Timothy Morano
Apr 16, 2026 17:48
Andreessen Horowitz researchers identify identity, payments, and governance as critical gaps where crypto rails can enable AI agents as economic actors.
AI agents have outpaced the infrastructure meant to support them. A new research piece from Andreessen Horowitz argues that blockchains offer the missing coordination layer these autonomous systems need to function as legitimate economic actors.
The six-author team from a16z crypto—including Christian Catalini and Andrew Hall—identifies five specific areas where traditional systems fall short and blockchain primitives already work.
The Identity Problem Is Real
Here’s a striking stat: In financial services alone, non-human identities (automated trading systems, risk engines, fraud models) outnumber human employees roughly 100 to 1. Yet these agents remain “effectively unbanked,” the researchers note. They can interact with financial systems but lack portable, verifiable credentials that travel across platforms.
The a16z team proposes “KYA” (know your agent) as the AI equivalent of KYC—cryptographically signed credentials linking agents to their principals, permissions, and reputation. Early implementations are already emerging: onchain agent registries, wallet-native agents using USDC, and ERC standards for “trust-minimized agents.”
“Until a common identity standard emerges, merchants will keep blocking agents at the firewall,” the paper states.
Payments Are Already Flowing
Agent-to-agent commerce isn’t theoretical. Stripe and Tempo’s MPP marketplace processed over 34,000 transactions in its first week, with fees as low as $0.003 and stablecoins as a default payment method. These “headless merchants”—just servers, endpoints, and prices per call—have no checkout pages or storefronts.
Coinbase’s x402 protocol is processing roughly $1.6 million monthly in agent-driven payments after filtering out wash trading, according to the researchers. That’s well below the $24 million figure Bloomberg recently cited, but the infrastructure is scaling fast. Stripe, Cloudflare, Vercel, and Google have all integrated x402.
Why crypto rails? Traditional payment processors struggle to underwrite headless merchants with no websites or legal entities. Stablecoins let any developer make an endpoint payable without signing a merchant agreement.
Governance Gets Complicated
When AI systems manage resources—allocating capital, running supply chains—who’s actually in control? The a16z team warns that formal governance can look decentralized while the operational AI layer remains centralized. Whoever controls the model ultimately controls outcomes.
Their proposed solution: cryptographic guarantees at multiple levels showing exactly what training data produced a model, what instructions govern an agent, and records of what it actually did. Without these, “governance by agents collapses back into governance by whoever controls the model weights.”
Trust Becomes the Scarce Resource
The paper’s most provocative argument: when AI drives execution costs toward zero, verification becomes the binding constraint. “Agent throughput already dwarfs human oversight capacity,” the researchers write. The “human in the loop” is becoming physically impossible at scale.
Blockchains shift the economics by providing auditable history rather than black-box outputs. Systems moving money via stablecoins and smart contracts can simultaneously carry cryptographic receipts showing who did what.
What This Means for Builders
Several crypto-native tools are already addressing these gaps. NEAR Intents has processed over $15 billion in cumulative DEX volume since Q4 2024 using intent-based architecture. MetaMask’s Delegation Toolkit, Coinbase’s AgentKit, and Merit Systems’ AgentCash let users define at the smart contract level what agents can and cannot do.
The infrastructure buildout is happening now. Whether it prioritizes transparency and user control—or gets bolted onto legacy systems never designed for non-human actors—remains the open question.
Image source: Shutterstock
