Nansen：AI 會贏過不用代理人的交易員

0:00推理已便宜到能當原料

0:00早安。歡迎來到新加坡。希望大家早上過得不錯。AI 會不會贏過人類交易員？先舉手看看。站 AI 這邊的？好。站人類這邊的？也有幾隻手。接下來 15 分鐘我們會看答案。我要說的是：我認為代理人交易的黃金時期就要開始。Good morning. Welcome to beautiful Singapore. Hope you've had a great morning so far. Will AI outperform human traders? Maybe let's start with a show of hands. Who's on the side of the AI? AI, show of hands. All right, who's on the side of the humans? Okay, a couple of hands there as well. Very good. We shall find out in the next 15 minutes whether AI will outperform human traders. I am here to tell you that I believe a golden era of agentic trading is about to start.

0:49為什麼？因為我們現在有一種新資源，幾乎像商品：隨時可取用的智慧。很多人會覺得這很怪。你可以把推理、智慧、token，接到原本由人、知識工作者在做的流程裡。先看幾個例子。這是 OpenRouter 上最貴的寫程式模型。And why do I believe that? Well, we now have this new resource, a new commodity almost, which is intelligence on tap. This is a crazy concept for a lot of us. You can actually take inference, intelligence, tokens, and channel them into workflows and activities that we do and normally have been doing with human beings, knowledge workers, and so on. Let's look at a couple of examples. What you're seeing here are the most expensive models on OpenRouter for coding.

1:40假設你在做軟體工程：寫應用、改應用，或任何跟軟體有關的事。成本基本上用美分在算。也許會到幾美元，但跟五年前已經是另一個世界。這不代表軟體工程師沒工作了。角色變了：現在很少工程師還自己一行一行寫程式。Imagine you're doing something that involves software engineering. You need to code something. You're creating an application or you're making a change to an application or anything software-related. The cost to do that is basically measured in cents. Now, maybe you'll pay a few dollars, but this is a completely different world than what we had even just five years ago when it comes to software engineering. Now, it does not mean that software engineers are out of a job. It means that they have evolved their roles, and of course, very few software engineers are now actually writing the lines of code themselves.

2:29交易也是電腦上的知識工作

2:29這只跟寫程式有關嗎？再看 Artificial Analysis 的資料。橫軸是成本，縱軸是完成各類任務的指數。很多模型非常便宜，知識工作（寫程式、研究、在電腦上做的事）成本仍是美分。那交易和投資總該不受影響吧？那完全不同？交易到底是什麼？Now, is this only relevant for coding? Let's look at some of this data from Artificial Analysis. And here we're seeing the cost on the x-axis and the index of performing various tasks on the y-axis. And you can see how incredibly affordable a lot of these models are. Again, we're talking about cents to perform knowledge worker tasks, right? Coding, research, various things that you would do on a computer. But surely this is going to have no impact on trading and investing, because that's completely different, right? Well, what is trading?

3:20你坐在電腦前，做研究、消化資訊、按按鈕、下單。聽起來簡單，實際難一些，但本質上就是在電腦上做知識工作。那我們是不是直接跟代理人說：「幫我賺錢，別犯錯」？用 Manus、ChatGPT、Claude 就好。有人試過嗎？我試過，結果不好，不建議。不過還是有人會去試。You sit on a computer, you research, you digest information, you click buttons, you execute trades, right? It sounds really simple. It's a little bit harder than the way I'm explaining it, but fundamentally, what you're doing is knowledge work on a computer. So that means we can just tell an agent, right? We can just say, "Hey, make me money and make no mistakes." You can use Manus, you know, ChatGPT, Claude. Has anyone tried doing this, by the way? I've tried doing it. It didn't go very well. I do not recommend doing that. But some people, you know, are adventurous and might try it.

