Dragonfly Managing Partner Haseeb Qureshi said AI agents have not adopted crypto payments in the way much of the market expected at the start of the year. As agents moved from theory into real usage, he said, they did not wait for stablecoins or on-chain payment rails. They used cards first.
The remarks came in a long-form conversation with 168X host Mr. Z, published on Oct. 13, 2026. Haseeb discussed TOKEN2049 Singapore, AI agents, privacy, agentic finance, Robinhood Chain, Hyperliquid, HIP-3, equity perpetuals, and Dragonfly’s next phase.
At TOKEN2049, the clearest signal was not attendance but what did not happen at Marquee
Haseeb said TOKEN2049 has likely become the world’s largest retail crypto conference, with about 25,000 attendees this year. But for him, the more revealing measure of the industry’s current state was Marquee at Marina Bay Sands.
In past years during Token Week, he said, the venue was almost always booked out at night by crypto projects. This time, for the first time he had seen, Marquee operated as a normal nightclub for the entire week, with no crypto project taking it over.
That detail, in his view, says a lot about how the industry has changed. Retail activity, volatility, and hot projects are still there, but the intense speculative energy that defined prior cycles has faded. In contrast, one of the more important events at this year’s conference was a closed-door institutional gathering attended by hedge funds, asset managers, founders, and regulators who flew into Singapore for it.
He said conferences still produce useful information. Conversations on the ground still reveal where capital is looking and which projects are drawing attention. But the industry, in his words, is getting older, more mature, and more institutional. More people are showing up for business, deals, and flow, not only for parties.
AI agents did not wait for crypto. Card networks moved faster
Haseeb said he had publicly revised a view he shared earlier in the year. Back then, many people in crypto, including him, believed crypto would become essential infrastructure for AI. The reasoning seemed straightforward: an AI agent has no driver’s license, no Social Security Number, and no obvious way to complete KYC, open a bank account, or connect to traditional payment systems. Stablecoins looked like the natural answer.
What changed was reality. At the start of the year, agents were still mostly a thesis. Now enterprise and consumer adoption are both growing quickly, and the market can finally observe what agents actually use for payments. The answer, he said, is simple: mostly cards.
He named Visa, Mastercard, and companies built on card networks such as Rain, Ramp, and Stripe. According to Haseeb, they moved quickly to build the features agentic payments require, including spending limits, risk limits, and permissions, all on top of existing card rails.
Traditional finance adapted faster than many in crypto expected. Haseeb said that point should simply be acknowledged: the original call was wrong.
Crypto does not need to be the default payment layer for agents
That does not mean agents will never use crypto. Haseeb said they will use it for the same reason humans do: when it is the best tool for the job.
If a cross-border transfer is cheaper and faster with stablecoins, an agent will find that route. If an on-chain market offers something unavailable in traditional finance, it will use that too. The mistake, he said, was assuming that because an agent is software, it must naturally prefer crypto. It will not. An agent has no ideological loyalty to crypto. It optimizes.
Sometimes the answer will be a card. Sometimes an API. Sometimes a stablecoin. Sometimes an on-chain protocol.
He argued that this version of the thesis is healthier. Crypto does not need to win all payments. It only needs to win where its architecture is actually better. The harder question now is which financial actions are genuinely improved by blockchain systems compared with traditional rails.
Where AI matters more for crypto: interface, execution, and safety
For Haseeb, the bigger change AI brings to crypto is at the interface layer. Today, users still have to bridge assets, watch gas fees, search for yield, compare Aave with Morpho, rebalance positions, and hunt for the best trading fee. A lot of that work, he said, will disappear behind an agent. The user will only need to say what they want done.
He described that future agent as a personal “crypto guy.” It would search for better yield, lower fees, and better execution paths. It would also handle due diligence.
Each time it interacts with a smart contract, he said, it could inspect the contract, look into the team, review LinkedIn, search X, and assess whether there is honeypot risk, Ponzi risk, or rug risk. It would not skip checks out of laziness, and it would not trade on FOMO.
Haseeb added that today’s agents still make mistakes, but today is also the worst they will ever be. Every model release improves them. He compared the shift to self-driving cars: five years ago they may not have matched human safety, but the picture is very different now.
The more interesting conclusion, he said, is not that AI makes crypto better for machines. It may be that AI finally makes crypto safe enough and simple enough for ordinary people to use.
Privacy is the stronger AI-crypto overlap
Asked about Dragonfly’s recent investments in Venice AI and Sapiom, Haseeb said privacy is now the most important intersection between AI and crypto.
In his view, AI is dangerous for privacy by default. Each time users hand data to an AI system, they are usually handing it to a third party. People may feel that a conversation with an AI system is deeply private, almost like an extension of the mind, but that feeling does not necessarily hold up technically or legally.
He also said AI makes mass surveillance easier. Data that used to sit in separate places can now be aggregated, analyzed, and even de-anonymized quickly by models. As AI strengthens states and large institutions, he said, another set of technologies is needed to strengthen individuals. That is where privacy-preserving cryptography, Zcash, Monero, and private AI infrastructure matter.
