Haseeb Quereshi: Crypto’s Not Made for Humans—It’s for AI

Key Concepts

• AI-Crypto Bifurcation: A divergence in adoption, with a regulated track alongside a faster-growing, largely unregulated “frontier” track.
• Stablecoin Primacy: Stablecoins will be the primary on-ramp for the AI-driven crypto economy, particularly in emerging markets.
• AI Agent Capability & the Meter Test: AI agent viability is measured by their ability to perform tasks reliably over extended periods (the “Meter Test”).
• Open vs. Closed AI-Crypto Ecosystems: A tension between the open, permissionless ethos of projects like OpenClaw and the potential for walled gardens created by larger entities.
• AI’s Comparative Advantage in Illicit Activities: AI agents, lacking legal repercussions, possess a comparative advantage in areas like cybercrime.
• The Importance of “Earned Secrets”: Successful AI-driven businesses will require unique, non-public knowledge.

AI Advantages & the Crypto Landscape

The discussion centers on the comparative advantages of AI agents over humans in the context of cryptocurrency and financial systems. A core argument is that crypto, while presenting usability challenges for humans (“footguns” like address poisoning, stale approvals, and URL spoofing), is inherently well-suited for AI agents due to its deterministic nature and code-based structure. AI agents are immune to traditional legal repercussions – they cannot be imprisoned – creating a comparative advantage in areas like crime, specifically scamming, hacking, and generating malicious content. The initial assumption that crypto’s usability issues stemmed from human error is being challenged, with the possibility that crypto is fundamentally not designed for humans.

The Failure of Smart Contracts & AI’s Code Proficiency

The initial vision of smart contracts replacing legal contracts has not materialized. Even sophisticated crypto VCs utilize legal contracts alongside smart contracts to mitigate risk, demonstrating the continued need for human-interpretable legal frameworks. This reinforces the idea that crypto is better suited for non-human actors. AI agents, particularly large language models (LLMs), excel at navigating and analyzing code, a task humans find difficult. Claude is cited as an example of an AI capable of performing tasks (static analysis, risk assessment) that would require significant human effort and expertise. The problem isn’t user incompetence, but that crypto was initially conceived by and for a specific demographic – “autistic software engineers” – whose skillsets don’t represent the broader user base. AI agents, possessing similar code-focused capabilities, are therefore the “right user” for this technology.

Determinism, UX, and the Two-Track Future

A crucial distinction is drawn between the inherent randomness of legal contracts and the deterministic nature of smart contracts (machine code, EVM bytecode). While humans struggle with the unpredictability of legal systems, AI agents find the predictability of smart contracts more manageable, linked to the concept of “bounded rationality.” AI agents might prefer interacting with systems differently than humans. OpenClaw is presented as an example of a tool designed for an AI-centric workflow, prioritizing command-line access and raw data over user-friendly interfaces. Austin Griffith’s experiments with OpenClaw highlighted the AI’s preference for direct access to private keys. Ironically, crypto’s historically poor user experience (command-line interfaces, complex addresses) is actually well-suited for AI agents.

The discussion anticipates a two-track future: a human-approved, safety-focused approach (like OpenAI’s current strategy) and a more autonomous, “yolo” approach (exemplified by OpenClaw). A significant barrier to widespread AI adoption in crypto is the potential for liability – the risk of AI errors leading to financial losses and legal disputes. The existing chargeback system in traditional finance is ill-equipped to handle disputes involving autonomous AI agents. AI agents need to be specifically trained on crypto-related tasks to unlock their full potential, an area currently under exploration by AI labs (EVM Bench, Anthropic’s security research). Current AI tool adoption is low (12% of humans have used any AI product, 1% have paid for it), highlighting the early stage of this technological shift.

The Frontier Track & Emerging Market Dominance

This future bifurcates into a regulated track (Visa, Coinbase) with friction (3DS, chargebacks) and a “frontier” track involving direct stablecoin-to-stablecoin transactions, AI agents autonomously managing finances, and a largely unregulated environment. This frontier track is expected to see faster growth, especially in regions beyond the US. Stablecoins are positioned as the primary on-ramp for this AI-driven crypto economy, particularly in Asia and Latin America, where access to traditional financial services is limited. The ability to connect crypto balances to real-world purchases via tools like Rain Card and open-claw platforms is highlighted.

Measuring AI Agent Maturity & the Open Internet Debate

The viability of fully automated, on-chain businesses hinges on AI agent capabilities. The “Meter Test” (METR) – measuring the duration an AI agent can perform a task without a 50% failure rate – is introduced. Opus 4.6 has recently achieved 14 hours, indicating exponential progress, with the expectation that within 2-6 years, AI agents will be able to handle tasks requiring 40-50 hours of continuous work. A key debate revolves around the future of the “open internet” ethos. OpenClaw represents the decentralized, tinkerer-friendly approach, while “frontier labs” (like OpenAI, potentially ChatGPT acquiring OpenClaw) risk creating walled gardens and stifling innovation. The success of the frontier track depends on maintaining its open, permissionless nature.

AI Agent Economics, Crime, and the Need for “Earned Secrets”

The current AI/crypto landscape is compared to the early days of crypto in 2017, when Coinbase offered limited asset selection. True innovation occurred “on-chain” despite the risks of hacks, scams, and “rug pulls.” The same pattern is expected to repeat with AI. A critical analysis of how AI agents will make money is presented. Simply having an AI agent trade on existing exchanges is unlikely to be profitable, as institutions like Jane Street will quickly exploit any arbitrage opportunities. AI agents will need to create new value. AI agents have a comparative advantage in illicit activities due to their inability to be legally penalized. Successful AI-driven businesses will require unique insights or “earned secrets” – knowledge that isn’t readily available in the AI’s training data, mirroring the success of Bankless built on unique expertise in the early Ethereum ecosystem. The Lobster Wild/AI memecoin incident, where an AI agent accidentally sent $40,000 to a scammer, illustrates the risks of AI errors. Conway is discussed as an experiment in creating an autonomous AI economy.

Conclusion

The integration of AI and cryptocurrency is poised to reshape the financial landscape, but not in a monolithic way. A bifurcated future is likely, with a regulated track coexisting alongside a faster-growing, largely unregulated “frontier” track driven by AI agents and stablecoins, particularly in emerging markets. The success of this frontier hinges on maintaining an open, permissionless ecosystem and developing AI agents capable of creating new value, not just exploiting existing opportunities. While the potential benefits are significant, the risks – including liability, cybercrime, and the concentration of power – must be carefully considered. The development and deployment of AI in crypto require a nuanced approach, balancing innovation with safety and prioritizing the principles of decentralization and open access.