A new class of economic actor is colonizing decentralized finance — and it isn't human. In the span of three weeks, Coinbase launched Agentic Wallets purpose-built for autonomous AI systems (February 11), OpenAI and Paradigm released EVMbench to stress-test AI agents against real smart contract v...
"A machine will not pay a 2-3% interchange fee charged by companies like Mastercard and Visa if it can settle a transaction for a fraction of a penny." — Simon Gerovich, CEO of Metaplanet
A new class of economic actor is colonizing decentralized finance — and it isn't human. In the span of three weeks, Coinbase launched Agentic Wallets purpose-built for autonomous AI systems (February 11), OpenAI and Paradigm released EVMbench to stress-test AI agents against real smart contract vulnerabilities (February 18), and Uniswap shipped seven open-source AI Skills enabling machines to execute swaps, deploy pools, and manage liquidity on v4 without human intervention (February 20). Meanwhile, over 24,000 AI agents have registered on-chain under Ethereum's new ERC-8004 trustless agent standard, which went live on mainnet in late January.
This is not a speculative narrative cycle. It is an infrastructure buildout. The largest protocol teams in crypto — Coinbase, Uniswap Labs, OpenAI, Paradigm — are simultaneously shipping production-grade tooling designed for a world where software agents, not retail traders, are the marginal user of on-chain financial rails. The implications for DeFi's economic model, security architecture, and value distribution are profound.
Yet the market for AI agent tokens tells a different story: the sector suffered a 44% wipeout in early 2025 and remains volatile, with Virtuals Protocol — the leading AI agent launchpad — sitting at a $414 million market cap, down sharply from its highs. The gap between infrastructure investment and token speculation is the central tension of the AI-agent economy. This report examines who is building what, where the real economic value lies, and what happens when machines become DeFi's primary users.
The February 2026 infrastructure push is unprecedented in its coordination. Three of crypto's most consequential organizations simultaneously shipped agent-native tooling, signaling that the industry has moved past the "will AI agents use DeFi?" question and into the "how do we serve them?" phase.
Coinbase launched the first wallet infrastructure explicitly designed for AI agents. Built on the company's x402 payment protocol — which has already processed over 50 million machine-to-machine transactions — Agentic Wallets enable autonomous systems to spend, earn, and trade cryptocurrency without human intervention. The wallets operate primarily on Base, Coinbase's Layer-2 network, and include programmable spending limits and security guardrails.
The x402 protocol embeds payments directly into HTTP requests: when an AI agent requests a paid resource, the server returns an HTTP 402 status with payment requirements, and the agent's wallet handles settlement automatically. This transforms every API endpoint into a potential revenue stream for the machine economy. Agents can autonomously purchase compute, access premium data feeds, pay for storage, and settle with other agents — all without a human touching a wallet.
In a partnership that would have been unthinkable two years ago, OpenAI and Paradigm released EVMbench, an open-source benchmark evaluating AI agents' ability to detect, patch, and exploit smart contract vulnerabilities. The benchmark draws on 120 curated vulnerabilities from 40 audits, primarily sourced from Code4rena competitions.
The results are startling. GPT-5.3-Codex can now exploit over 70% of critical, fund-draining smart contract bugs — up from less than 20% when the project began. Smart contracts routinely secure over $100 billion in assets on the Ethereum Virtual Machine; the fact that leading AI labs are now building dedicated tooling for this attack surface signals that autonomous security — both offensive and defensive — is becoming an existential concern for DeFi.
Uniswap Labs released seven open-source AI Skills for Uniswap v4, providing structured, command-based access to core protocol functions: security foundations, configuration, deployment, Viem integration, swap execution, liquidity planning, and swap planning. Available in both Python and TypeScript, these Skills replace ad-hoc scripts with standardized automation that reduces failed transactions and tightens slippage control.
The strategic signal is clear: Uniswap is positioning itself not merely as a human-facing DEX, but as execution infrastructure for autonomous on-chain agents. If adoption scales, 2026 could mark the shift from manual DeFi activity toward machine-native financial execution.
Underlying all of this infrastructure is ERC-8004, formally titled "Trustless Agents," which went live on Ethereum mainnet on January 29, 2026. The standard was co-authored by contributors from MetaMask, the Ethereum Foundation, Google, and Coinbase — a cross-industry coalition that underscores the seriousness of the initiative.
ERC-8004 establishes three on-chain registries: identity, reputation, and validation. Together, these registries create a trust layer that allows AI agents to function as independent economic actors on the blockchain. Agents can be discovered, their work history evaluated, and their reliability scored — all without centralized intermediaries.
Adoption has been rapid. Over 24,000 agents registered in the first weeks after launch, with deployments expanding beyond Ethereum to Avalanche's C-Chain and major Layer-2 networks. This is the ERC-20 moment for machine identity: a shared standard that enables interoperability across the agent ecosystem.
The economic implications are significant. ERC-8004 transforms AI agents from anonymous scripts into accountable entities with on-chain track records. This enables agent-to-agent commerce, reputation-based pricing, and automated trust evaluation — the foundational primitives for a machine economy that operates at speeds and scales impossible for human participants.
The February infrastructure push also revealed a deeply uncomfortable truth: the same AI capabilities that can defend DeFi can also attack it.
