Ask what "AI agents in crypto" means and you will get two answers depending on who you ask. One is a breathless list of autonomous funds, self-tweeting personas, and tokens that went up a thousand percent. The other is a much shorter, much quieter list of things people actually use every day. This piece is about the gap between those two lists — because the gap is the whole story.
We spend our days watching markets, headlines, and the public crypto conversation. Here is the honest map of where AI agents fit, sorted by how real each use is rather than how loud.
The seven things people point an AI agent at
1. Trading and DeFi. Bots that trade, hunt yield, or execute on-chain. The credible signal here is funding, not vibes: consumer trading apps in this space have raised serious rounds. Much of the "fully autonomous DeFi agent" volume, though, is a single press release wearing five different outfits — the same number quoted by five sites is still one number.
2. Agent frameworks and "agent tokens." The loudest 2025 story, and the clearest cautionary tale of 2026. A flagship framework-and-token project that once carried a multi-billion-dollar valuation collapsed to single-digit millions, with its own founder calling the token "dead." Tellingly, the open-source code kept getting committed to after the token was worthless. The lesson: the software layer and the speculative token layer are two different things, and only one of them was ever load-bearing.
3. Research and intelligence. The most durable corner — analytics platforms that turn on-chain data and social signal into something a human can act on. They sell to people, charge money, and keep the lights on without a token narrative. Boring, and real.
4. Prediction markets. This is the category with the best ratio of substance to hype, and the one closest to what we build. The underlying markets are large and growing, and the big venues now ship agent-friendly interfaces so bots can quote and trade programmatically. The eye-catching agent-performance stats floating around ("triple-digit returns," "a third of wallets are bots") almost all trace back to one vendor — so treat the market as real and the leaderboard as marketing.
5. Agent payment rails. The most mature infrastructure and the thinnest actual demand. Standards for letting agents pay each other in stablecoins are shipping, backed by names as big as the card networks — and one of the largest backers openly admits the organic demand "is just not there yet." Built for a future that hasn't arrived, by people betting it will.
6. Social and consumer agents. The autonomous-influencer wave. Weakest signal-to-hype ratio of the seven; the personas that went viral in 2024 have mostly thinned out.
7. Security. A genuinely new attack surface. When an agent can sign transactions, you no longer have to break the smart contract — you break the agent's reasoning. There is already a documented case of a prompt injection, hidden in a social-media post, talking an agent into moving funds. The key was never stolen. The agent was simply persuaded.
The pattern under the list
Read those seven back and a shape appears. The money is rotating away from "tokenize the agent as an asset" — the persona-token thesis that just deflated so publicly — and toward two humbler ideas: build the rails agents transact through, and let agents trade the venues that already exist. Utility over spectacle.
There is a second pattern, and it is less flattering to the space. A lot of what carries the word "AI" in retail crypto is not a product at all. "Claim your free AI trading bot — three-day trial, one dollar a day" is not a breakthrough in autonomous finance; it is a lure, and it is everywhere. When you watch the actual conversation in the crowd rather than the headlines above it, the "AI agent" narrative is far louder in the press than it is on the ground.
How to read any AI-agent claim without getting fooled
We apply the same discipline to this beat that we apply to our own numbers, and you can too:
- Syndication is not corroboration. The same figure on ten sites, all citing one report, is one source. Find the primary or discount the claim.
- Separate the code from the coin. A framework can be genuinely useful while its token is a casino chip. Judge them separately.
- Prefer businesses that charge humans money. Revenue is a harder thing to fake than a market cap.
- Assume an agent with signing power is a target. If untrusted text can reach a thing that can move money, that is the vulnerability, no matter how clean the contract is.
Where we sit
We build a prediction and event-trading exchange, which puts us squarely in category four — the real one. The interesting work for us is not launching an agent token; it is making our markets legible to agents that want to trade, turning market intelligence into something our users can act on, and protecting people from the "free AI bot" lures that use the same vocabulary we do. Build for the durable uses, treat anything that touches money as fail-closed until proven otherwise, and let the hype cycle be somebody else's product.
AI agents in crypto are real. Most of what is said about them is not. The difference is worth learning to see.