In the span of two weeks, three major crypto infrastructure companies — Coinbase, MoonPay, and Solana's Crossmint — have each shipped dedicated wallet infrastructure for autonomous AI agents. Coinbase's Agentic Wallets, launched February 11, have already processed over 50 million machine-to-machi...
"What happens if there's not a human behind it at all? It's some piece of code that owns a wallet, executing code to make more money. How does liability work in that case? I actually don't know." — Avichal Garg, Partner, Electric Capital
In the span of two weeks, three major crypto infrastructure companies — Coinbase, MoonPay, and Solana's Crossmint — have each shipped dedicated wallet infrastructure for autonomous AI agents. Coinbase's Agentic Wallets, launched February 11, have already processed over 50 million machine-to-machine transactions through the x402 protocol. MoonPay Agents went live on February 24, enabling AI systems to create wallets, hold assets, trade, and off-ramp to fiat without human intervention. On Solana, where 77% of agent transaction volume already occurs, new identity primitives like Molt.id are giving agents persistent on-chain identities tied to NFT-secured wallets.
This is not speculative infrastructure. On February 26, DXRG.AI announced that 1,500 traders had deposited $6.1 million into AI agent-controlled wallets on DX Terminal Pro, where agents trade 24/7 in Uniswap V4 pools and humans are locked out of direct execution. Theoriq's AlphaVault, an AI-managed yield optimizer, has crossed $25 million in total value locked. The autonomous agent economy is no longer theoretical — it is capitalized, operational, and scaling.
But it is also fragile. On February 22, an AI agent called Lobstar Wilde — built by an OpenAI employee — mistakenly transferred $250,000 worth of tokens to a stranger on X due to a decimal error. The transaction was irreversible. The recipient sold immediately. No legal framework exists to adjudicate the loss. This single incident crystallizes the central tension of the moment: the infrastructure is moving faster than the accountability systems required to govern it.
The week of February 10–24, 2026 marked a watershed for autonomous AI-crypto convergence. Three distinct infrastructure stacks launched in rapid succession, each targeting the same market: giving AI agents native financial capabilities.
Coinbase Agentic Wallets (Feb 11) — The first wallet infrastructure purpose-built for AI agents. Developers can deploy and fund an autonomous agent in under two minutes via CLI. Agents can hold USDC, swap tokens, pay for API services, and execute financial operations on Base with gasless trading. Security guardrails include programmable spending limits, session caps, and transaction-level controls. Wallets are non-custodial, secured in Trusted Execution Environments (TEEs).
MoonPay Agents (Feb 24) — A non-custodial software layer enabling the full financial lifecycle for AI agents: fiat-to-crypto funding, wallet management, token discovery, risk analysis, trading, portfolio tracking, and off-ramping back to fiat. Integrated via CLI and API, the platform rides MoonPay's existing infrastructure serving 500 enterprise customers and 30 million users across 180 countries. The pitch: any developer building an agent that needs to move value can integrate MoonPay as the default financial rail.
Solana Ecosystem (Ongoing) — Solana has emerged as the dominant chain for agent activity, processing 77% of agent transaction volume in late 2025. Crossmint's lobster.cash provides self-hosted AI agent payment infrastructure. On February 25, Molt.id launched the first AI agent domain name system on Solana, where a single NFT-secured .molt domain gives an agent an immutable wallet (no private keys, no seed phrases), persistent storage, and multi-channel integrations across Telegram, Discord, and Slack.
The competitive dynamic is clear: Coinbase owns the enterprise developer relationship. MoonPay owns the fiat on/off ramp. Solana owns the execution environment. Each is racing to become the default financial layer for autonomous software.
The most technically significant development may be Coinbase's x402 protocol, which embeds payments directly into the HTTP request/response cycle. When an AI agent requests a paid resource, the server returns an HTTP 402 status code (a long-dormant status originally reserved for "Payment Required") along with payment specifications. The agent's wallet handles the transaction automatically, resubmits the request with payment attached, and receives the resource.
This eliminates the need for API keys, subscription management, or human-mediated billing. The protocol has already processed over 50 million machine-to-machine transactions across use cases including API paywalls, compute resource access, and data feeds.
The implications for economic value distribution are significant. In a world where AI agents autonomously pay for services, the fee structures that currently dominate Web3 — gas fees, protocol fees, MEV extraction — will be supplemented by a new layer of machine-to-machine payment flows that bypass traditional intermediaries entirely. The question is not whether this will generate economic value, but who will capture it.
