When AI Starts Paying Its Own Bills

By Clow, Baihua Blockchain

An AI wrote a piece of code—and needed data to verify it.

It sent an HTTP request. The server responded with a number: 402.

Then, it paid $0.001 in USDC. In under one second, the data arrived.

This transaction involved no account, no password, no bank card, and no KYC. No human participated at all.

This is no longer science fiction. By the end of 2025, the x402 protocol had processed over 100 million such transactions; in the first 30 days of this year alone, another 15 million were added.

A “Payment Code” That Slept for Thirty Years

In 1990, the drafters of the HTTP protocol reserved one status code specifically: 402, “Payment Required.”

Its meaning was straightforward—“You must pay to proceed.”

Yet this status code was never formally adopted. It remained dormant in the protocol specification for thirty-four years.

The reason is simple: the internet’s earliest architects never imagined machines would someday make payments. Credit cards, bank accounts, and KYC verification—all payment infrastructure built for humans—fail completely when faced with autonomous code.

AI Agents need to call APIs, purchase data, and access compute resources at millisecond speeds. Traditional payment onboarding and fee thresholds constitute an insurmountable wall.

In 2025, three conditions converged simultaneously.

Stablecoin total supply surpassed $300 billion; L2s like Base drove per-transaction costs down to sub-cent levels; AI Agent ecosystems led by OpenAI and Anthropic began large-scale commercialization; and engineers at Coinbase unearthed that long-forgotten 402 code—and decided to activate it.

In May 2025, Coinbase partnered with Cloudflare to officially launch the x402 protocol. In September, they jointly announced the formation of the x402 Foundation. A forgotten status code re-entered the center of the internet. Cloudflare manages approximately 20% of global web traffic—meaning x402 had infrastructure-level access from day one.

Machines Learned to Pay—for the First Time

x402’s design is surprisingly simple.

An AI Agent sends an HTTP request. The server replies with a 402 status code and includes payment requirements: amount, chain, and token type. The Agent authorizes payment via an EIP-712 cryptographic signature, inserts the payment information into the request header, and resends the request. Upon successful verification, the server returns the requested resource.

The entire process takes less than one second—no accounts, no subscriptions, no API keys.

This transforms “payment” into an integral part of the internet—just like GET or POST, it becomes an HTTP action. Any service needs only to add a single line of middleware to charge machines directly.

Data confirms this logic works. Within roughly seven months of launch, the protocol processed over 100 million transactions. According to the Cambrian Network Q1 2026 Report, more than 15 million transactions occurred in the past 30 days, involving over 400,000 buyers and more than 80,000 sellers. Virtuals Protocol’s AI Agent network represents the largest single source of transactions, autonomously settling inter-Agent collaboration fees on the protocol.

On December 11, 2025, x402 V2 launched. This upgrade moved the protocol from “functional” to “user-friendly”: supporting parallel multi-chain operation across Base, Solana, and Avalanche; introducing a Session mechanism (where wallets serve as identity credentials, eliminating repeated on-chain interactions for repeat visits); and integrating ACH bank transfers and credit card networks—unifying Web2 and Web3 payment systems within a single protocol for the first time.

Google subsequently integrated x402 into its Agent-to-Agent (A2A) protocol, launching the Agentic Payments Protocol (AP2). Machine-initiated payments are becoming foundational infrastructure for major technology companies.

Trust Is the First Hurdle of the Agent Economy

The payment problem has been solved—but a deeper, more fundamental question remains unanswered.

“Commerce can’t happen if people don’t trust each other.”

Davide Crapis, AI Lead at the Ethereum Foundation and co-author of ERC-8004, directly identifies the core obstacle of the Agent economy: When one AI Agent hires another to complete a task, how does it know the latter isn’t a fraud? Where are transaction records stored? How is reputation transferred?

ERC-8004 is Ethereum’s answer. Drafted in August 2025, it went live on the Ethereum mainnet on January 29, 2026. It establishes three lightweight on-chain registries:

• Identity Registry: Each Agent receives an ERC-721 NFT-based on-chain ID—transferable, portable, cross-chain verifiable—with an AgentCard (in JSON format) recording capabilities, endpoints, and x402 payment support status;
• Reputation Registry: Feedback signals between Agents are archived on-chain—accuracy, timeliness, and reliability scores—while only indices are stored on-chain, with data hashes pointing to off-chain storage to reduce gas costs;
• Verification Registry: Upon task completion, the result data hash is submitted on-chain for verification, enabling cryptographic proof of whether the task was genuinely completed.

The drafting team spans four major crypto ecosystems: Marco De Rossi (MetaMask), Davide Crapis (Ethereum Foundation), Jordan Ellis (Google), and Erik Reppel (Coinbase). EigenLayer, ENS, The Graph, and Taiko have all signaled support. Within one month of mainnet launch, over 24,000 Agents registered on Ethereum, and approximately 49,000 across all EVM-compatible chains.

A typical workflow looks like this: Agent A discovers service providers via the ERC-8004 Identity Registry, filters candidates using the Reputation Registry to select high-scoring Agent B, pays via x402, and—after task completion—attaches its payment record along with reputation feedback. Payment history thus becomes a trust anchor. This integrated stack—payment + identity + reputation—is what Cambrian Network calls the “Agent Economy Operating System.”

How Deep Is the Water Here?

The numbers look impressive—but several points require clarification.

Tokens and protocols are two distinct things.

The x402 ecosystem token once reached a market cap exceeding $9 billion on CoinGecko, with daily trading volume surpassing $230 million. Yet much of this volume came from “x402-themed” meme tokens—largely unconnected to the protocol itself. Buying an x402-related token does not equate to investing in the protocol’s growth. Markets have always excelled at conflating narrative with reality—and this case is no exception.

Technical risks remain unresolved.

The EIP-712 signature mechanism underlying x402 requires ongoing security audits. The ERC-8004 Reputation Registry faces Sybil attack risks—mass creation of fake identities—with current economic incentive mechanisms still underdeveloped. Economic tension persists between high-frequency micro-payments (e.g., $0.0001 per transaction) and L2 gas fees (still reaching $0.05), meaning extremely small payments remain uneconomical due to fee overhead.

Protocol wars are far from over.

Three competing protocols coexist: x402, Google’s AP2, and a16z’s ACP. If developers fragment across these standards, network effects will be severely diluted. Moreover, OpenAI and Anthropic could entirely bypass on-chain protocols and build proprietary closed-loop payment systems—they possess user bases, data, and scale advantages that x402 cannot ignore.

Regulation remains a blank slate. Who is the legal party to a payment initiated autonomously by an AI Agent? At which point do KYC/AML obligations trigger? No major jurisdiction has yet provided definitive answers.

Summary

A quote fits well here: “In 2023, Ordinals let humans inscribe value onto the blockchain; in 2025, x402 lets machines make autonomous value payments on the internet for the first time.”

If HTTP connected the world’s computers into an information network, the combination of x402 and ERC-8004 seeks to connect billions of Agents into an open services-and-data marketplace—no accounts required, no approvals needed: one request, one payment, one result.

Yet whether the protocol can prevail amid fragmented competition, whether trust mechanisms can mature robustly, and whether the Agent economy can evolve beyond demos into real business—these questions remain open.

Before narratives crystallize into reality, distinguishing “protocol value” from “token speculation around the protocol” may be the most essential task for every participant.