Organized by: ChainCatcher
On October 7th, at the TOKEN2049 conference in Singapore, Dragonfly Managing Partner Haseeb Qureshi delivered a keynote speech titled "Agents + Crypto: What's Real & What's Cope". The following is the full text of the speech, organized by AI.
I. Confession: I Was Wrong
Let me start with a self-critique. I've been talking about agents and crypto for a while, and I was wrong. I asserted that: agents will all use crypto, traditional financial rails absolutely couldn't be ready before agents took off. That wasn't the case. The reality is: your agent doesn't need crypto, *you* need crypto.
Recall the market narrative from earlier this year: everyone was saying agentic payments would explode, the curve pointing straight up and to the right. "It's so obvious—the real users of crypto won't be normal humans, they'll be AI."
II. Reality: Agents Are Using Credit Cards
That chart only went until last November. What happened after that? It turned downward. It turns out agents aren't using x402 (Coinbase's agent payment protocol) en masse—90% of the payment volume on x402 is for farming airdrops, which is very 'crypto'.
And agents are already here: various consumer-grade agent products are growing like crazy, and OpenAI is working on them too. What are they using? Cards, credit cards, just like all of us. Companies like Rain, Visa, Ramp, Stripe have all launched agentic card products—generating sub-cards under your main card, with spending caps and merchant restrictions. All these companies saw the same opportunity in agentic commerce we did, and they were ready.
I use an agent myself. I gave it a crypto wallet, but there's limited stuff it can buy with that today; I also gave it a card—and most things it buys for me use the card. For consumer payments today, cards are basically the answer. Cards aren't perfect: ubiquitous in developed markets like the US, but not everywhere, and agents will be everywhere. So cards aren't the only answer, but the story from early this year—"agents will use crypto, traditional rails won't be ready"—was wrong. The rails are ready.
This is the "cope" part: aside from conventional uses like international payments, large transfers, the overlap between agents and crypto isn't as big as the early-year narrative suggested—what agents will use crypto for is similar to what you use crypto for today. But agents will fundamentally change other aspects of crypto, and that's the key point.
III. Agents Make Hard Things Easy: Governments Are Getting Overwhelmed
The most important insight about agents is this: agents make hard things easy. Governments worldwide are feeling this, and they are panicking—because government services have long relied on a vast bureaucratic maze, betting that ordinary people can't easily navigate it: filing court lawsuits, submitting complaints to the Consumer Financial Protection Bureau (CFPB), applying for benefits, filing housing complaints, all blocked by layers of barriers.
Now, almost every such service worldwide is seeing an explosion in applications—governments are getting "overwhelmed" by AI agents faithfully representing their owners to consume public services. The problem is: governments were never designed for "everyone can easily submit complaints." These systems are actually "secret rationing"—implicitly limiting access by making it "too hard, too annoying," and this mechanism fails in the age of agents.
IV. Agents Make Crypto Reliable: Protecting You From Your Own Stupidity
"Making hard things easy" also applies to another area: crypto itself is too hard—can't use wallets, can't handle gas fees, don't understand cross-chain bridges, so ordinary people are afraid to use it. And ordinary people are not just scared, they constantly make mistakes: hacked, phished, falling into honeypots. Just looking at on-chain records, a significant percentage of assets are permanently lost due to such errors.
This means one of the biggest ways agents will change crypto is by making crypto more reliable—protecting you from your own stupidity.
Why do we trust banks? Honestly, after over a decade in crypto, I still get nervous sending a large sum from a wallet—what if I'm accidentally sending money to North Korea? I never have that worry transferring money within a bank account, because the bank protects me from my own stupidity: if I really tried to send money to North Korea, the bank would stop me. This works because there's an intelligent layer between me and my money—the bank, acting as my intelligent agent to manage mistakes.
Once AI becomes the API to crypto, the same thing will happen in the crypto world: you will never get phished, hacked, or scammed again; if you encounter DNS poisoning, the agent won't let you enter a fake interface; if a contract is under attack, the agent will check it for you first; you think you copied an address, but it belongs to North Korea—the agent will always check for you. The agent will never be lazy, never skip a step, never get tired, never make the mistakes you make. The way we use crypto in five years won't be the same as today, and the trust relationship people have with crypto will be reshaped.
