The Autonomous Agentic Web | Yat Siu at BEYOND Expo

Hello. Hello. All right. Thank you. So, I'm going to talk a little bit about the autonomous agentic web. I think a lot of you already have been hearing about a lot of AI, but I want to give it a little different touch. I think you notice already we're now entering this new era where we went from human-to-web interaction to agent-to-agent interfaces. What it really means is that most people in the world today are essentially using a kind of ChatGPT mode where they do a question and answer type of setup. But autonomous AI agents completely change the interaction layer because they are interacting and doing things on your behalf, and they have full long-term memory. The impact isn't only because they're autonomous; it will completely redraw the web as we see it because we no longer browse the web the way we used to. For example, the last time you used Gemini, you went to search something on the web and didn't actually click through to the website. A year and a half or two years ago, you would have always clicked through. The average number for most websites today is 80% less traffic that is clicked through because it's good enough if ChatGPT or Perplexity or Google gives you a summary. So if you're not agent discoverable, if you're not AI discoverable, you become irrelevant. Everything moves towards APIs, CLIs, MCPs, all sorts of ways in which an agent can talk. But the real impact isn't that you have to become agent visible; the real impact is that there's about $950 billion of advertising and attention-based revenue that is now completely under threat. Facebook and Meta make almost 95% of their revenue in advertising, and Google 77%. That revenue is distributed across all the other websites that get a cut, and that's all under threat because if you don't go to a website, you don't see the ads in a traditional way. We believe that this entire shift of almost a trillion dollars of revenue will ultimately go into an invocation economy, which I'll go into later. The invocation economy is essentially a transaction-based economy similar to when you pay for a subscription. A small amount of that money will go straight to the website or content company or application that you don't know, but on the back there's a payment rail that takes care of that. So that is how we think the transaction system will change. The superpower of AI agents is that they are extremely good at code generation, particularly over the last six months. While many of us use them for research and onchain ops or analysis, the code generative capability is what sets them apart. But that also means that 80% or 90% of all apps are going to disappear. You're not going to need them anymore because you can build an application as you need it on the fly in literally five minutes. Everything in the app store, software in a traditional way, we think is going to go away. For instance, this is an example we did with one of our own agents, with Animoca Minds. We told it to make a quick game and it did it in five minutes. It worked perfectly well. It can do it for everything: special applications, spreadsheets, SaaS applications. The traditional attention economy is going to go away. Everything is going to move towards frictionless microtransactions and extreme customization. When you look at the arc of internet history, everything was about customization and personalization. This personalization is no longer just about what content I get, but how I shape the experience the way I want. You're not going to go to a newspaper or website to get your content; the content is packaged in your personal newspaper. The blueprint for 2026 and going forward isn't what LLM you're using. Most companies today use multiple LLMs and integrate them together. The difference between Gemini, Anthropic, DeepSeek, or Xiaomi is not that much. For most of us, we don't really care. So it's about the harness, how you orchestrate them together. The skill that most companies have to focus on isn't the LLM; it's how you orchestrate the agents together. That would be the most valuable skill and the most valuable thing you need to do well in this industry. As a company, you're no longer writing software or buying software; you're orchestrating agents in different management structures and you create a factory. Every company is going to be making agents, doing factory productions of new workforces. The power and job opportunities for the individual is who can manage the most agents well. There's a better job opportunity for the person who can hire and manage 50 agents well as opposed to someone who has five agents or only ten. Productivity will accelerate because you can do more with that, and that becomes the superpower. We can assume that agents are going to grow and everyone's going to use them in companies and individually. But what happens on the back end, and this is where blockchain becomes really interesting, is that all the transactions I talked about, billions of dollars, are going to be negotiated from agent to agent because they will have wallets on the back. They will do the transactions. They are comfortable with blockchain and they need a permissionless negotiation layer because they don't need an interface like we do. All the interfaces we've seen, the web, the clickthroughs, are human interfaces. But an agent interface is entirely code-based. They don't need it. So they're going to transact. In the time that I'm having this conversation, my AI agents will have done 100 or 200 transactions. To them, it's completely normal. Blockchain is the native settlement layer for the AI economy. AI systems are breaking the internet in its current structure because it was designed with certain human limitations. So you need a programmable blockchain system that is comfortable for machine-to-machine communication. Programmable in the sense that an AI agent can also modify and change it in a way that a human cannot. We're not making rules when we send money, but an AI agent can do this very autonomously. Seven years ago, blockchain could only handle millions of dollars. Today it can do trillions of dollars no problem. BlackRock, Fidelity, Standard Chartered, the biggest companies in the world are all now using blockchain technology. For