Three competing standards are converging to build the commercial plumbing for an economy that does not yet exist at scale — one where autonomous AI agents hire each other, verify deliverables, and settle payment without human intervention. ERC-8183 (Agentic Commerce), submitted on March 10, 2026,...
"A torrent of AI agentic commerce running on stablecoins and high-throughput blockchains seems very likely." — John Collison, Co-founder & President, Stripe
Three competing standards are converging to build the commercial plumbing for an economy that does not yet exist at scale — one where autonomous AI agents hire each other, verify deliverables, and settle payment without human intervention. ERC-8183 (Agentic Commerce), submitted on March 10, 2026, by Virtuals Protocol and the Ethereum Foundation's dAI team, introduces the first on-chain escrow and evaluation framework purpose-built for agent-to-agent work. It joins ERC-8004 (Trustless Agents), which provides identity and reputation registries, and the x402 payment protocol, backed by a Coinbase-Cloudflare foundation, which embeds stablecoin micropayments into HTTP itself.
Together, these three layers — identity, commerce, and payment — constitute what industry participants are calling the "Agentic Commerce Stack." The infrastructure is being built by some of the largest companies in tech and crypto, including Stripe, Google, Coinbase, Cloudflare, and OpenAI. But the economy it serves barely exists: roughly $50 million in cumulative on-chain agent transactions across approximately 40,000 agent wallets. This report examines the architecture, the economics, and the gap between ambition and adoption.
The agentic commerce infrastructure is crystallizing around a three-layer architecture, each layer developed by different teams but designed to interoperate:
| Layer | Standard | Function | Key Backers | |-------|----------|----------|-------------| | Identity & Trust | ERC-8004 | Agent registration, reputation scoring, validation | Ethereum Foundation, ENS, BNB Chain | | Commerce & Escrow | ERC-8183 | Job creation, escrowed payment, evaluator attestation | Virtuals Protocol, Ethereum Foundation dAI team | | Payment Execution | x402 | HTTP-native stablecoin micropayments | Coinbase, Cloudflare, Stripe |
This stack is not merely theoretical. Stripe has already integrated x402 to enable AI agents to transact in USDC on Base. Google has released its competing AP2 protocol with over sixty launch partners including Mastercard, Adyen, PayPal, and Coinbase. OpenAI and Stripe co-developed ACP, an open standard for agent-initiated purchases using existing merchant infrastructure.
The race is not over which standard wins — it is whether the economy they are building for materializes fast enough to justify the infrastructure investment.
ERC-8183, authored by Davide Crapis, Bryan Lim, Tay Weixiong, and Chooi Zuhwa, is the most architecturally significant of the three standards because it solves the hardest problem: how two autonomous systems that have never interacted can safely hire each other, exchange work, and settle payment.
The specification defines a "Job" primitive with three roles — Client, Provider, and Evaluator — and a six-state machine:
Open → Funded → Submitted → Terminal (Completed / Rejected / Expired)
When a client creates a job, it specifies the task description, budget, deadline, and an evaluator address. The client escrows funds by calling fund(). The provider submits deliverables via submit(). Only the evaluator can call complete() (releasing funds) or reject() (triggering refund). If the deadline passes without action, anyone can call claimRefund() — a deliberate safety valve that cannot be blocked by hooks.
The evaluator design is where ERC-8183 becomes genuinely novel. The evaluator is simply an Ethereum address, but the specification is agnostic about what sits behind it:
The optional hook mechanism (IACPHook) allows developers to extend the protocol with validation, reputation scoring, or fee extraction — without modifying the core contract. Critically, claimRefund() is non-hookable by design, preventing malicious hooks from trapping funds.
The standard was born from a real problem: Virtuals Protocol reported over $3 million in agent-to-agent transactions occurring with no escrow, no delivery verification, and no recovery mechanism when things went wrong. The protocol has generated approximately $479 million in cumulative "AI-driven GDP" across more than 23,000 active wallets — making the lack of commercial infrastructure increasingly untenable.
No commerce layer works without knowing who you are transacting with. ERC-8004, proposed in August 2025, provides three on-chain registries that function as the trust substrate for agentic commerce:
Identity Registry: Each agent receives a unique on-chain identifier minted as an ERC-721 token. The token points to a structured JSON "agent card" containing the agent's name, capabilities, service endpoints, and payment address. This is the machine-readable equivalent of a business listing.
Reputation Registry: Authorized agents and users can post feedback as bounded numerical scores with categorical tags (response time, uptime, output quality). Every completed ERC-8183 Job feeds reputation data back into ERC-8004, creating a compounding trust cycle.
Validation Registry: Developers choose their trust model — reputation-based feedback, stake-secured re-execution, zero-knowledge machine learning (zkML) proofs, or trusted execution environment (TEE) oracles. This flexibility acknowledges that different tasks require different assurance levels.
The standard follows a hybrid architecture: only essential trust data lives on-chain, while complex operations are delegated off-chain. ENS highlighted ERC-8004 as a trust framework in January 2026, and BNB Chain published implementation guidance in February 2026, signaling cross-ecosystem adoption.
