Here They Come: AI Agents

A few weeks ago, I mentioned a16z’s “Big Ideas 2026” report, which paints a clear picture of a new financial architecture combining AI decision layers, blockchain execution layers, and tokenization.

Last week, I got a chance to experience a small-scale version of that vision: I set up skills on Claude. Two different assistants to help with news research and draft writing for my newsletter. I ended that piece saying, “I want to set up something like this for trading too,” but added a caveat: if I don’t define the strategy clearly, the agent can’t make the right decision for me.

This week, we’re at the intersection of those two articles.

Beyond Enthusiasm: The Building Process

Our first two stories come from the blockchain/Web3 world: On March 3rd, OKX made an AI-focused update to its OnchainOS platform. Over 60 blockchains, more than 500 decentralized exchanges, and 1.2 billion API calls daily. The goal? To create an infrastructure where AI agents can operate autonomously.

On the same day, NEAR co-founder Illia Polosukhin said, “Blockchain users won’t be human anymore. AI agents will handle payments, manage assets, participate in governance votes. Humans will just talk to AI agents.”

So, a future where blockchain forms the infrastructure and AI takes the front end is being built today.

The First Crack: “February Wick”

Another story involves a brief but intense market movement in February 2026: 15,000 AI agents were using the same open-source model. They received the same signal simultaneously, trying to exit the same liquidity pool at once. This “collective” movement cost $400 million.

Classic financial theory relies on “diversity”: different actors think differently, balancing the market and prices. But if everyone, or most, thinks the same way, that assumption collapses.

Economic systems were designed for human use; when users become machines and all rules are set from one point, we might face situations the current economic system hasn’t encountered before.

“February Wick” might not be an accident, but rather a kind of warning.

Today’s Targets Are Humans, But What About Tomorrow?

This week, Moonlock Lab highlighted another issue. Scams targeting professionals in the web3/blockchain/crypto ecosystem, which had been ongoing on X, have now spread to LinkedIn. Fake VC profiles, fake meeting invites, fake Zoom or Google Meet links. Persistent suggestions like “You must download and install the code from our GitHub,” or “We use our infrastructure for meetings, please download and install our app.” You click, download the file, and malware is instantly installed on your device. This is more about social engineering than a technical vulnerability.

At this point, I can’t help but ask a question I’ve posed before: What if the victim wasn’t a human, but an agent?

Would agents also receive “invitations”? Would they click on links? If rules, risks, and dangers are well-defined beforehand, agents could respond more clearly, decisively, and safely than humans in such situations. But on the flip side, they might follow instructions from a seemingly trustworthy source. Just like how antivirus programs can’t catch zero-day attacks and there are always a few or more victims, or how users or servers within the same secure zone can’t be protected from attacks from each other.

The Common Denominator, and That Question

OKX is building infrastructure on the blockchain for AI agents, NEAR is describing what AI agents will do in our near future, February Wick is testing the classic and established economic system, and cyber attackers are seeking new targets in new environments. They all point in the same direction: AI agents are on their way to becoming significant economic actors.

In an ideal scenario, this idea is truly exciting. A faster, more efficient system, free from emotional decisions. In my last article, I said, “I want exactly such an assistant for trading”, and I still do.

But what if we’re wrong and not everything goes perfectly? 15,000 agents making the same decision at the same time, or an agent manipulated by a fake invitation acting on my behalf; to me, these all sound like scenes from a horror movie.

“Know Your Agent” is no longer just a philosophical concept. If an agent’s identity, authority, rules, and limits aren’t clearly defined before deployment, emotional decisions will be replaced by algorithmic errors.

And who will bear the cost of those errors?

Sources

OKX OnchainOS: https://www.coindesk.com/tech/2026/03/03/okx-jumps-into-ai-agent-race-with-new-onchainos-toolkit

NEAR co-founder: https://www.coindesk.com/tech/2026/03/03/ai-agents-will-be-primary-users-of-blockchain-near-co-founder-says

February Wick / DeFYZ: https://coincub.com/blog/best-ai-crypto-agents/

Fake LinkedIn VC attack: https://cyberpress.org/fake-linkedin-vcs-spread-malware/

a16z Big Ideas 2026: https://a16zcrypto.com/posts/article/trends-stablecoins-rwa-tokenization-payments-finance/