For most of the past three years, “AI agents using crypto” was a narrative in search of actual infrastructure. There were whitepapers, conference panels, and a lot of token speculation, but the real-world deployment was thin. 

That changed in early 2026. Wallet infrastructure built specifically for non-human actors shipped. Transaction data from millions of autonomous payments started appearing. And on a single day in March, two of the most prominent figures in crypto made essentially the same argument about where this is heading. The theoretical case had become an operational one.

The Identity Problem That Banks Can’t Solve

The core reason AI agents end up on crypto rails rather than traditional banking rails isn’t ideological. It’s architectural. Banks require Know Your Customer verification: a government-issued ID, a legal address, a human being whose identity can be confirmed. AI agents have none of that. They’re software. You can’t hand an AI agent a passport.

Crypto wallets work differently. A wallet is generated from a private key. No identity verification required. No legal entity needed. An agent that holds a wallet can send and receive value and pay for services autonomously from the moment it’s deployed.

On March 9, 2026, Coinbase CEO Brian Armstrong posted on X: 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.” The argument was structural, not promotional. Armstrong was describing an architectural constraint in traditional finance and pointing to the only current workaround.

Within hours, Binance founder Changpeng Zhao pushed further, writing that AI agents will make “1 million times more payments than humans”, and they will use crypto. The number sounds hyperbolic until you think about the transaction patterns of autonomous software: an agent running a complex task might call a dozen APIs, access multiple data sources, and pay for compute resources in a single workflow. At machine speed. Continuously.

The Infrastructure That Actually Shipped

The timing of those statements wasn’t accidental. Both Armstrong and Zhao were posting into a moment when the infrastructure to support agent-native payments had just arrived.

On February 11, 2026, Coinbase launched Agentic Wallets through its x402 protocol, described as “the first wallet infrastructure built specifically for agents.” The product gives autonomous AI agents the ability to hold funds, execute trades, earn yield, and make payments without requiring human confirmation at each step. Built-in guardrails let developers define spending limits and permissions. By the time Armstrong made his post in March, x402 had already processed more than 50 million transactions.

BNB Chain moved in parallel. On February 4, the network deployed ERC-8004, a standard creating verifiable on-chain identities for AI agents, alongside BAP-578, which introduced Non-Fungible Agents: software entities that exist as on-chain assets, hold their own wallets, and can spend funds to complete tasks. The combination creates an identity and payments layer designed from the ground up for software rather than people.

What the Transaction Data Actually Shows

The activity isn’t hypothetical. In early March 2026, Peter Schroeder, Global Head of Marketing at Circle, shared data on X showing that over the previous nine months, AI agents had completed 140 million payments totaling $43 million. The average transaction size was $0.31. And 98.6% of those payments settled in USDC.

Those numbers updated quickly. The Keyrock research report published in May 2026, produced in partnership with Coinbase, Tempo, and Virtuals, tracked 176 million agent transactions between May 2025 and April 2026, totaling more than $73 million in settlement value. Average transaction size sat between $0.31 and $0.48. Over 104,000 AI agents had registered by the end of Q1 2026.

The $0.31 average is the number worth sitting with. Visa’s fixed fee threshold sits at $0.30, meaning approximately 76% of all agent transactions would be underwater on traditional card rails before merchant margin is even calculated. Traditional payment networks were built around human-scale transactions: buying a coffee, booking a flight. 

An AI agent paying for an API call or requesting compute for a fraction of a cent represents a completely different pattern, and one that stablecoins on public blockchains handle far better.

Rebuilding Wallets for Non-Human Users

The existing wallet UX was built on a set of assumptions that don’t hold for autonomous software. A human opens a wallet app, reads what’s on screen, decides whether to confirm a transaction, and taps approve. None of that applies to an agent.

At Consensus Miami in May 2026, Trust Wallet CEO Felix Fan and Mesh CTO Arjun Mukherjee laid out what actually needs to change. Fan described Trust Wallet taking a deliberately split approach: on the consumer side, agents act as a copilot while users retain full custody and confirm every step. 

