui。我們今天稍後還會有更多分享。未來代理人流程會比人類驗證更快,整個金融體系都會往代理人化發展,需要能跟上代理人速度的系統,才能為使用者創造價值。Now, in a world where we have autonomous risk systems, in a world where we have the ability for a network to heal itself, to halt transactions that are malicious, we actually have a flywheel because now you have a trustable ecosystem. That is impervious to the hacks that AI runs today, and certainly can react faster than human beings can ever react in a scenario where there's fraud or hacks that happen in real time. So we'll have an ecosystem that detects fraud, takes care of fraud in real time. Therefore, more capital is coming into Sui in regards to that. We have more to talk on that later on today. And certainly, we have agentic workflows that are going to run faster than humans can verify. So in a world where everything is moving towards agentic finance, you're going to need systems that work at the speed of agents to provide the value for consumers at that same rate as well.

Sui 的優勢與全球合作布局

我們相信鏈上金融和代理人系統的結合,就是 Sui 的定位和優勢。昨天我們剛創下 4000 萬 TPS 的新紀錄。在代理人交易速度遠超人類的世界,無論是幾分錢還是幾百萬美元,都需要一個永遠快速、便宜、穩定的層來處理支付。Sui 現在穩定幣轉帳完全免手續費,無論金額大小都不用付 gas,因為我們能隨著驗證者增加而線性擴展網路吞吐量。So we believe the marriage of on-chain finance and how it intersects directly with agentic systems is where Sui plays and where Sui wins. As I mentioned, we're now at over 40 million transactions per second. We broke the record yesterday. In a world where agents are going to transact faster than human beings, or need to transact, whether it's a quarter of a cent all the way to millions of dollars, you need a layer that is always going to be fast, cheap, and robust to process payments. Sui today does not charge money for gas for moving stablecoins. So to move any level of stablecoin, whether it's a cent all the way to billions of dollars, there is no fee. So it's the first network where agents pay nothing for moving money without any middleman required in the process to actually handle gas management. The network itself doesn't charge because we have the ability to scale throughput of the network as we add more validators. And of course. As an execution layer is very, very key.

我們打造的引擎,將吸引史上最多的代理人使用者。更重要的是,代理人會為使用者執行交易,變得越來越有用。昨天我們也宣布和 Google 合作,Sui 的驗證引擎將進入企業市場,Google 客戶可以用 Sui 來驗證代理人行為、設定規則、取得完整稽核紀錄。目標是讓代理人能大規模在生產環境部署,Sui 是唯一能支撐超大規模吞吐量的層。So we have an engine that we feel ultimately is going to lead itself to adopting more agentic users than has ever existed in any formal history that's ever existed for us. But more importantly for Sui, we believe that agents are going to transact for users. They're going to become more useful. And yesterday we announced our partnership collaboration with Google, where we will be taking our engine into enterprise, where Google customers will be able to run the Sui verification engine for agents to run in enterprise and set the rules of what the agents can or can't do and get an audit trail of every agent interaction. So our goal is to make agents deployable in production at scale. And Sui is the only layer that can handle the throughput of large hyperscalers.

同時我們也宣布和阿里雲合作,他們會用 Sui 的代理人支付,讓產品和服務能直接在鏈上銷售。未來你用代理人處理聊天機器人、退款、雲端服務買賣、銀行服務等,這些互動都需要能跟上網路速度的基礎設施,而我們用 Sui 打造了這一切。謝謝大家!At the same time yesterday, we announced the partnership and collaboration with Alibaba Cloud. They're going to be using Sui agentic payments to enable products and services to be sold directly on-chain using our infrastructure. So in a world where you're going to handle your chatbots using agents, you're going to handle your refund process using agents. You're going to handle buying and selling of cloud and platform services using agents. You're going to handle your banking service using agents. These kind of interactions demand infrastructure that ultimately scale at the speed of the internet, and we built that with Sui. Thank you very, very much. Thank you.

下載 Word 摘要(.docx)

更多來自 TOKEN2049 Singapore 2026

訂閱