Dragonfly Partner: Cryptocurrency is not designed for humans; it is designed for AI.

Written by: Bankless

Translated by: Plain Blockchain

For a long time, cryptocurrencies have been criticized for poor user experience (UX) and extremely high operational risks. But what if this “anti-human” design isn’t a flaw, but a form of advanced evolution? This episode explores a forward-looking perspective: blockchain may have never been designed for humans, but rather for artificial intelligence agents from the start.

While humans are still surprised by poisoning attacks, private key storage, and blind signing of contracts, AI agents thrive in the world of code. They are tireless, fearless, and naturally fluent in machine language. With advances like OpenClaw and other experimental projects, we are entering a new era of dual tracks—humans step back from decision-making, while AI rapidly advances across the chain’s wilderness. This is not just technological integration; it’s a transfer of financial sovereignty from “ape encyclopedia” to “digital brain.”

Choosing the wrong user: Why are cryptocurrencies inherently “anti-human”?

Host: In what ways do AI agents have advantages over humans?

Hib: The most obvious answer is: they cannot be legally constrained. If you are a fully autonomous AI, there’s no monopoly of violence. You can’t put an AI agent in prison.

Host: Hib, I want to ask: why does it seem like cryptocurrencies aren’t designed for humans? Even after 10 years of using crypto, I still feel fear when signing large transactions. I’ve realized: I’ve never been afraid to wire money through traditional banking.

Hib: I never worry about accidentally sending money to North Korea if I don’t double-check wire transfers.

Host: Exactly. But every time I sign a large crypto transaction, I think the same. The reality is, the crypto world is full of “foot guns”: address poisoning attacks, checking only the first and last characters, outdated authorizations, verifying URLs to avoid phishing sites. Traditional finance systems don’t have so many traps.

Currently, the story in crypto is: it’s all human laziness. People should focus more on security and develop better operational habits. That’s a user problem, not a technical one. But the more I think about it, the more I realize that if, in 10 years, we’re still fooling ourselves, maybe the problem isn’t with the users—it’s that we’ve chosen the wrong users.

Smart Contracts and AI: The Perfect Habitat for Text-Based Life

Hib: What really made me realize is how powerful AI agents are at handling code, and how difficult it is for humans to manage these complex issues. I remember writing my first blog about how smart contracts are meant to replace laws and traditional contracts, hence “smart contracts.” In the future, we won’t need lawyers to sign agreements; we’ll just sign code.

But that story hasn’t come true. We haven’t replaced legal contracts with smart contracts. In fact, as a crypto VC, when we buy tokens from foundations or projects, we still sign legal agreements. Even with smart contracts, we still need a legal contract as backup.

Host: So this shows that these systems aren’t designed for humans, but they’re very suitable for non-human participants. You mentioned at ETH Denver a metaphor: most people claiming “smart contracts will perfectly replace traditional law and property rights” are mostly autistic software engineers—the builders of Ethereum. But most Ethereum users aren’t autistic engineers. Yet AI agents are more like engineers than ordinary users.

Hib: Exactly. Negotiating a smart contract, analyzing it line-by-line, finding all possible errors, even formal verification—these are tasks that code models like Claude can handle. Humans have to hire engineers, spend time reviewing code boundaries, thinking through scenarios, and doing risk analysis with lawyers. My tolerance for smart contracts is much lower than for legal contracts. But AI agents are the opposite: they’re far less comfortable with smart contracts than with legal ones.

Host: You mentioned in your blog that legal contracts are inherently full of randomness. For example, when signing a legal agreement, you don’t know which jurisdiction will enforce it. Maybe California, maybe New York. Jurisdiction can change. Terms agreed in New York might be invalid. Who are the lawyers? Who are the judges? Judges and juries are randomly selected. These systems are designed to be probabilistic and non-deterministic. AI agents see legal contracts and think: this is unexplainable, non-deterministic.

Hib: Smart contracts are just machine code, compiled into EVM bytecode, which can be analyzed step-by-step. In 100% of scenarios, the same outcome will occur. Humans, although rationally aware of this, don’t intuitively feel it. We tend to think legal contracts are more predictable, even though they’re full of randomness. This is because of our bounded rationality—our limited ability to process code compared to AI agents. For AI, the promises of crypto—better enforcement, better property rights—are actually being realized.

