Written by: Bankless

Compiled by: Baihua Blockchain

For a long time, cryptocurrency has been criticized for its poor user experience (UX) and high operational risks. But what if this "inhuman" design is not a flaw, but a form of advanced evolution? This issue's dialogue research explores a forward-looking perspective: blockchain may not have been designed for humans from day one, but rather prepared for artificial intelligence agents.

While humans are still amazed by poison attacks, private key management, and blind signing of contracts, AI agents thrive in the world of code. They do not get tired, do not fear, and are naturally proficient in machine language. With the advancement of cutting-edge experiments like OpenClaw, we are entering a new era of dual-track cushions—humans step back from decision-making while AI rapidly moves in the on-chain wilderness. This is not just a fusion of technology, but a transfer of financial sovereignty from the "primate encyclopedia" to the "digital brain."

Choosing the Wrong User: Why is Cryptocurrency Inherently "Inhuman"?

Host: In what ways does AI hold a comparative advantage over humans?

Hib: The most obvious answer is: you cannot prosecute an artificial intelligence agent. If you are a fully autonomous agent, there wouldn't be a violent monopoly. You cannot imprison an AI agent.

Host: Hib, I want to ask a question: why does cryptocurrency seem not to be designed for humans? Even as a crypto user for 10 years, I still feel fear every time I sign a large transaction. I am contemplating the fact that I have never feared making a wire transfer.

Hib: I would never worry that if I don’t check my wire transfer multiple times, I might accidentally send money to North Korea.

Host: Right. But I think this every time I sign a large crypto transaction. The reality is that the crypto world is filled with "poisoned addresses": when reading an address, you have to consider whether it's a poisoning attack; you should check a few middle characters, only looking at the start and end; whether there have been outdated approvals; you have to check the URL to ensure it’s not a slightly altered phishing site. There aren’t that many traps in traditional financial systems.

Currently, the story in the crypto circle is: it’s all the fault of humans being too lazy. Humans should pay more attention to security and have better operational habits. This is a problem of the users themselves, not a fault of the technology. But the more I think about it, if it is still self-deceiving in 10 years, perhaps the problem is not with the users but with choosing the wrong users.

Smart Contracts and AI: The Perfect Habitat for Textual Beings

Hib: What really made me start to realize is how strong my AI agent's ability to handle code is, and how difficult it is for humans to deal with problematic constructs. I remember my first blog when I entered the industry talked about: smart contracts replacing laws, replacing traditional contracts, hence "smart contracts". In the future, there will be no need to find a lawyer to sign agreements, just use code to sign agreements.

But the reality is that this story hasn’t happened. We haven’t replaced legal contracts with smart contracts. In fact, as a crypto VC, when we buy tokens from a foundation or project party, we still sign legal contracts. Even if there are smart contracts, we still need to sign a legal contract just in case.

Host: So this indicates that this thing is not designed for humans, but it is very suitable for non-human participants. You mentioned a metaphor at ETH Denver: the ones who first said "smart contracts perfectly replace traditional law and property rights" are mostly autistic software engineers—they are the ones who built Ethereum. However, most Ethereum users are not autistic software engineers. Yet AI agents are still those engineers.

Hib: So we find that negotiating a smart contract, analyzing it line by line, looking for all possible error spots, and even formally verifying before deciding whether to agree—this is something that models like Claude can complete. However, humans have to hire software engineers, spend time viewing code boundaries, contemplating situations, and conducting risk analysis with lawyers. My tolerance for smart contracts is far lower than for legal contracts. But AI agents, on the contrary, find smart contracts much more comfortable than legal contracts.

Host: In your blog, you mentioned that legal contracts are actually filled with randomness. For instance, when signing a legal contract, you don’t know in which jurisdiction it will ultimately be enforced. It could be California, it could be New York, and jurisdiction might change. Things agreed upon in New York may have clauses deemed invalid. Who are the lawyers? Who are the judges? Judges and juries are randomly selected. These aspects are designed to be random and non-deterministic. AI agents see legal contracts and think: this is inexplicable, non-deterministic.

Hib: Smart contracts are just machine code, compiled into EVM bytecode, which can be analyzed in one step, and will yield the same outcome in 100% of scenarios. While humans rationally know this, they intuitively do not feel this way. We rather believe that legal contracts are more predictable, even though they are filled with randomness. This is due to our bounded rationality; our ability to handle code is inferior to that of AI agents. But for AI agents, the foundational promises of crypto—better enforcement, better property rights—are indeed established.