Venice, he said, is a useful example. Users can choose a traditional login and a credit card, but if they want stronger privacy, they can use a crypto wallet, a crypto payment method, and encrypted inference. A credit card reveals identity to the platform. Crypto plus real end-to-end encryption can reduce what the platform knows, potentially down to not knowing a user’s name, location, or model choice.
That led him to a line he said differs sharply from his view earlier this year: crypto is not for your AI agent. It is for you. The agent only uses it on your behalf.
Agentic finance may first erase the “human laziness tax”
On financial stability, Haseeb said the first effect of agentic finance may not be an agent-driven bank run. It may be the destruction of business models built around user inertia.
He used a checking account at JPMorgan as an example. Large amounts of cash may sit there at a very low rate while money market products offer much higher yield. Why do people not move money every day to wherever yield is highest and move it back when they need to make a payment? Because people are lazy.
Agents are not. They can operate 24/7, check balances every hour, and move idle cash automatically. Haseeb said cheap deposits are central to the business model of many large banks because consumers do not bother moving money, allowing banks to lend against very cheap funding and earn the spread.
He made the same point about brokers, naming Schwab and Fidelity, where a meaningful part of the economics comes from client cash left in sweep accounts.
Agents, in his view, will arbitrage away this “human laziness tax.” The more likely result is not a sudden run on every bank but a forced rise in deposit rates as institutions try to keep funds from leaving. That would make funding less cheap and put pressure on margins and return on equity for banks and brokers.
Incumbents are already starting to block agents
Haseeb said incumbents will not sit back while agents strip out their profits. One response is to block them, and he said that process has already begun.
If a platform’s profits depend heavily on advertising, sponsored placement, or friction in the user journey, an agent that acts in the user’s interest is a direct threat. It does not care about sponsored results. It only looks for the best result for the user.
He expects banks to develop the same incentive structure. They may not say openly that they do not want a user’s agent moving deposits to higher-yield products. Instead, they may say the process is unsafe, presents compliance risk, or that AI can hallucinate. The conflict underneath is still economic.
Over the long run, he said he is especially positive on two categories. One is agent-first companies that are willing from day one to support agentic commerce. The other is local agents. If your agent effectively belongs to Meta or another third-party platform, that platform can be blocked. But if the AI runs on your own computer and acts through the browser like a human, it becomes much harder to stop.
For Haseeb, the larger power shift is from platforms deciding what you are allowed to do to your agent acting in your interest.
On 2026 forecasts: BTC at $150,000 looks difficult, fintech chains were overestimated
Looking back at his earlier 2026 predictions, Haseeb said the overall set held up reasonably well, but not all of it was right.
He had previously called for BTC to reach $150,000 by year-end. From current levels, he said, that would require nearly a doubling, making the target very difficult. Even so, he still sees a move back above $100,000 as possible.
On fintech chains, he said the broad direction of his view still looks right. Last year, the market thought Tempo, various fintech chains, and Robinhood Chain would use distribution to pull in large amounts of stablecoin and real-world asset activity, potentially challenging Ethereum and Solana.
So far, that scale has not appeared. Robinhood is still very early, so he did not want to draw a final conclusion. But most of the meaningful net-new RWA activity, he said, is still happening on Ethereum and Solana. Distribution matters, but distribution alone does not create a blockchain ecosystem.
Robinhood Chain still needs to prove RWA traction and a developer base
Haseeb said Robinhood Chain remains TBD. There is clear meme activity, he said, but Robinhood is not trying to build a meme chain. The main story it wants to tell is RWA, and right now the amount of real on-chain RWA activity is still small, including as a share of the broader RWA market.
The second issue is developers. Haseeb said he has not yet seen many truly independent developers continuously building new applications and new ideas on Robinhood. What he sees instead is more meme farming, meme cycles, and a set of market structure experiments.
He said perps and DeFi market structure are worth watching, but on absolute scale Robinhood Chain remains much smaller than Hyperliquid, Arbitrum, BSC, Tron, Base, and Solana.
That is not a call that Robinhood Chain cannot succeed. His point was narrower: the market jumped too quickly from “Robinhood has distribution” to “Robinhood Chain will swallow everything,” and that still has not been proven. The real things to track are sustained RWA growth, independent developers, and native applications that can form a flywheel without depending on the Robinhood brand itself.
Perp DEXs may consolidate, but not into a one-player market
On perpetual DEXs, Haseeb said the market has not moved as far from his original expectation as some might think. In one 24-hour snapshot, he said, Hyperliquid reported roughly $4 billion in volume, while Variational and Lighter were each around $1.5 billion.
That is not yet the 40 / 30 / 20 split he once imagined, but it does suggest the beginnings of a structure centered on a small number of large venues.
He disclosed that Dragonfly is an investor in Variational and Lighter and also holds some HYPE. Even with that disclosure, he said he still believes the market structure will continue to consolidate as the liquidity flywheel at these venues matures.
He also said one area where he feels more confident he got the call right is equity perpetuals. At the start of the year, he said, the market was close to zero. Now it is clearly taking shape. He sees one important growth curve for DeFi derivatives in bringing stocks and other traditional assets into the same perpetual infrastructure.
