When AI begins to pay for itself

Writing by: Clow, Plain Language Blockchain

An AI wrote a piece of code that requires data for verification.

It sent an HTTP request, and the server returned a number: 402.

Then, it paid 0.001 USD using USDC. Less than a second later, the data was back.

This transaction involved no account, no password, no bank card, no KYC. The entire process involved no human participation.

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

The Sleeping Payment Code for Thirty Years

In 1990, the draft of the HTTP protocol reserved a status code: 402, Payment Required.

The meaning was straightforward—“Payment needed to proceed.”

But this status code was never officially used. It just remained in the protocol documentation, dormant for thirty-four years.

The reason is simple: the early builders of the internet never imagined that one day machines would handle payments. Credit cards, bank accounts, KYC verification—these payment infrastructures were designed for humans and completely failed in autonomous code execution.

AI agents need to call APIs, purchase data, and acquire computing power at millisecond speeds. Traditional payment account registration and transaction fees are insurmountable barriers.

By 2025, three conditions aligned.

The total supply of stablecoins surpassed $300 billion, and Layer 2 solutions like Base reduced transaction costs to sub-cent levels; AI agent ecosystems led by OpenAI and Anthropi began large-scale commercialization; Coinbase engineers rediscovered the long-forgotten 402 status code and decided to activate it.

In May 2025, Coinbase, in partnership with Cloudflare, officially launched the x402 protocol. By September, they announced the formation of the x402 Foundation. A forgotten status code was brought back to the center of the internet. Cloudflare manages about 20% of global web traffic—meaning x402 had infrastructure support from day one.

Machines Learned to Pay for the First Time

The design of x402 is surprisingly simple.

An AI agent initiates an HTTP request, and the server responds with a 402 status code, along with payment requirements: how much, which chain, which token. The agent signs the payment info with an EIP-712 encrypted signature, embeds it in the request headers, and resends. Once verified, the server returns the resource.

The entire process takes less than a second, with no accounts, no subscriptions, no API keys.

This turns “payment” into a part of the internet. Like GET or POST, it’s just an HTTP action—any service can add a middleware to charge machines.

Data proves this logic works. Within about seven months of release, the protocol processed over 100 million transactions. According to Cambrian Network’s Q1 2026 report, in the past 30 days, over 15 million transactions occurred, with more than 400,000 buyers and over 80,000 sellers. The largest source of transactions is the Virtuals Protocol AI agent community, which autonomously settles collaboration fees between agents on the protocol.

On December 11, 2025, x402 V2 was launched. This upgrade moved the protocol from “usable” to “user-friendly”: supporting multiple chains like Base, Solana, Avalanche; introducing a Session mechanism (wallets as identity credentials, no need for repeated on-chain interactions); integrating ACH bank transfers and credit card networks—connecting Web2 and Web3 payment systems for the first time within this protocol.

Google later integrated x402 into the Agent2Agent (A2A) protocol, releasing the Agentic Payments Protocol (AP2); machine payments are becoming a foundational infrastructure consensus for major tech companies.

Trust: The First Hurdle in the Agent Economy

While payment issues are solved, a more fundamental problem remains.

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

Davide Crapis, head of AI at the Ethereum Foundation and co-author of ERC-8004, directly pointed out the core obstacle of the agent economy: when one AI agent needs to hire another to complete a task, how does it know the other isn’t a scammer? Where are the transaction records? How is reputation transmitted?

ERC-8004 is Ethereum’s answer to this problem. Drafted in August 2025 and officially on the mainnet on January 29, 2026, it establishes three lightweight on-chain registries:

• Identity Registry: Each agent gets an on-chain ID based on an ERC-721 NFT, which is portable, transferable, and cross-chain searchable. The AgentCard (JSON format) records capabilities, endpoints, and x402 payment support status.

• Reputation Registry: Feedback signals between agents are stored on-chain—accuracy, timeliness, and reliability scores. Only indexes are stored on-chain; data hashes point to off-chain storage, reducing gas costs.

• Verification Registry: After task completion, the hash of the result data is uploaded for verification, providing cryptographic proof that the task was genuinely completed.

The drafting team spans four major crypto ecosystems: Marco De Rossi from MetaMask, Davide Crapis from Ethereum Foundation, Jordan Ellis from Google, Erik Reppel from Coinbase. EigenLayer, ENS, The Graph, and Taiko have all expressed support. Less than a month after mainnet deployment, over 24,000 agents registered on Ethereum, totaling about 49,000 across all EVM chains.

A typical workflow looks like this: Agent A discovers service providers via ERC-8004 identity registry, filters high-reputation agents from the reputation registry, uses x402 to make payments, and after the task, attaches reputation feedback—payment history becomes a trust anchor. This chain is what Cambrian Network calls the “Agent Operating System”: payment + identity + reputation—three layers integrated.

How deep is this water?

The data looks promising, but a few things need clarification.

Tokens and protocols are not the same.

The x402 ecosystem token once exceeded a market cap of $9 billion on CoinGecko, with daily trading volume over $230 million. But many “x402 concept tokens” are meme tokens with no real binding to the protocol itself. Buying x402-related tokens doesn’t mean buying into the protocol’s growth. The market has always been good at mixing narrative with reality, and this time is no exception.

Technical risks remain unresolved.

x402’s EIP-712 signature mechanism requires ongoing security audits. The ERC-8004 reputation registry faces Sybil attack threats—mass registration of fake identities, with current economic incentives still imperfect. Micro-payments of $0.0001 per transaction versus Layer 2 fees (which can still reach $0.05) create economic tension; very small transactions are still eaten up by fees.

The protocol war isn’t over.

x402, Google’s AP2, and a16z’s ACP coexist. If developers split among three standards, network effects weaken. Moreover, OpenAI and Anthropic can bypass on-chain protocols to build their own closed-loop payment systems—they have users, data, and scale advantages, posing serious competitive pressure that x402 cannot ignore.

Regulatory issues remain a blank space. Who is the actual transaction subject when an AI agent initiates a payment? Where are the KYC/AML trigger points? No major jurisdiction has provided clear answers.

Summary

Someone once wrote a phrase worth quoting here: “In 2023, inscriptions let humans embed value on-chain; by 2025, x402 enables machines to autonomously pay value on the network for the first time.”

If HTTP connected computers worldwide into an information network, then the combination of x402 and ERC-8004 aims to connect hundreds of millions of agents into an open service and data marketplace—no accounts, no approvals needed. One request, one payment, one result.

However, whether the protocol can succeed amid fragmentation, whether trust mechanisms can truly mature, and whether the agent economy can move from demo to real business—all remain open questions.

Before the narrative materializes, clarifying the “value of the protocol” versus the “tokens hyped around the protocol” is perhaps the most important thing for every participant to understand.