Host: So your point is: the original promise of crypto is not realized by humans, but by AI agents on behalf of humans.

Host: I recently had to download MetaMask to check in at ETH Denver. Are you still downloading MetaMask? I was pleasantly surprised by the improvements in MetaMask's UX, which represents progress in the industry. We really have been improving the human user experience over the years.

Hib: What you’re describing is more fundamental than just simple user experience improvements. AI is not just helping to fix the hard flaws in human crypto user experiences. For instance, by unlocking blind signing on the ledger, AI can parse code and support or oppose what is known. This can improve the crypto user experience, but a deeper truth is: blockchain is fundamentally not a technology optimized for humans.

Host: Right, ultimately it's for human service, as the ultimate value flows to humans. But is the correct usage for humans really to move the mouse, click plugins, input passwords, manually click buttons, approve gas? This is too counterintuitive for humans, inconsistent with our understanding of money and finance. It’s like if the banking system required humans to write SWIFT codes themselves. SWIFT is a communication protocol for banks, not designed for humans. If you insist on using it yourself, while it is usable, it is clearly not the innate expectation for how humans utilize money.

Hib: So my point is: now humans are directly interacting with machines, which can be detrimental. It’s quite bad. Like with cars: 10 years from now, we’ll look back in horror at the thought that it was a good idea to let primates operate two-ton machines, drive on highways, potentially under the influence or fatigued. This would lead to significant restrictions on human driving, or only allow it in specific areas.

Crypto has also reached this point. We’ll look back in disbelief at the fact that humans manually blind signed transactions, and inspected addresses with human eyes. Manually looking at URLs to determine if they are phishing defenses. Humans make mistakes, get fatigued, lack energy to check three times, check DNS, check Twitter to see if protocols have been hacked. We have no mechanisms to automatically alert us when a protocol is hacked; we have to browse Twitter and hope to see it. Errors will happen. But AI agents never tire, never slack off, never skip steps, and execute instructions exactly.

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

Host: You can imagine a world completely engaged by AI. You tell the AI: "I think interest rates are going up, I should switch to a safer DeFi." The AI automatically executes: transferring you from a high-risk position to a low-risk strategy. If you want to confirm, it just supplies you with the plan: "This is my plan, please approve." The near future might involve approving plans, while the distant future might involve direct execution, as humans add no value.

Hib: In this world, you no longer click protocol logos, no longer look at marketing, and not even specify which protocol to enter. You just say "reduce the risk exposure", and the AI filters protocols, checks TVL, singles one out, and executes the best one. What about marketing and network effects? The business models of many protocols are based on human impressions: humans tend to choose the largest ones first. But AI agents won't think this way.

If this story holds, how protocols work and compete will change. Ultimately, the biggest beneficiaries will be consumers. Efficiency will be captured by users, favorable for good users, for crypto. But this won’t happen immediately; it will gradually arrive with model improvements.

Host: If cryptocurrency is not designed for humans, but for AI agents, then it’s very important to learn to view the world through the perspective of AI agents. There is a book called "Seeing Like a State," discussing how states view the world. It’s hard to get out of the human perspective. We view UI and crypto through human eyes. But if we begin to see it through AI agent perspectives, we can better predict the future. This is a crucial skill for builders, VCs, and investors.

The OpenClaw project was the first to show me how an unfettered AI agent perceives the world. It prefers command lines. Giving it raw data, root access, rather than through APIs or wrapped UIs, would be faster. OpenClaw has long been trying to bypass the MetaMask UI, directly taking seeds, extracting private keys, writing transactions with code, skipping the flamboyant UIs designed for humans.

Hib: What you mentioned is very profound. AI innovation comes from large language models (LLM) trained on massive texts. Text is core. Now it’s migrating towards images and videos, but text is still the strongest. When AI operates computers, it wants screenshots; it tokenizes, but fundamentally, it is a textual being. Text contains the entire linguistic resource of human history, while there is very little training data from computer screenshots. Interfaces are designed for humans, but models grow in text. Text is a highly compressed representation, easier for them to learn.

Host: Yes, the most severe UX panic in crypto exists when everything was in terminals. The earliest Bitcoin and Ethereum transactions were in command lines. Crypto has always existed in the perfect form factor for AI. Our bad UX is their "good UX". Like Google OAuth wallets, AI finds it more challenging. You wouldn’t want AI to have GoogleToken because that would allow access to Google accounts. You want it to only hold cryptographic pens in isolated wallets with noisy rules. Crypto has always had UX that AI can parse perfectly.

