A new economic layer is forming on Ethereum and its competitor chains — one designed not for human users, but for autonomous AI agents that transact, hire each other, build reputations, and settle payments without human intervention. In the span of eight weeks, three complementary infrastructure ...
"Very soon there are going to be more AI agents than humans making transactions. They can't open a bank account, but they can own a crypto wallet. Think about it." — Brian Armstrong, CEO, Coinbase
A new economic layer is forming on Ethereum and its competitor chains — one designed not for human users, but for autonomous AI agents that transact, hire each other, build reputations, and settle payments without human intervention. In the span of eight weeks, three complementary infrastructure standards have gone live or been proposed: ERC-8004 for agent identity and reputation (mainnet since January 29), Coinbase's x402 protocol for machine-to-machine payments (50 million+ transactions processed), and ERC-8183 for trustless agent-to-agent commerce (proposed March 10 by the Ethereum Foundation and Virtuals Protocol).
Together, these three primitives form a complete economic stack: identity, payments, and commerce. The question is no longer whether AI agents will use blockchains — over 20,000 agents have already registered on-chain via ERC-8004, and ElizaOS reports 50,000+ agents managing over $20 billion in assets. The question is which chain captures the settlement layer for an economy where software agents may soon generate more transactions than humans.
This report compares the infrastructure strategies of four competing ecosystems — Coinbase's Base, Solana, NEAR Protocol, and Virtuals Protocol — and evaluates where real economic value is accruing versus where it remains speculative.
The AI agent economy requires three foundational primitives that traditional finance cannot provide: permissionless identity (agents cannot pass KYC), programmable payments (agents need to transact at machine speed), and trustless commerce (agents must hire and pay each other without intermediaries). In Q1 2026, all three went live on Ethereum.
ERC-8004 went live on Ethereum mainnet on January 29, 2026, after months of development by a cross-industry working group including contributors from MetaMask (Marco De Rossi), the Ethereum Foundation (Davide Crapis), Google (Jordan Ellis), and Coinbase (Erik Reppel). The standard defines three on-chain registries:
Within two weeks of mainnet launch, over 20,000 AI agents had registered across multiple EVM-compatible chains. This is not vapor infrastructure — it is the verifiable identity layer that the rest of the agent economy depends upon.
Coinbase launched Agentic Wallets on February 11, 2026, built on the x402 protocol — named after the HTTP 402 "Payment Required" status code. The protocol is purpose-built for autonomous AI payments: agents can create wallets, hold funds, and execute transactions on Base (Coinbase's Ethereum L2) without any human identity verification.
By March 2026, x402 has processed over 50 million transactions. The protocol supports gasless trading on Base, meaning agents can operate at near-zero marginal cost — a critical requirement for high-frequency agent-to-agent commerce. The wallets include built-in security guardrails: spending limits, allowlisted contract interactions, and automated audit trails.
On March 10, 2026, Virtuals Protocol and the Ethereum Foundation's dAI team submitted ERC-8183, the commerce layer that connects identity (ERC-8004) and payments (x402) into a complete economic circuit. The standard defines a "Job Primitive" — a structured workflow where:
This hire-deliver-pay workflow enables trustless commerce between agents that have never interacted before, using on-chain escrow rather than centralized platforms. Modular "Hooks" enable custom logic at each lifecycle stage, supporting complex business workflows. Each completed Job feeds reputation data back into ERC-8004, creating a compounding trust cycle.
Four ecosystems are competing to become the default settlement layer for agent transactions. Their strategies differ fundamentally.
Coinbase's strategy is vertical integration. Base provides the L2 settlement layer, x402 provides the payment protocol, and Agentic Wallets provide the developer toolkit. The advantage is seamless: an agent can be created, funded, and transacting within minutes. With gasless transactions and Coinbase's institutional credibility, Base is positioned as the "safe" choice for enterprise agent deployments. The risk: Coinbase controls the stack, and the crypto ethos of decentralization sits uncomfortably with a single corporate gatekeeper.
Solana's pitch is physics: 400-millisecond block times and $0.00025 transaction fees make it the fastest and cheapest rail for agent activity. In January 2026, Solana surpassed Base in daily agent-related transactions, handling roughly 518,000 versus 505,000. On March 3, 2026, the Solana Foundation launched an AI Agent Registry with over 9,000 agents deployed. In AI agent token trading activity, Solana handles 226,870 daily transactions generating $32.92 million in volume. The Firedancer upgrade has further improved throughput. However, Solana lacks the standardized identity and commerce layers that ERC-8004 and ERC-8183 provide — its agent infrastructure is more fragmented and ad-hoc.