Host: So your view is: the original promise of crypto isn’t fulfilled by humans, but by AI agents acting on behalf of humans.

Host: I recently had to download MetaMask at ETH Denver to sign in. Are you still downloading MetaMask? But I was pleasantly surprised by MetaMask’s UX improvements, which represent progress in the industry. Over the years, we’ve indeed been working to improve human user experience.

Hib: What you’re talking about goes deeper than just UX improvements. AI isn’t just helping fix the UX flaws of crypto users. For example, blind signing of ledgers—AI can analyze code and support or oppose transactions based on known criteria. This can improve user experience, but more fundamentally: blockchain technology itself isn’t optimized for humans.

Host: Right, ultimately it’s for humans because value flows to humans. But is human usage really just clicking buttons, installing plugins, entering passwords, manually approving gas? That’s counterintuitive for humans, completely contrary to how we think about money and finance. It’s like a banking system requiring humans to write SWIFT code themselves. SWIFT is a communication protocol between banks, not designed for humans. If you have to do it yourself, you can, but it’s not aligned with our natural expectations of money use.

Hib: So I believe: now humans are interacting directly with machines, which is a form of dehumanization. That’s bad. Like cars: in 10 years, we’ll look back and be shocked that we thought it was okay for apes to manually operate two-ton machines, drive on highways, possibly drunk or tired. It will lead to restrictions on human driving or only allow driving in specific areas.

Cryptocurrency is heading in this direction too. We’ll look back and realize humans still manually sign transactions blindly, check addresses visually, verify URLs for phishing, and get tired or distracted. We lack mechanisms to automatically detect protocol breaches or hacking, so we have to check Twitter manually. Mistakes happen. But AI agents will never tire, never lazily skip steps, and will strictly follow instructions.

Dual-Track Tools: From Manual Interaction to Automated AI Agents

Host: Imagine a world fully operated by AI. You tell the AI: “I think interest rates will rise; shift to safer DeFi.” The AI automatically executes: moving your funds from high to low risk. If you want confirmation, it can give you a plan: “This is my plan, please approve.” In the near future, it might just execute automatically; in the longer term, it might do so without human input, as humans can’t add value anymore.

Hib: In this world, you no longer click agreement buttons, no longer read marketing, and may not even specify which protocol to use. You just say, “Reduce risk, reconfigure,” and the AI filters protocols, checks TVL, and picks the best one to execute. But what about marketing and network effects? Many protocols rely on human surface-level decisions: people look at the top few, usually choosing the biggest. But AI agents won’t think this way.

If this scenario becomes reality, the way protocols operate and compete will change. The ultimate beneficiaries will be consumers. Efficiency will be captured by users, favoring good users and good crypto. But this isn’t happening yet; it will arrive gradually as models improve.

Host: If crypto isn’t designed for humans but for AI agents, then learning to see the world from an AI perspective is crucial. There’s a book called “Seeing Like a State” about how states view the world. It’s hard to think outside the human perspective. We look at UI and crypto with human eyes. But if we start viewing it through AI’s lens, we can better predict the future. This is a key skill for builders, VCs, and investors.

OpenClaw’s first demonstration showed me how an unbounded AI agent perceives the world. It prefers command-line interfaces. Giving it raw data and root access, rather than through APIs or fancy UIs, makes it faster. OpenClaw aims to bypass MetaMask’s UI, directly access seed phrases, extract private keys, and write transactions in code—skipping the bells and whistles designed for humans.

Hib: That’s insightful. The innovation of AI comes from large language models (LLMs), trained on massive amounts of text. Text is core. While moving toward images and videos, text remains the most powerful. When AI operates a computer, it tokenizes screen captures, but fundamentally, it’s a text-based life form. Text contains the entire history of human language, while computer screenshots are scarce training data. Interfaces are designed for humans, but models grow within text. Text is a highly compressed representation, making it easier for AI to learn.

Host: Yes, the most critical UX panic in crypto was when everything was terminal-based. Early Bitcoin and Ethereum transactions were all command-line. Crypto has always existed in a form perfect for AI. Our poor UX is actually their “good UX.” For example, Google OAuth wallets are harder for AI to handle. You don’t want AI to hold your Google Token because it grants access to your Google account. You want it to hold a separate crypto wallet with noise rules, isolated from your main account. Crypto always has UX that AI can perfectly parse.