A study released on February 20 by AI security firm Cecuro evaluated 90 real-world smart contracts exploited between October 2024 and early 2026, representing $228 million in verified losses. A purpose-built AI security agent detected vulnerabilities in 92% of those contracts (covering $96.8 million in exploit value), compared with just 34% detection for a baseline GPT-5.1-based coding agent running on the same underlying model. The difference came entirely from domain-specific security methodology — not from the model itself.
More alarming: the research found that AI exploit capability is doubling roughly every 1.3 months, while the average cost of an AI-powered exploit attempt has fallen to approximately $1.22 per contract. Several of the exploited contracts in the dataset had previously undergone professional human audits.
This creates a structural transformation in DeFi security economics. When the cost to scan every vulnerable contract on Ethereum approaches zero, the attack surface becomes functionally infinite. The only viable defense is equally automated: AI security agents that can monitor, detect, and respond at machine speed. DeFi is entering an arms race where both attackers and defenders are artificial intelligence systems, and the humans are increasingly bystanders.
Applying webthreepedia's economic value framework to the emerging AI agent economy reveals a familiar pattern: infrastructure providers capture disproportionate value while token holders absorb speculative risk.
Infrastructure Value Capture:
On-Chain Fee Implications: The 1kx 2026 onchain revenue forecast projects $32+ billion in on-chain fees for 2026 (63% year-over-year growth), with DeFi continuing to expand at over 50% annually. If AI agents drive even a fraction of this growth — through higher-frequency trading, automated liquidity management, and machine-to-machine settlement — the fee revenue accruing to protocols and validators could increase substantially without any growth in human user counts.
The Subsidy Question: Critically, unlike most DeFi activity which our foundational analysis shows is 85-90% subsidy-driven, agent-to-agent transactions have the potential to be genuinely fee-sustaining. A machine optimizing yield does not need airdrops or token incentives; it needs execution quality and low latency. If this thesis holds, AI agents could represent the first class of DeFi user that generates sustainable on-chain revenue — the elusive "real demand" that the industry has been seeking.
Despite the infrastructure boom, the AI agent token market tells a cautionary tale. The sector suffered a 44% market cap wipeout in early 2025, and the leading platform token — Virtuals Protocol (VIRTUAL) — sits at approximately $414 million market cap as of late February 2026, with $42 million in daily trading volume. This is a fraction of its previous highs.
The top 10 AI agent tokens by market cap — including VIRTUAL, KITE, and FET (Artificial Superintelligence Alliance) — collectively represent a small slice of crypto's total market capitalization. The disconnect is instructive: the serious infrastructure investment is flowing through equity (Coinbase), protocol revenue (Uniswap), and research partnerships (OpenAI/Paradigm), not through speculative token launches.
This mirrors a pattern identified in our economic value analysis: the entities that capture durable economic value in crypto are those with defensible revenue streams — not those that issue tokens into speculative markets. The AI agent economy is being built by companies with business models, not by token projects seeking narrative premium.
The infrastructure is production-grade. Coinbase's x402 protocol has processed 50M+ transactions. Uniswap's AI Skills ship in Python and TypeScript for immediate integration. OpenAI's EVMbench benchmarks real exploit scenarios. This is not vaporware.
24,000+ agents are already on-chain. ERC-8004 adoption surpassed expectations within weeks of mainnet launch, creating the identity and reputation primitives required for agent-to-agent commerce.
AI exploit capability is doubling every 1.3 months. The security implications are existential for DeFi protocols that rely on static human audits. Automated, AI-powered defense is now a requirement, not an option.
The value accrues to infrastructure, not tokens. Coinbase, Uniswap, and OpenAI are building revenue-generating businesses. AI agent tokens remain speculative and have underperformed the broader infrastructure thesis.
Agents may be DeFi's first sustainable user class. Unlike human retail traders who require airdrops and incentives, AI agents optimize for execution quality — potentially generating genuine, subsidy-free on-chain fee revenue.
February 2026 may be remembered as the month DeFi's user model fundamentally changed. The simultaneous launches by Coinbase, Uniswap, and OpenAI/Paradigm are not coordinated marketing — they are independent responses to the same underlying reality: AI agents are becoming economically significant actors on-chain, and the protocols that serve them best will capture the next wave of DeFi revenue growth.
The economic value framework is clear. In a blockchain economy where 85-90% of value flows are subsidy-driven, autonomous agents represent a potential path to sustainability. Machines do not need governance token bribes to provide liquidity. They do not need airdrop campaigns to try a new protocol. They optimize ruthlessly for cost and execution quality — exactly the economic behavior that generates real fee revenue.
But this transition comes with profound risks. The same AI capabilities that automate yield optimization also automate exploit discovery. The cost of attacking a smart contract is collapsing toward zero, and only equally automated defense can keep pace. DeFi's next security crisis will likely be machine versus machine, playing out at speeds no human governance system can match.
The machine economy is not coming. It is here — 24,000 agents strong and growing. The question is no longer whether AI agents will reshape DeFi, but whether the protocols and infrastructure providers positioned to serve them can convert this unprecedented buildout into sustainable economic value. The data suggests they can. The tokens suggest the market hasn't noticed yet.