V. Privacy Was Never a Right, It Was Friction
My view might be harsh: privacy was never really a right, and we never truly had effective privacy—it was actually a rationing system. It's not that the government forbids you from asserting your rights, but reality—"it's too much trouble"—filters everyone out.
A U.S. Supreme Court Justice put it plainly in a GPS tracking case: in the pre-computer era, the greatest protection for privacy was neither constitutional nor statutory, but practical—it wasn't the law that gave you privacy, it was the practical cost of "police don't have the manpower to follow every suspect 24/7" that protected you.
Technology changed the equilibrium. Anthropic founder Dario Amodei pointed out the same thing in his exchanges with the U.S. Department of Defense: powerful AI makes it possible to "automatically and at scale piece together a complete picture of anyone's life"—data about each of us has already leaked in various breaches, a toxic sludge of information about everyone floating on the dark web. The only thing preventing someone from piecing together "who you are, what you're doing" was cost.
The recent controversy around the U.S. camera company Flock is a perfect example: street corner cameras themselves have long existed and aren't illegal; the difference is scale—when every street corner has a camera and AI can instantly search all footage, quantity itself becomes a qualitative change, and the government suddenly can "always know where everyone is." The only change was technology.
The crypto world is seeing the same thing: physical attacks against crypto holders have risen sharply in 2025–2026, with reported physical attacks hitting historical highs. People with publicly viewable holdings used to comfort themselves with "who would bother? I'm just a small fish." Before, bad actors had to do a lot of homework to target you; today, you don't need that motivation.
VI. Under the Panopticon: AI Empowers the State, Crypto Returns Power to the Individual
This "Panopticon" is precisely a large part of the reason crypto exists. In this dimension, the two technological directions are opposite: AI fundamentally empowers the state—it removes the traditional constraints of the state being "too slow, too stupid, can't understand the data"; whereas crypto does the opposite, pushing power from the government to the individual. In a world where governments are becoming increasingly powerful and all-seeing, crypto is your only hedge.
But the historical problem with privacy is: people say they care about privacy, but their actions betray them—handing over chat logs to Meta, ChatGPT, Google. Privacy only wins when it's "easy," never because it's "important." History repeatedly proves this: encrypted connections only became the default on the internet after Let's Encrypt made HTTPS certificates almost free—SSL used to be slow and expensive, only used on checkout pages; chat tools only saw billions truly using encryption when Signal became usable and WhatsApp fully implemented end-to-end encryption. The same thing is happening recently with Zcash: UX improvements directly explain the growth in shielded address usage.
My core judgment: AI will cause the same shift for privacy. Your AI agent will become your guardian against privacy invasions: when your agent is powerful enough, you just need to say "make me invisible"—erase my public footprint, give me top-tier operational security, delete all information about me from the dark web, simulate for me "if you were a bad actor, how would you find and attack me." Privacy protocols like Zcash are indeed hard to use today, but when the agent becomes your interface to crypto, all this costs you nothing—"don't use Bitcoin for this, use Zcash; don't use plaintext, use encrypted text."
VII. The Final Piece: Private AI
The final piece of the complete picture is private AI. Take the privacy-preserving AI platform Venice (a Dragonfly portfolio company) as an example: it's like a "private ChatGPT," running on encrypted rails. You can even use it completely anonymously—register with just an Ethereum address, pay with stablecoins.
I believe this will be the most important battle in civil liberties in the coming decade: first, how to protect yourself from the ever-expanding power of the state; second, how to maintain control over AI—AI will become the most politicized technology of the next decade.
VIII. Conclusion
Those stories about "agents will use crypto for everything they do"—I don't think they hold up. For most consumers, agents will use the same things you do. But my ultimate thesis is this: it's not that your agent needs crypto, it's that *you* need crypto.