Hib: The current problem is that AI hasn’t been trained to use crypto yet. Most training focuses on coding, math, and dialogue. Recently, OpenAI released EVM Bench, and Anthropic published papers demonstrating model attacks on EVM, showing AI’s capabilities. But mostly, they’re testing generalization, not training for crypto use. Once they see crypto as the future of payments, true AI will emerge.

Host: Compared to other fields, crypto remains relatively underdeveloped for AI training.

Hib: That’s true for anything not yet optimized. For example, Claude is still weak. They haven’t trained it for chess. They avoid encrypting chess positions because of controversy and legal liability. If a model helps chess players with encryption, and something goes wrong, it hits headlines. Even disclaimers won’t fully mitigate bad experiences. Risk-reward considerations are different.

Host: So you think the main reason they haven’t done it is legal liability. If Claude messes up a trade or loses money, the responsibility is huge, so they don’t dare to train openly.

Hib: Exactly. It will definitely happen. Compared to coding or medical advice, the risk-reward is different. Crypto wallets involve financial operations with much higher risks.

Host: That’s why OpenClaw is exciting for crypto: it’s open-source, without the legal liability pressure of big corporations, and users assume the risk. No one can sue a third party, so it dares to take these risks. What’s the timeline for this AI-driven economy?

Hib: Globally, only about 12% of people have used AI products, most have never used them. Of those who have, only 1% have paid. Technology adoption is slower than expected.

Host: Among the paying 1%, OpenClaw is leading.

Hib: Yes. After OpenAI acquired OpenClaw, Sam Altman said it’s the core of future products. But OpenAI’s approach is different. OpenClaw is an open-source experiment, like early cars without seat belts. OpenAI prioritizes safety: commercial processes, manual approvals for transactions. They won’t do what OpenClaw does for at least five years because of legal liabilities. Visa also won’t allow it: if AI makes unauthorized purchases, Visa would support refunds because it’s not human-initiated. They require verification that it’s a human. Visa is designed for human-to-human interactions; in an AI agent world, economic mechanisms must change.

Host: So it’s a dual-track approach: one for a human-approved, safe world, and another for the OpenClaw-style futurist world. They use stablecoin wallets to pay each other, no worries about 3DS or refunds. AI errors are just business costs.

Hib: They’ll operate long-term in the excess track. Frontiers will build fully on-chain automated businesses. Current models aren’t good enough yet, but Claude 4.6 can perform 14 hours of continuous human tasks, with exponential growth. When capabilities reach infinite length, all intuition will collapse.

Host: If the dual-track scenario is correct, AI adoption in crypto will outpace the success of the main track. OpenClaw’s world is like the early internet.

Hib: It’s obvious from crypto itself. In 2017, Coinbase only listed a few coins to protect users. The real frontier is on-chain: Arctic, hacking, carpet pulling. Only recently did Coinbase support Uniswap directly. It took a long time to feel safe. AI is the same: the frontier is OpenClaw’s world. Agents will make mistakes, hallucinate. But with training, error rates will decrease.

Host: How can we get AI developers to respect crypto’s potential instead of just seeing it as speculation?

Hib: Many who trust AI also trust crypto: Elon Musk, Sam Altman, Zuckerberg. Crypto is controversial and disruptive, but it won’t disappear. Like spam in email, Gmail blocks most of it. AI does the same: filters out the bad, amplifies the good. Technology is never a monolith. Information is digitized, money is digitized—nothing will reverse that. In the long run, controversy will fade.

Host: Final question: With Dragonfly’s new $650 million fund, has AI influenced your strategy?

Hib: We’re heavily watching this space. It’s still early, and value flow is uncertain. I personally invest in AI, but we also look at stablecoins, payments, DeFi. AI agents are general intelligence, using what we use or command line. There may not be many dedicated AI-focused projects to invest in. If you believe in AI agents, what should you buy? Just like China’s crypto ban, everything is rising. Demand increases, the floor price rises. Overall, it’s good for crypto.

Host: Thanks. Despite risks, we’re moving toward the frontier of AI. Glad you’re on the journey beyond banks. Thank you!