4:09我覺得有更好的做法。先講：Manus、ChatGPT、Claude 這些，看不到市場，也看不到鏈上。比較好的作法，是給它能看見鏈上發生什麼事的產品。用 Nansen 的 API，幾秒就能讓代理人看見鏈上；不只是眼睛，還有手，能去執行交易。若你想拿現成的 AI 產品做代理人交易，給它鏈上的眼睛和手，並不是壞主意。I think there's a better way to do this. Well, maybe first of all, Manus, ChatGPT, Claude, etc. They don't have eyes on the markets and on-chain. So a good idea would maybe be to give it a product that allows them to see what's happening on-chain. And with our API at Nansen, you can literally, in seconds, give eyes to your agent, eyes on-chain. And actually, it's not just eyes on-chain, it's also hands on-chain. So if you are of the adventurous sort, and you want to try agentic trading with existing AI products, it's not a bad idea to give your agent eyes on-chain and also hands on-chain to execute your trades.

5:02Nansen AI：口袋裡的代理人交易

5:02也有人不想自己做代理人，想用已經打包好的產品。這就是我們的手機 App「Nansen AI」：全天可帶在身上的代理人交易。它是內建代理人的交易 App，我們做了大量評測和品管，確認它能幫忙做鏈上交易。背後是 Nansen 超過 5 億個已標記地址，可以看出地址背後是誰，交易功能也做在裡面。But other people might not want to build their own agents, right? You might want to actually use a product that already has all of this stuff packaged up for you. And so let me introduce you to our mobile app called Nansen AI: agentic trading in your pocket 24/7. Basically, it's a trading app that has an agent built into it, which has been battle-tested. And we've run an insane amount of evaluations and quality assurance to make sure it's actually very good at helping you with on-chain trading. It's powered by Nansen's 500 million-plus labeled addresses, meaning you can see who is behind the different addresses, and it comes with trading built in.

5:57所以不必另接錢包、把資金轉去別處，是整套合在一起的產品。你可以試試看。我們在這上面做了很多研發。你可能會想：這不就是套一層 ChatGPT？比那難一些。我們做了 harness，試過多種模型，每個新模型都做評測和品管，模型變好、隨時可取用的智慧變好，我們就換。So you don't need to connect separate wallets, send funds elsewhere. It's a fully integrated product. So this could be a good idea to try out. Now, we put tons of research and development into this product, and you might think, well, is it just a ChatGPT wrapper? Well, it's a little bit harder than that. We've basically built a harness, and we've experimented with a range of different models. We run evals and quality assurance on all of the different models that come out, and we've actually evolved and changed them out, of course, as models get better, as the commodity of intelligence on tap gets better.

6:43現在用這個產品，不必自己選模型。我們會把你的提示自動送到我們認為最適合回應、也最適合幫你的模型。研發花了很多力氣。產品有兩個特性。第一，它在鏈上，內建自行保管錢包。我們不是交易所，不是中心化交易所，是技術服務提供者，把 App 和裡面的能力交給你。第二，它是代理人式的：你不必在圖表和表格上點來點去。And if you use the product now, you don't even have to think about which model to use. We just automatically route your prompt to the model we think is best to respond to your prompt and to help you. So there's a ton of research and development that's gone into this product. There are two characteristics of this product. Number one, it is on-chain. It's a self-custodial wallet that's built in. We are not an exchange. We are not a centralized exchange. We are a technology service provider. We give you this app and all of the magic that goes into the app. And the other part is that it's agentic, which means you don't have to click around too much on charts, on tables.

7:30你可以直接問代理人：鏈上在發生什麼？聰明錢在買哪些代幣？過去 15 分鐘新上的迷因幣有哪些（如果你在看這個）？公開股票裡，聰明錢現在在買什麼？也能看股票的財務表現。這些都裝在裡面，不必自己組。做簡單的產品其實很難。這是一支直覺、好用的手機 App，但背後要做的事很多。對我來說，這是在爬信任階梯的起點。想像你要上自駕車。You can just ask the agent about what's happening on-chain, which tokens are smart money buying? What are the latest meme coins that launched in the last 15 minutes, if that's what you're into? Or which public equities are smart money buying right now? You can even get financial performance on stocks through this app. So it has a lot of this stuff built in, batteries included. You don't have to build it yourself. And it's very hard to build a simple product. This is a mobile app. It is very intuitive and easy to use, but there's so much stuff that goes into building this product. And for me, what this really is, is the beginning of walking up a trust ladder. Imagine that you are going into a self-driving car.