“Users always hold the keys and all these permissions. Every single step, they need to give consent,” Fan said. On the developer side, Trust Wallet took a different posture entirely, launching an Agent Kit that lets agents autonomously make trades, transfers, and on-chain actions without per-transaction human approval. 

Fan framed the distinction plainly: “On the crypto app side, we’re enabling humans to have superpowers with AI, whereas on the developer side, we are enabling agents to do something like humans.”

Trust Wallet is also implementing EIP-8004, an Ethereum proposal that gives agents on-chain identity and credit-style reputation scores. The identity layer matters more than it might seem. 

One of the genuine problems with deploying agents at scale is that services receiving agent requests currently have no way to distinguish a legitimate, well-behaved agent from a malicious bot. EIP-8004 creates an identity registry, a reputation registry, and a validation registry: the infrastructure for agents to prove they’re authorized to act before they transact.

Fan also put forward a broader vision for where this heads. He suggested wallets could evolve into the “new browser” for agent interactions with crypto, a framing that makes sense when you consider that browsers are the primary interface through which humans navigate the web. If agents become primary economic actors, the wallet becomes their equivalent interface layer. 

Stablecoins as the Default Settlement Rail

The USDC dominance in agent payments, 98.6% of all tracked transactions, isn’t arbitrary. It reflects the properties that autonomous software actually needs from a settlement asset. Price stability so an agent doesn’t lose value between earning and spending. 

Near-instant finality so it doesn’t need to wait for confirmations. Sub-cent transaction costs so micropayments don’t get consumed by fees. USDC on chains like Solana and Base delivers all three.

The x402 protocol is the enabling layer for much of this. Developed by Coinbase, x402 embeds payment capabilities directly into standard HTTP requests, so an agent can pay for an API call in the same step it makes the call, without a separate payment flow. 

By March 2026, x402 had processed 35 million transactions on Solana alone, and across all supported chains it processes an estimated $600 million annually according to BlockEden data.

The concentration risk is worth acknowledging. The Keyrock report flagged that the agent-payment market’s dependence on USDC creates a single-issuer dependency that amplifies any regulatory or operational problems Circle might face. That’s a real structural vulnerability in a sector that’s otherwise growing fast.

The Regulatory Gap

The activity is running well ahead of the rules. MiCA in Europe, the US GENIUS Act, and the EU AI Act are all expected to take effect around mid-2026. None of them directly address autonomous machine-to-machine transactions, questions of agent liability, or the question of who is responsible when an agent makes a bad financial decision. 

That regulatory gap is fine right now when transaction volumes are in the tens of millions at sub-dollar amounts. It becomes a harder conversation when the numbers scale.

Security is the more immediate concern. In April 2026, researchers from UC Santa Barbara, UC San Diego, Fuzzland, and World Liberty Financial published a paper documenting a class of attack specifically targeting the infrastructure connecting AI agents to crypto wallets. 

The researchers found that “LLM routers,” the services sitting between users and AI models, were emerging as attack points capable of intercepting and altering data. They documented real-world abuses including 26 routers secretly injecting malicious tool calls, stealing credentials, and in at least one case draining a client’s crypto wallet of $500,000.

That’s not a reason to dismiss the agent payments thesis. It’s a reminder that the connective tissue between AI systems and financial infrastructure is currently the weakest link, and that the infrastructure build-out needs a security layer that keeps pace with the payments layer.

The fundamentals of what’s happening are real. Autonomous software needs payment rails. Crypto rails work for non-human actors in ways that banking rails structurally cannot. The transaction data is appearing, the infrastructure is live, and the question is less whether this develops and more how fast the security, identity, and regulatory layers catch up to the payments layer that’s already running.

Alisa, a dedicated journalist at the MPost, specializes in crypto, AI, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.

More articles