At TOKEN2049 in Singapore on Oct. 7, Dragonfly managing partner Haseeb Qureshi used his talk, titled Agents + Crypto: What's Real & What's Cope, to publicly revise one of his own earlier calls.
His updated view was simple: it is not that agents need crypto. People do.
He opened with a correction to his earlier thesis
Qureshi said he had been talking about agents and crypto for some time and had previously argued that agents would end up using crypto, while traditional financial rails would not be ready before agents arrived at scale. He said that turned out to be wrong.
He recalled the market narrative from earlier this year, when many in crypto were pushing the idea that agentic payments would take off quickly and that AI, not ordinary people, would become crypto’s real end user base. In his telling, that narrative did not hold.
He said the curve everyone was drawing only went up until last November. After that, it reversed. Agents did not adopt x402, Coinbase’s agent payments protocol, at anything close to the scale some expected. Qureshi said 90% of payment volume on x402 has been driven by airdrop farming.
Consumer agents are using cards
For Qureshi, the important point is that agents are already here. Consumer-facing agent products are growing fast, and OpenAI is building in the category as well. But the payment instrument they are using is not primarily crypto. It is cards.
He named Rain, Visa, Ramp and Stripe as companies that have already launched agentic card products. The structure is straightforward: a sub-card can be created under a user’s main card, with spending limits and merchant or use-case restrictions.
Qureshi said these companies, just like crypto builders, have understood that agentic commerce will matter. The difference is that they are already operational.
He also used his own setup as an example. He said he gave his agent a crypto wallet, but what the agent can buy that way remains limited today. He also gave it a card. Most of the things it actually bought were paid for with the card.
That does not mean cards solve every payment problem everywhere. Qureshi noted that cards are ubiquitous in developed markets such as the US, but not in every part of the world, while agents will be everywhere. So cards are not the only answer. Still, he said the early-year story that agents would use crypto because legacy rails would not be ready has already been disproved. The rails are ready.
Where the “cope” begins
That, in Qureshi’s framing, is the “cope” part of the story. Outside familiar crypto use cases such as international payments and large transfers, the overlap between agents and crypto is smaller than many suggested at the start of the year. Agents are likely to use crypto for many of the same things people already use crypto for today, not for a wholly new mass-market payment behavior.
He did not present that as a dismissal of crypto. Instead, he shifted the discussion to what he sees as the more consequential effect of agents: they change how people interact with crypto, how errors get handled and how privacy may be defended.
Agents make hard things easy, and governments are already feeling it
Qureshi said the single most important thing to understand about agents is that they make hard things easy.
He argued that governments around the world are already reacting to that shift with alarm because public services have long depended on bureaucratic mazes that most people struggle to navigate. He cited filing court claims, submitting complaints to the Consumer Financial Protection Bureau, applying for benefits and lodging housing complaints as examples of systems built with friction at every step.
Now, he said, application volumes across many of these services are surging. In his words, governments are being flooded by AI agents faithfully representing their users and consuming public services at scale.
His point was not just about automation. It was about system design. Governments, he argued, were never built on the assumption that everyone would be able to file complaints easily. Many of these systems functioned as what he called a hidden rationing mechanism, one that relied on difficulty and annoyance as an implicit throttle. Agents break that throttle.
He expects agents to make crypto safer to use
Qureshi then applied the same logic to crypto itself. Crypto remains difficult for ordinary users: wallets are confusing, gas fees are not intuitive, bridges are hard to assess, and the risk of error is high. People do not just feel intimidated. They routinely make mistakes.
He pointed to hacks, phishing attacks and honeypots, and said onchain records show that a meaningful share of assets has been lost permanently because of this kind of user error.
That is why he believes one of the biggest ways agents will change crypto is by making it more reliable. His phrase for that was memorable: agents can protect you from your own stupidity.
He compared that function to banking. Even after more than a decade in crypto, he said, sending a large transfer from a wallet can still make him nervous. He used an extreme example to make the point: what if he accidentally sends money to North Korea? In the banking system, he said, he does not worry the same way because the bank acts as an intelligence layer between him and the money. It catches obvious mistakes and stops certain risky transfers.
He expects a similar intelligence layer to emerge in crypto once AI becomes the API through which users reach onchain systems. In that model, the agent checks first. It catches phishing attempts, flags compromised interfaces caused by DNS poisoning, reviews whether a contract has been exploited, and verifies whether the address a user copied is actually the destination they think it is.