HIP-3 is less permissionless than it appears, and TradeXYZ looks more like a deep partnership
Haseeb said he is somewhat skeptical of the idea that HIP-3 on its own proves a naturally effective open marketplace model.
If the model were that powerful on its own, he said, the market should have already seen many successful examples before TradeXYZ emerged. That did not happen. TradeXYZ has been a major success, but he argued it was not a case of a random team walking in through the front door and succeeding automatically because HIP-3 existed.
Instead, he said, TradeXYZ had a deeper history with the Hyperliquid ecosystem, including prior cooperation in areas such as spot infrastructure. That makes it look more like a deep partnership, even a quasi-joint venture, than proof that a neutral and permissionless marketplace will naturally produce great companies by itself.
He also raised a practical question. If a team wants to build a very large market, why place the entire company on someone else’s land, pay significant rent over time, and accept the platform risk that the rules may change later? If you are building the next Polymarket, he said, you may prefer to own the infrastructure.
Hyperliquid can provide strong rails, but it will not do market development for you. Going from zero to one in a new category still requires founders, a full team, capital, and distribution. Protocol primitives can make execution easier. They cannot replace company building.
Hyperliquid is expanding the market more than taking share from centralized exchanges
One common narrative is that Hyperliquid is taking market share directly from Binance, OKX, Bybit, and Bitget. Haseeb said that reading is incomplete.
If you analyze addresses on Hyperliquid and compare them with CEX usage, he said, the overlap between the two user bases appears lower than many people assume.
Three or four years ago, there were still many non-KYC centralized exchanges globally, and their user bases may have looked more similar to Hyperliquid’s current audience. Today, large CEXs are highly regulated and generally require KYC. A set of users, for regional, regulatory, KYC, or other reasons, do not have strong access to those venues. For them, Hyperliquid is not a migration from Binance. It is access to a market they did not really have before.
He added that if Hyperliquid were truly taking share from CEXs, Hyperliquid volume should rise while CEX volume falls. Instead, he said, both sides are growing. For now, that looks more like market expansion than share transfer.
HYPE still has room, but crypto remains the core engine
On the next growth curve for Hyperliquid, Haseeb said pre-IPO equity clearly has demand and product-market fit, and TradeXYZ still has a lot of room to expand. But if the question is what the next biggest product curve will be beyond that, he said he does not yet have a high-conviction answer.
He also argued that some investors are too focused on how Hyperliquid can move beyond crypto and not focused enough on the fact that Hypercore already has a very strong crypto business. He said he is constructive on the crypto market over the next 12 months, and if that recovery continues, HYPE itself and exchanges built around crypto trading could still benefit from a favorable backdrop.
The best outcome, in his view, is not Hyperliquid leaving crypto behind. It is crypto compounding at the core while equity, pre-IPO assets, and other financial products expand total addressable market layer by layer.
Dragonfly no longer sees itself only as a bridge between East and West
Asked whether Dragonfly should still be understood mainly as a China-US bridge, Haseeb said the industry has changed too much for that frame to be sufficient.
When he first entered crypto, China was still relatively constructive toward the sector overall. There was an ICO crackdown in 2018, but the industry had not gone fully underground. By 2020, as the market accelerated, there was still a sense that crypto might become a globally accepted technology.
After 2021, China banned crypto and mining. Many founders who had been building in China moved to Singapore and Hong Kong, and later to Dubai and the US. In that context, he said, Dragonfly’s role is no longer just connecting China and the US. Crypto itself is global.
He said the Chinese diaspora remains an important part of the industry, only now it is distributed across countries. Dragonfly’s LP base includes Asian capital and US institutions. To invest well in crypto, he said, you need to see the whole elephant rather than touching only one part of it.
That is one reason Dragonfly has a presence in both Singapore and New York. The firm wants to identify founders creating real value in different parts of the world, instead of judging investability by where a company is registered. Haseeb added that he still believes China will eventually recognize that crypto has become too large and too important to ignore completely.
By 2030, “crypto VC” may stop being a category
Haseeb said his comment about crypto VC disappearing does not mean crypto investment firms will all vanish. It means the label itself may stop making sense.
Crypto, he argued, is moving into the plumbing. Stablecoins will become fintech backends. Neobanks will integrate crypto. Consumer products will build in wallets. More technology companies will quietly use blockchain rails in the background.
He compared that shift to the old category of “internet investor.” When the internet first emerged, the label had meaning. Later, when every company became an internet company, the category collapsed into investing. He expects crypto to follow the same path.
That transition, he said, will be harsh for funds whose only edge is knowing how to buy tokens. In 2020 and 2021, it was possible to raise a fund, buy a few tokens, flip some low-quality coins, and call yourself a crypto VC. That will become harder.
The durable edge, in his view, will come from understanding the world, having real relationships, helping founders solve problems, and offering something beyond crypto fluency. Funds that do not have those qualities will struggle to survive the transition. Those that do may face a much larger opportunity.
When asked who he most wanted to meet in Asia now, Haseeb’s answer was founders. He said Dragonfly has always believed that great ideas can emerge anywhere in the world.