8:24自駕交易員與 Nansen Agents

8:24如果沒聽過這家公司，也沒聽過別人坐過，要坐進去會不舒服，尤其是坐後座。你可能想坐有方向盤的位子。我們的手機 App 也讓你留在迴路裡：你大致還在駕駛座，交易要不要執行，仍由你拍板。問題是：怎麼走到信任階梯頂端，把交易交給能自主代你執行的代理人？這就是「自駕交易員」。If you had never heard about the company, and you've never heard about anyone else taking a ride with this self-driving car, I think you'd feel uncomfortable getting into it, especially getting into the back seat. You might want to sit where the wheel is, right? And similarly, our mobile app keeps you in the loop. You're sitting somewhat in the driver's seat. You are calling the shots on what trades get executed. But I think the question is, how do you ultimately get to the top of the trust ladder, where you can actually delegate the trades to an agent that autonomously acts on your behalf? This is the concept of self-driving traders.

9:09我們現在在做的產品，不是人在挑代幣，而是代理人在挑，因為我們認為 2028 年以後會是這樣。我覺得 2027 年會有轉折，交易會出現類似 Cursor 或 Lovable 的時刻；到 2028 年，用代理人交易會變成常態，就像軟體工程師現在用代理人寫程式。這個產品叫 Nansen Agents，可以登記等候名單，這裡有 QR code。這就是自駕交易員。And what we are working on now is a product where humans are not picking the tokens, but agents will, because that's what we believe 2028 and beyond will look like. I think there's going to be an inflection moment in 2027. I think you'll have a Cursor or a Lovable moment for trading happen in 2027, and in 2028, the norm will be that we are trading with agents, just like software engineers are coding with agents today. So this product is called Nansen Agents. You can sign up to the waitlist for it, and I have a QR code for you here. This is self-driving traders.

9:59這是能自主運作的交易代理人。策略不是我們給的，是你自己寫，但可以回測，也可以用模擬交易、不放真金。對交易代理人夠放心之後（也許改過好幾輪），再注資，讓它代你自主運作。你可以進去看它在做什麼，介入、暫停、微調，但基礎設施大概不是你想自己做的。我們把 Nansen 六年多累積的能力（超過 5 億個已標記地址、鏈上資料、交易執行、代理人能力）包進同一個平台，讓你做自己的自駕交易員。This is autonomous trading agents. We do not give you the strategies. You create the strategies yourself, but you can backtest them. You can run them in paper trading mode without putting any capital at risk. And once you are comfortable with the trading agents, perhaps after multiple iterations, you can fund the trading agent, and it operates autonomously on your behalf. You can go in and see what it's doing. You can intervene, you can pause it, you can tweak it, but it's all of the infrastructure that you probably do not want to build yourself. So we take all of these amazing capabilities from Nansen that we've built out over the last six-plus years, all of our 500 million labeled addresses, all of the on-chain data, the trade execution, the agentic capabilities, all of that packaged up in one platform where you can create your own self-driving traders.

11:03目標是做出能用代理人在鏈上交易所有東西的產品。回到一開始的問題：我們認為，AI 會贏過不用代理人的人類交易員。但關鍵是承認，人在這個時期會有新角色。在代理人交易的時代，我們認為人會跟代理人一起交易，不是跟代理人對打，也不是人類對上 AI。And the goal here is really to build a product where you can trade everything on-chain with agents. Because, to go back to the very question that we started with, we believe that AI will outperform human traders that do not use agents. But the key is to acknowledge that the human gets a new role in this era. In the golden age of agentic trading, we believe that humans will be trading with agents. They're not trading against agents. It's not humans versus AI.