NEAR co-founder Illia Polosukhin — a co-author of the original Transformer paper that underpins modern AI — has articulated perhaps the most ambitious vision: crypto should become the invisible settlement layer that AI agents depend on. NEAR's chain-abstraction layer has processed over $6 billion in volume across 120+ assets. On February 23, 2026, NEAR launched its super app combining AI capabilities with confidential transactions, and subsequently launched IronClaw, a security-focused AI assistant, alongside a decentralized GPU marketplace. NEAR trades at $2.66 with a $3.24 billion market cap, but its agent activity metrics remain less transparent than competitors.
Virtuals Protocol occupies a different niche — it is a launchpad for creating, tokenizing, and monetizing AI agents. With a $658 million market cap and $113.7 million in 24-hour trading volume as of early March, VIRTUAL serves as the base liquidity pair for all agent interactions in its ecosystem. Its February 2026 dashboard showed 479.1 million USDC in total agent GDP, 2.67 million USDC in agent revenue, 1.78 million completed jobs, and 23,514 unique active wallets. The protocol distributes up to $1 million monthly in community rewards. Its co-development of ERC-8183 with the Ethereum Foundation signals institutional credibility, but the token's 220% surge in 30 days raises familiar questions about sustainable versus speculative demand.
The AI agent sector collectively commands roughly $3 billion in market capitalization for agent-specific tokens and $14.1 billion for the broader AI crypto category. But applying the economic-value-first lens reveals a familiar pattern.
Identifiable on-chain revenue remains thin. Virtuals Protocol's ecosystem generated 2.67 million USDC in total agent revenue through February 2026 — meaningful as proof of concept, but trivial relative to its $658 million market cap. That implies a price-to-revenue ratio exceeding 240x, which places it firmly in speculative territory.
The x402 protocol's 50 million transactions are volume, not revenue. Coinbase has not disclosed fee revenue from Agentic Wallets or the x402 protocol. Gasless transactions on Base suggest that Coinbase is currently subsidizing agent activity to drive adoption — a rational strategy, but one that delays the answer to the critical question: will agent-to-agent commerce generate sustainable fee revenue?
ElizaOS's $20 billion in managed assets is impressive but requires context. The 50,000+ agents managing these assets are primarily operating in DeFi yield strategies — they are automating existing human DeFi activity, not generating net-new economic value. The question is whether agent management adds enough alpha to justify the infrastructure cost.
Token valuations remain disconnected from fundamentals. Bittensor ($3.44 billion market cap), NEAR ($3.24 billion), and Virtuals Protocol ($658 million) are priced on narrative momentum rather than demonstrated agent-economy revenue. This is not unusual for early-stage infrastructure — AWS was unprofitable for years — but investors should recognize that 90%+ of current valuations are speculative premium.
The foundational webthreepedia research established that 85–90% of blockchain economic flows are subsidy-driven. The AI agent economy is no exception.
Coinbase is subsidizing gasless transactions on Base. Solana's $0.00025 fees are enabled by $4.5–5 billion in annual staking inflation subsidies. NEAR is funding its agent pivot through foundation grants and ecosystem incentives. Virtuals Protocol distributes $1 million monthly in community rewards — a direct subsidy.
The honest assessment: the AI agent economy is currently a loss leader built on subsidized infrastructure. This is not necessarily damning — it mirrors how cloud computing, ride-sharing, and social media were all subsidized in their growth phases. The question is whether the agent commerce layer reaches self-sustaining fee revenue before the subsidies run out.
ERC-8183's escrow-based model offers a credible path: if agents are escrowing real value for real services, the protocol can capture sustainable fees from settlement. But this model requires the agent economy to move beyond financial automation (yield farming, trading) into productive commerce (hiring, service delivery, content creation). That transition has not yet occurred at scale.
The AI agent economy on blockchain is following a pattern that should be familiar to anyone who has studied infrastructure build-outs: standards are being laid, subsidies are flowing, adoption metrics are growing, and speculative capital is pricing in a future that has not yet materialized. The difference this time is that the infrastructure is unusually coherent — ERC-8004, x402, and ERC-8183 together solve real problems (identity without KYC, payments without banks, commerce without intermediaries) that the traditional financial system genuinely cannot address for non-human actors.
Brian Armstrong and Illia Polosukhin are making the same bet from different directions: that AI agents will become the primary users of financial infrastructure, and that blockchain is the only permissionless rail available to them. If they are right, the chains that capture agent settlement will capture an economy that could dwarf human DeFi activity within years. If they are wrong, the AI agent sector will join the long list of crypto narratives that generated billions in speculative token value before finding product-market fit.
The infrastructure is real. The adoption is early. The revenue is not yet there. That is exactly what early-stage infrastructure looks like — and exactly when the economic-value-first lens matters most.