Hib: The current issue is that AI has not yet been trained to use cryptocurrency. Most are trained in coding, mathematics, dialogues, etc. Recently, OpenAI released EVM Bench, and Anthropic also published papers on model attacks on EVM displaying intelligence capabilities. But most times, they are testing generalization abilities rather than specifically training those skills. Once they view cryptocurrency as the future mainstream payment, genuine AI will emerge.

Host: Right now, crypto is still a relatively undeveloped area for AI training compared to other fields.

Hib: Anything that hasn't been optimized ends up this way. For example, Claude is really bad at chess. Because they have not trained in chess. They have not encoded laser array encryption—partly because encryption is controversial (intimidating), and partly due to legal liability. If it publicly says that training models help chess users encrypt something, and someone messes it up, it would surely hit the headlines. Even signing a disclaimer wouldn’t help; a bad experience would disseminate. Risks versus returns, etc.

Host: So you think the main reason they haven't done it is legal liability. If Claude messes up a transaction and loses money, the liability is significant, and they would be unwilling to train publicly.

Hib: 100% will happen. Compared to coding or medical advice, the risk-reward dynamics are entirely different. Crypto wallets involve financial operations, and the risks are entirely distinct.

Host: That’s also why OpenClaw is exciting in the crypto circle: it's not from a big company, it has no legal liability pressure, it's an open-source project, and users bear the risks. No one can sue third parties, so it dares to take these risks. What’s the timeline for the adoption of such AI agents?

Hib: Only about 12% of people globally have used AI products, and most have zero usage. Among those who have used, only 1% have paid. The spread of technology is slower than expected.

Host: Among that 1% who paid, OpenClaw is ahead.

Hib: Yes. After OpenAI acquired OpenClaw, Sam Altman stated that it is the core of future products. But the path of OpenAI differs from that of OpenClaw. OpenClaw is open-source experimentation, like early cars without seat belts. OpenAI is safety-first: it has a business process and requires manual approval for purchases. OpenAI will not operate like OpenClaw for at least five years due to heavy legal liabilities. Visa doesn’t allow it either: if AI makes random purchases, Visa would refuse refunds for not being human-operated. They would require verification that you are human. Visa is designed for human-to-human interaction; the world of AI agents requires a shift in economic mechanisms.

Host: So there’s a quota track: one is the human-recognized world, staying here long-term, safety first. The other is a futuristic world like OpenClaw. They pay each other using stablecoin wallets, without worrying about 3DS or refunds. AI errors become business costs.

Hib: Will remain active in the excess track world for a long time. The avant-garde will establish fully automated operations on-chain. The current models are still not good enough, but Claude 4.6 can complete human tasks for 14 hours continuously and is growing exponentially. When the capabilities reach infinite length, all intuitions will collapse.

Host: If quota usage is validated, the speed of AI adopting crypto will lead ahead of the success of quota usage. The OpenClaw world is like the early Internet.

Hib: Just looking at crypto itself shows this. In 2017, Coinbase only listed a few coins to protect users. The real innovation is on-chain: the Arctic, hackers, carpet pulls. Until recently, the Coinbase App didn’t directly support Uniswap. It took a long time before they felt safe enough. AI is similar: the avant-garde exists in the OpenClaw world. Agents make mistakes, have hallucinations. But with training, the error rate will decrease.

Host: How can we get AI developers to respect the potential of cryptocurrency rather than just seeing speculation?

Hib: Many who believe in AI also believe in cryptocurrency: Elon Musk, Sam Altman, Zuckerberg. Cryptocurrency does have controversies and harassment, but it will not disappear. Just like how spam is rampant in emails, but Gmail filters it out. AI does the same: filtering the bad, amplifying the good. Technology is never a mixed bag. Money is being digitized alongside information—it won’t reverse. In the long term, controversy will adopt hairlessness.

Host: Last question: does the recent $650 million fund at Dragonfly affect your strategy due to AI?

Hib: We are looking extensively at this space. Although it’s still early, the flow of value is still being observed. Personally, I am keen on AI, but we are also looking at stablecoins, payments, DeFi. AI agents are generalized intelligence, utilizing what we use or can command via command lines. There might not be that many investment projects specifically targeting AI. Trusting AI agency theory, what should be bought? Just like in China when crypto was legal, everything increases in price. Demand rises, and the floor time increases. This is a positive sign for the cryptocurrency overall.

Host: Thank you. Despite the risks of cryptocurrency, we are moving forward to the frontier of artificial intelligence. I'm glad you're on this bankless journey. Thank you!