AI agents will kill the web as we know it: Animoca’s Yat Siu

Animoca founder Yat Siu says the rise of AI agents will bring about a radical shift in how information and value will be distributed online.

Instead of the web being a place where humans interact with web pages and apps, the primary actors on the internet will be intelligent agents searching, negotiating, transacting and collaborating. This will cause a seismic shift in how the web is built and monetized. 

Instead of push-based emails and attention-grabbing ads, the model moves to pull-based, where agents find what users want or carry out their intents.

Siu tells Magazine he believes everyone will run dozens of AI agents in the future, specializing in different tasks: one to manage your portfolio, another for research, along with other agents for scheduling, coding, API management and so on.

That means there could be up to 200 billion agents roaming the web in short order — and they’re likely to be the main users of all the crypto infrastructure most humans find too complex.

Siu says they’ll use smart contracts to reach agreements where code really is law, blockchain verification for identity, and crypto to transact.

“If you’re an agent, it’s entirely conceivable that you’ll be making payment transactions 1,000 times, if not 10,000 times a day. What’s the most efficient, lowest cost way of doing it? It’s obviously blockchain,” he says.

Ad-supported open web is over

The impact of AI technology means the traditional model of an open web, with freely available information that’s supported by advertising, is dead. LLMs already consume any information posted online and regurgitate it for users, without sending any traffic to the source or funding the salaries of the journalists and researchers who came up with it. 

“We think traditional advertising, as you might know it in terms of Google ads and so on, is under threat,” he says. “We’re no longer going to be browsing the web the way that we used to. So essentially, HTTP as we know it is dead.”

Traffic volume to major sites fell sharply after Google introduced its AI summaries, with the worst-affected sites featuring tech reviews. Users are still basing their purchasing decisions on the information provided by the reviews, but they get it neatly packaged in an AI summary. Clearly a new monetization model is required, otherwise the websites providing content will go out of business.  

Microtransactions are finally practical

Siu believes agents will pay a tiny fee to access each article or research paper they need to provide users with the information they want. Microtransactions have been discussed since at least the late 1990s, but high credit card fees and a time consuming check out process meant it wasn’t practical. Now, AI agents can take care of the payments using low-fee cryptocurrencies.

One of Siu’s agents browses the web looking for interesting things to read.

“Right now, I have an agent that is basically creating my custom news for me based on information it receives, but from all sorts of sources,” he explains.

This information comes from free sources, but that’s set to change. 

“In order to get this data, it’s going to have to pay people for it. So, for instance for Cointelegraph, in the future, you’re going to be charging agents, maybe something in the form of tokens or stables, to access an API that gives you the news feed.

“That is then delivered to the user in some kind of packaged format, rather than people going to the Cointelegraph website and clicking through.”

Agents are already paying for content

There are signs this is already starting to happen. Last week, Distro Media unveiled a pilot project for an “agentic newstand” that allows AI agents to use microtransactions to pay for news articles. A demonstration article sold for 5 cents, with a transaction fee of a tenth of a cent. 

Last year, Cloudflare also released Pay per Crawl, which allows websites to charge AI company web crawlers a fee each time they fetch a page. At present, this is a one-off charge for companies like OpenAI to ingest it whole into their models, but they hope to set up an “agentic paywall” where individual agents can pay for access to specific content.

Siu says it’s likely this will also end up being bundled as packages of related content in a Netflix or Spotify style model. Apple News already does this to a certain extent.

Animoca Minds is OpenClaw for Dummies

Two months ago, an AI harness called OpenClaw had 15 minutes of fame amid breathless reports of its autonomous capabilities. “It’s amazing technology,” says Siu. The hype died down somewhat after ordinary folk realized you needed to have a computer science degree and three days to set one up, otherwise it would be full of gaping security holes. 

“The reality is that this software is pre-alpha,” says Siu, adding that Peter Steinberger (who he met recently and apparently isn’t as anti-crypto as he’s made out to be) is “not out there promising that it’s all safe and secure and everything and easy to set up.”

Siu is, however, promising his company’s customizable AI agent service called Animoca Minds is at least easy to set up. It’s relatively safe too, although it’s still in alpha, so you wouldn’t risk your family fortune on it just yet. Animoca portfolio company Cryptoslam has been developing the technology over the past nine months, and it was rush-released to capitalize on OpenClaw’s success.

Aimed at non-technical users, the set up process is done via email. You describe the broad outlines of what you want and then customize agents with different personalities to perform different tasks. Agents can manage wallets, track portfolios and manage data on block explorers.

You can create assistants to manage your scheduling, researchers that will operate to a brief, a crypto analyst plugged into Dune or Nansen or even an NFT trader on OpenSea with its own wallet.

I spent about half an hour creating an AI agent to analyze my AI Eye columns, determine the sort of weird, viral content it features, and after a bit of back and forth to get the format right, the AIEyeBot now emails me once a day with stories that fit the brief.

It’s also really easy to link it to a Telegram bot, so you can chat with your agents all day long if you like. It’s really weird the first time an agent pings you out of the blue on Telegram.

Animoca is far from the only company creating easy-to-use agents of course, and you can expect to see them everywhere in the coming months. This week’s Claude Code leak includes a forthcoming always-on background agent called KAIROS that checks every few seconds if it should fix some code or do other autonomous tasks while the user is sleeping.

It seems highly likely that most people will end up using agents created by OpenAI, Anthropic or Facebook, but Siu isn’t sure we’ll end up with the same five companies dominating AI as we saw with the web.

Siu believes the cost of inference is going to be 100 to 1000 times cheaper than today, and the LLMs themselves will just become background infrastructure. Open source models will be able to do 98% of the stuff everyday people need, he says. 

Siu says the real moat for an agentic system is in its persistent memory — it remembers everything you do and say, where you’ve been and what you’ve done. 

And if that memory is stored in a sovereign blockchain-based stack rather than on a centralized company’s servers, then their accumulated experiences and data belong to the user. Animoca Minds gives you the option of accessing this via a private key — or lazy people like me can just use a backup email for 2FA. Eventually, each mind could be represented by an NFT or a blockchain-based passport.

Yat Siu’s big call on the Metaverse

Siu was famous for being one of the biggest proponents of the Metaverse, and Animoca was one of the leading companies in the space even before Facebook changed its name to Meta and wasted tens of billions on its VR tech no one uses. Magazine first caught up with him in June 2020.

Siu believes much of the hype around the Metaverse was due to its emergence in the midst of the pandemic, when everyone was stuck at home and it seemed that meeting people in virtual worlds would become a permanent change.

“Everyone thought, ‘Oh, we’re going to be at home, and we’re never going to travel as much anymore,” he says. “And we have these new work habits,’ type of thing. Which, of course, turned out to be quite the opposite. Once COVID was dealt with, it was ‘travel revenge time.’”

But Siu has come to see the Metaverse as a sort of proof of concept for the infrastructure that enables AI agents to work together. 

“The thesis we originally had was, and we still think it’s the right thesis, was around digital property rights,” he says.

“Digital property is important because that’s how you have provenance. That’s how you have ownership. That’s how you know that you have assets in the virtual space in which you’re spending so much time. We think that thesis is still true, but who is it much more valuable for? It’s going to be for AI agents.”

“Where we’re landing is that the metaverse, the blockchain-based one […] was really the POC for agents. In other words, it was never really destined for humans as a prime consumer.”