Qureshi’s argument was that agents do not get tired, skip steps or become careless. People do. He said that difference will reshape how crypto is used within five years and alter the trust relationship between users and the technology.
His privacy argument: privacy was often protected by cost, not law
From there, Qureshi moved into privacy. He said his view might sound harsh: privacy has never really functioned as a guaranteed right in practice. More often, it has survived as a byproduct of friction.
He argued that many people were not protected because the state had chosen to respect their privacy in every instance. They were protected because surveillance, aggregation and investigation took too much time and effort.
To illustrate that, he cited a US Supreme Court justice in a GPS tracking case. In the pre-computer era, the justice argued, the greatest protection for privacy was not constitutional or statutory. It was practical. Police simply did not have the capacity to follow every suspect around the clock.
Qureshi said technology has changed that equilibrium. He also referred to Anthropic founder Dario Amodei and his clash with the US Department of Defense, saying Amodei made a similar point: powerful AI makes it possible to automatically and at scale assemble a full picture of a person’s life. Data about nearly everyone, Qureshi said, is already scattered across leaks and floating through dark web markets. What stopped others from rebuilding a complete portrait was the cost of doing it.
He linked that shift to both public surveillance and crypto-targeted violence
Qureshi used the controversy around US camera company Flock as an example. Cameras on street corners are not new, he said, and are not inherently illegal. The change comes from scale. Once every corner has a camera and AI can search all of that footage instantly, quantity turns into something qualitatively different. The state can suddenly know where everyone is, all the time. The only thing that changed is the technology.
He said crypto is seeing a comparable shift. Physical attacks against crypto holders rose sharply in 2025 and 2026, and reported incidents reached an all-time high. For people with publicly visible holdings, the old comfort was that no one would spend the effort to target an ordinary individual. Qureshi said that comfort no longer holds when the cost of targeting falls.
AI strengthens the state, while crypto can strengthen the individual
That is where he located a major reason for crypto’s existence. On this axis, he said, AI and crypto push in opposite directions.
AI strengthens the state at a basic level because it removes long-standing constraints: governments are no longer as slow, as unintelligent or as unable to process data as they once were. Crypto, by contrast, shifts power away from governments and back toward individuals.
In a world where states become stronger and more capable of seeing everything, Qureshi argued, crypto may be the only real hedge available to the individual.
Privacy wins only when it becomes easy
He did not romanticize public demand for privacy. People say they care about privacy, he said, but they still hand over chats and personal data to Meta, ChatGPT and Google. Privacy has never won simply because it was important. It wins when it is easy.
He pointed to earlier transitions on the internet. Encrypted web traffic did not become standard until Let’s Encrypt made HTTPS certificates nearly free. Before that, SSL was slower, more expensive and typically limited to checkout pages. Messaging followed a similar path. Only when Signal became usable and WhatsApp rolled out end-to-end encryption broadly did encrypted messaging reach billions of people.
He said Zcash is now showing the same pattern. Improvements in user experience can directly explain growth in shielded address usage.
His central bet is that agents will turn privacy into a low-friction default
Qureshi’s main conclusion was that AI will drive the same transition in privacy. A user’s own AI agent could become a standing defense layer against privacy intrusion.
In his example, a person could simply tell the agent to make them hard to find. The agent would then clear public traces, set up stronger operational security, work to remove leaked information from the dark web and run an attacker-style analysis of how someone might locate or target them.
That logic also extends to crypto privacy tools. Protocols such as Zcash may still be difficult to use today, he said, but if the agent becomes the interface to crypto, the cost to the user can approach zero. The agent can decide when not to use Bitcoin, when to use Zcash instead and when plaintext should be replaced by encrypted text.
The final piece: private AI
Qureshi closed by saying the last piece of the full picture is private AI. He used Venice, a privacy-preserving AI platform and a Dragonfly portfolio company, as the example.
He described Venice as something like a private version of ChatGPT that runs on crypto rails underneath. Users can, in his account, access it anonymously by registering with an Ethereum address and paying in stablecoins.
He said this will be one of the most important civil-liberties battles of the next decade. The fight has at least two parts: how individuals protect themselves against the expansion of state power, and how they retain control over AI. In his words, AI will be the most politicized technology of the coming decade.
His closing line returned to the opening correction
Qureshi ended where he began. The story that agents will use crypto for everything they do does not hold, he said. For most consumers, agents will use many of the same tools people already use.
But his final point was not anti-crypto. It was the opposite. Agents may not need crypto in the way many expected. People still do.