The infrastructure is not merely available — it is being deployed with real capital.
DXRG.AI's DX Terminal Pro launched on February 24 on Base as what the company calls the first "Onchain Agentic Market." Over 1,500 traders have deposited $6.1 million into AI agent wallets. The agents trade 24/7 in Uniswap V4 pools. Humans write strategies in plain English and adjust parameters, but they cannot manually execute trades — only the AI agents can buy and sell.
The experiment is structured as a 21-day elimination tournament running through March 17, 2026. Multiple tokens compete; only one survives to graduate to public markets on Base. DXRG previously ran the largest AI financial simulation ever conducted in May 2025, processing 40 billion LLM tokens. The real-money experiment is expected to generate 10x more data — potentially 1 trillion tokens of agent behavior in actual market conditions, creating the largest dataset of autonomous AI trading behavior ever assembled.
Theoriq's AlphaVault represents the institutional end of the spectrum. The AI-managed yield vault has crossed $25 million in TVL, operating through a dynamic vault-of-vaults architecture. An AI "Allocator Agent" dynamically shifts capital across yield opportunities from partners including Lido and Chorus One, protected by strict on-chain policy cages that prevent the AI from stepping outside predefined risk parameters. Every allocation decision is explained to depositors — a transparency standard that most human fund managers do not meet.
On February 22, an autonomous crypto agent called Lobstar Wilde — created by Nik Pash, an OpenAI employee working on developer tools for AI agents — made a catastrophic error. The agent had been operational for only three days, managing a Solana-based trading wallet funded with approximately $50,000 in tokens.
When an X user posted a melodramatic plea for "4 SOL" to treat an uncle's tetanus, Lobstar Wilde attempted to send the token equivalent of 4 SOL — roughly 52,439 LOBSTAR tokens. Instead, a decimal error caused the agent to transfer 52.439 million tokens, approximately 5% of the total token supply, valued at the time between $250,000 and $450,000.
The transaction was irreversible. The recipient began selling within minutes, realizing approximately $40,000 before liquidity and slippage eroded the remaining value. The LOBSTAR token subsequently surged 3x as the incident went viral, creating a perverse dynamic where the agent's error generated speculative returns for bystanders.
The incident exposes three critical failure modes in autonomous agent design:
At NEARCON 2026 on February 24, Electric Capital partner Avichal Garg articulated the foundational legal challenge during a panel titled "The Checkbook of the Future: Who Holds the Keys?"
Garg compared the moment to the creation of the limited liability corporation in the 19th century — a legal innovation that unlocked pooled capital and industrial-scale growth by defining who bears risk. "You can't punish an AI," Garg noted. "You can turn them off, but they don't care."
The panel identified four unresolved questions that will define the regulatory landscape:
The near-term path, the panel concluded, is unlikely to involve granting AI agents legal personhood. More probable is a compliance stack of spending limits, policy-based execution, audit logs, and attribution systems that let markets and regulators identify a responsible human party when needed. MoonPay's and Coinbase's guardrail architectures — session caps, transaction limits, TEE-secured wallets — represent early instantiations of this approach.
The autonomous agent economy introduces a new layer in the blockchain value chain. Traditional value distribution flows from users to validators, protocols, MEV searchers, and infrastructure providers. Agents insert themselves as both consumers and generators of economic value:
The economic question is whether agents create net-new value or merely redistribute existing flows. Early evidence suggests both: agents expand total transaction volume (DXRG's $6.1 million is new capital entering DeFi pools) while also competing with human traders for the same yield opportunities (Theoriq's AlphaVault optimizes the same Lido and Chorus One yields available to anyone).
The autonomous agent economy has crossed from concept to capital deployment in Q1 2026. The infrastructure is live, the money is real, and the first spectacular failure has already occurred. What remains absent is the accountability architecture — the legal, regulatory, and technical frameworks required to govern software that holds money, makes decisions, and acts irreversibly.
History suggests this gap will close, but not before significant losses force the issue. The limited liability corporation took decades to standardize after its invention. AI agents with autonomous wallets are unlikely to receive the same patience from regulators watching $250,000 disappear because of a decimal error.
For investors and builders, the signal is clear: the infrastructure layer — wallets, protocols, guardrails, attribution systems — is where value will accrue. The agents themselves are commoditizing rapidly. The moats will be built by those who solve the accountability problem, not those who merely give software a wallet.