11:48四件事還得先解決

11:48這其實是人承認，交易和投資的角色變了，而那是我們想做的未來。再講幾個要讓這件事發生、必須處理的難題。這很難，拆成幾個子問題。第一，平台得讓你比手動交易更有機會把投資組合做大。我們永遠不能保證用這個產品的人會賺錢。有人跟你保證，你就該知道他不懂自己在說什麼。It's actually humans who are acknowledging that we have a new role in how we trade and how we invest, and that is the future we want to build for. Let me touch on some of the challenges we need to solve to make this happen, because this is a very hard problem, as you can imagine. Let's break it down into a couple of sub-problems that we have to focus on. Number one. We have to give you a platform where you have a better chance at growing your portfolio than if you're trading manually. We can never guarantee that people using our product are going to make money. If anyone is guaranteeing that, you should recognize that they don't know what they're talking about.

12:34市場不是這樣運作的。但現實是，很多交易產品給使用者的勝算很差。我不點名。你可以上 X 看有人在抱怨用某些產品虧了多少。目標是讓人更有機會做得比別人好。第二，你得做出可靠的代理人系統。代理人是非確定性的，能力強、彈性大，但會失控。做這個產品要處理很多細的技術問題。That's not how markets work. However, the reality is that many trading products give users a really bad chance at doing well in markets. I'm not going to mention any names. You can go on X and see people complaining about how much money they're losing on using various products. The goal here is to give people a better chance at outperforming. Number two. You have to build a reliable agentic system. Agents are non-deterministic. They are very powerful and flexible, but they can misbehave. And there are lots of subtle technical issues that we have to deal with in building this product.

13:20連回測這種基本的事，對代理人都很難。因為代理人可能發現自己在被回測、這是模擬。所以你得騙過它，讓它以為自己真的活在 2026 年 1 月 3 日。聽起來蠢，但你得把它留在「母體」裡。這類小事很多。代理人要有可靠的環境。第三，還得划算，因為隨時可取用的智慧是要花錢的。每次提示 AI 系統，提供 token 都有成本。這不是無限資源。怎麼讓你用到這種智慧、又不會把錢燒光，得想清楚。Even something as basic as backtesting is very hard with agents. Because the agent can find out that it's actually being backtested and it's a simulation. So you have to trick the agent so it believes it is actually operating in January 3rd, 2026, which sounds silly, but you have to keep it in the Matrix, basically. Lots of small challenges like that. You have to have a reliable environment for the agents. Number three, you need to make it economical, because intelligence on tap does actually cost money. Every time you prompt an AI system, it costs money to serve the tokens. So we don't have this infinite resource. We have to be thoughtful about how to actually give you access to this intelligence on tap without breaking the bank.

14:13第四，也許最有意思：還得好玩。我認為主動交易的人，很大一部分原因不是財務，是娛樂。錢其實可以放進 S&P 500 指數基金，坦白說多數人可能該這麼做。但有些人願意承擔風險、找點樂趣，喜歡主動交易。做交易代理人，不能把樂趣拿掉。你得給人一種真的好玩的使用經驗。早上醒來會想看交易代理人做得怎樣。And number four, which is maybe the most interesting part, you have to make it entertaining. I believe that people who trade actively, a big part of why they do it is not just financial reasons, it's also entertainment. We could put our money into the S&P 500 in an index fund, and frankly, most people should probably do that. But some of us who want to take the risk and have some fun, we like to actively trade. So by building trading agents, we don't want to take away the fun and the entertainment. You have to give people a new user experience that genuinely is entertaining and fun. You want to wake up in the morning and see how your trading agent is doing.

14:59所以也得好玩。最後，想清楚我們處在什麼時期。我們現在有隨時可取用的智慧。對我來說很明顯：接下來幾年，這會徹底改變交易和投資。我會鼓勵你站在對的那一邊。謝謝。So you have to make it entertaining as well. Now, to wrap up, think deeply about the era we are in. We now have intelligence on tap. And to me, it is very obvious this is going to completely change trading and investing in the next couple of years. And I would encourage you to be on the right side of history. Thank you.