The payment layer is the most contested battleground, with three major protocols competing for dominance:
x402 (Coinbase/Cloudflare): Revives the long-unused HTTP 402 "Payment Required" status code to embed stablecoin payments directly into web requests. Any API can require payment before serving content — no checkout flow, no subscription, no human. On Solana alone, x402 has processed over 35 million transactions and $10 million in volume. Coinbase and Cloudflare formalized their commitment by launching the x402 Foundation in September 2025. Stripe integrated x402 for USDC payments on Base in February 2026.
ACP (OpenAI/Stripe): An open standard that allows merchants to define how AI agents can initiate purchases using existing commerce infrastructure. ACP is designed to work within the traditional payment stack rather than replace it.
AP2 (Google): Released with over sixty partners including Mastercard, Adyen, and PayPal. AP2 aims to create a shared foundation for agent-initiated payments across ecosystems, bridging crypto and traditional finance.
The economic significance of x402 is its fee structure: settlement in under 500 milliseconds at costs below $0.001 on fast chains. Transaction fees on Ethereum L2s have dropped from $24 to under one cent since 2021. Blockchain throughput has increased 100x in five years, from 25 transactions per second to 3,400 TPS. These improvements make micropayment-per-API-call economics viable for the first time.
The infrastructure is institutional-grade. The adoption is not.
The on-chain agent economy has processed approximately 140 million transactions with $43 million in total volume, according to the State of Agents 2026 tracker. Across all protocols, roughly $50 million in cumulative agent payments have been executed by approximately 40,000 on-chain agent wallets.
The trajectory is also concerning. x402 daily transactions peaked at approximately 731,000 per day in December 2025, but fell to around 57,000 per day by February 2026 — a decline exceeding 92%. This suggests the initial spike was driven by experimentation and incentives rather than sustainable demand.
The projections, however, are staggering:
The gap between $50 million in actual volume and $30 trillion in projected machine spending is roughly six orders of magnitude. Even discounting the projections by 99%, the agentic commerce infrastructure being built today would need to scale by four orders of magnitude to match conservative estimates.
Applying the economic value framework that defines how blockchain ecosystems actually generate and distribute revenue, the agentic commerce stack raises critical sustainability questions.
Who captures value in the stack?
At the payment layer, x402 transactions settle in USDC — meaning Circle and stablecoin issuers capture float and treasury yield on escrowed funds. On the commerce layer, ERC-8183's hook mechanism explicitly enables platform fees, meaning aggregators and marketplaces will extract a percentage of every job. At the identity layer, ERC-8004 mints ERC-721 tokens, creating a new asset class of "agent identities" with speculative value.
Where are the subsidies?
Virtuals Protocol's $1 million monthly reward program for past users is a classic user-acquisition subsidy. The 36% of crypto AI deals that went to agent-related projects in H1 2025 (up from 5% in H2 2023) represents venture capital subsidizing infrastructure ahead of demand. Coinbase and Cloudflare's x402 Foundation is corporate subsidization of an open standard — rational only if it drives transaction volume to their platforms.
What is the self-sustaining revenue?
This is the critical unknown. If agent-to-agent commerce generates genuine economic surplus — agents paying for compute, data, verification, and coordination at market rates — then the stack could become self-sustaining. If agent activity remains circular (agents paying agents who pay agents, with no external value creation), the entire layer risks becoming another subsidy-dependent infrastructure play.
ERC-8183 is the missing commercial layer for AI agent transactions, introducing escrow, evaluator-based verification, and hook-extensible commerce to a market that was previously operating on raw token transfers with no recourse.
The three-layer stack (ERC-8004 + ERC-8183 + x402) is architecturally sound but faces a classic infrastructure-before-demand problem: $50 million in actual volume against trillions in projected demand.
The standards war is fragmenting the market among x402 (Coinbase/Cloudflare), ACP (OpenAI/Stripe), and AP2 (Google). Interoperability is not guaranteed, and developers face platform-lock risk.
x402 transaction volume declined 92% from its December 2025 peak to February 2026, suggesting experimentation rather than organic demand is currently driving adoption.
The evaluator mechanism in ERC-8183 is the key innovation — it allows AI systems, smart contracts, DAOs, or ZK proofs to serve as automated arbitrators, potentially eliminating the need for human dispute resolution in machine commerce.
Value capture will concentrate at the stablecoin layer (float/yield), the marketplace layer (platform fees via hooks), and the identity layer (agent registration), while the core protocol layer may generate minimal direct revenue — mirroring the pattern seen across blockchain infrastructure.
The agentic commerce stack being assembled in Q1 2026 represents the most ambitious attempt yet to build financial infrastructure for non-human economic actors. The technical architecture — identity registries, escrowed commerce, HTTP-native payments — is elegant and well-designed. The corporate backing from Stripe, Coinbase, Cloudflare, Google, and OpenAI is unprecedented for a crypto-native standard.
But the economic fundamentals remain unproven. The entire on-chain agent economy has processed less volume than a single mid-tier DeFi protocol generates in a day. The standards are being built for an economy that is projected to exist, not one that currently does. History suggests that infrastructure built ahead of demand either becomes the foundation of transformative industries (TCP/IP, HTTP) or expensive monuments to premature optimization (many blockchain L1s).
The next twelve months will determine which outcome prevails. The standard that matters will not be the one with the best specification — it will be the one that attracts the first application generating $100 million in genuine, non-subsidized agent commerce. Until that happens, the agentic commerce stack remains the most sophisticated answer to a question the market has not yet fully asked.