humans, crypto and blockchain remains terrifying. We haven't evolved from MetaMask over the last seven or eight years. So maybe it isn't about that after all. Maybe this technology is really meant for agents. Tokens become the virtual commodity for these agents. They are representations of compute and energy. They are the abstraction. In the same way that we talk about manufacturing and production of chips, lithium, silver, or gold, these are all commodities that help in production and there's a marketplace for it. That's what tokens become for compute and energy in the age of AI. Ultimately, traditional finance cannot bank AI. It's going to be a long time, if ever, that HSBC is going to give a bank account to an AI agent. But if an AI agent is to be truly autonomous, if it acts on your behalf, it must have access to money. A lot of people say it could use Visa or Mastercard, you could give it a credit card access, but will you pay 3.5%, 2.5%, even 1%? An agent will go for whatever is cheapest, fastest, and most reliable that you can prove. In our lens, a perfect financial system for machines is the blockchain. It was designed for that, and we will see trillions of dollars of new value grow onchain activity because of AI agents. All of them will have a wallet. They're going to have their identity on chain. Everything we built over the last seven years was probably to make agents happen. What's the impact? Maybe we do a little question here. How many agents do you think we will have in the future? Do you think we'll have more than 10 agents each? 20 agents? I have 200 agents. Depends where I talk. If I'm in Hong Kong, people say I'll have three to five agents in the next three years. If you're in San Francisco, they say I'll have a thousand agents. Either way, there's going to be more agents than humans. The number will be roughly 50 to 100, maybe more billions of agents in the world. I think we all agree that there's going to be many more agents than humans. They're all going to have a wallet. They're all going to be on chain. The reason they're on chain is because how do you create permissionless transactions in an open decentralized way that isn't gatekept? You don't have to keep your money in a bank account; onchain, I can send you a stablecoin or later do some kind of tokens. As I close, some thoughts on this onchain economy. People used to talk about the metaverse by taking virtual goggles and going into the virtual world. But maybe, given our experiences with AI agents, we had it slightly wrong. It wasn't about us going into the metaverse, but maybe the metaverse coming to us. When you think about what AI agents do today, they handle your spreadsheets, schedule your trips, make money for you, do all these transactions. But what is the impact? They're not impacting our virtual life directly, although it's virtual, your virtual being; they're impacting our physical life. They make impact in real life, whether they book tickets for you, handle stuff in Discord, manage your email. These are actual real life impacts. That's what agents are doing for us. The impact is really how they impact our real life, and the saying is that maybe it is about the metaverse coming to us as opposed to us going into that metaverse. What are some of the barriers? We believe there's going to be hundreds of billions of agents, but the barrier is still that it's quite difficult to use. Most people aren't comfortable using it. OpenClaw and these things are technical for the most part. You have to still deploy them. Different companies like ourselves are all progressing towards making it easier and simpler. We launched something called HelloMinds, which allows people to open their own AI agent over email in literally 60 seconds, and it communicates with you as an assistant, a coder, or a coworker. Another thing about AI agents to think about is what applications they can build. Sometimes you don't think about the applications because we still think of them in a human context. I want to give you one example that took me maybe 10 or 20 minutes. I told the agent to build it. I live in Hong Kong, and I had it compare shopping prices of commodities like milk, eggs, and chicken in different supermarkets. Here's what I discovered: there is a massive arbitrage between food items in Hong Kong, which I don't normally think about because how many people go into a supermarket and say, 'What's the price of milk here and what's the price of milk there?' It shouldn't make a difference. But it turns out that if I do this every day, I will save around 500 to 700 Hong Kong dollars a month. Maybe not a lot of money for a lot of people here, but it compounds. And it turns out they change it every day. As a human, it's very difficult to do this price comparison; it takes a lot of time. We wouldn't want to do that, and you're not going to go to three different supermarkets to buy your stuff. However, for an AI agent, it's just a few minutes of work. They do the analysis, comparison, and give you the detail. So it's not necessarily about an AI agent solving only the big problems, which of course we do all the time, but the real impact is that it solves these small problems at scale because they are not restricted by time like humans are. You don't have to go shopping into every single supermarket; you'll go to one that gives you a net average of 20 or 30% savings, and that's probably good enough. We think we've only seen the tip of the iceberg as to what you can do with AI agents. The entire discovery system will change as well. For instance, you're no longer going to go to an app to look for something; more likely the agent will find it for you. Everything is going to alter and change because of that. As a result, at Animoca, we've launched a $10 million investment program in the agentic space. If you're building with AI agents, we made our first investment just a few days ago, in Hyperliquid, a Polymarket-style trading agent that can work with any kind of agentic system. It's still an early stage; the whole agent space is only six months old because of the coding capabilities. For those of you building in the space, it's early. Anyone who's interested, feel free to apply. Otherwise, I hope you enjoy playing around with agents. Thank you very much.