AI agent payments could create a new source of stablecoin demand through small automated transactions. Early growth is more likely to appear in payment volume and pre-funded wallet balances than in circulating supply.
An agent can pay for data, computing power or a digital tool while completing a task. A connected wallet approves the purchase only when it fits the owner’s budget, permitted recipients and spending rules.
Stablecoins have an early lead because they settle around the clock and work across borders. USDC dominates the x402 activity analyzed by Keyrock, USDT is lagging behind, while Visa, Mastercard and other payment networks are developing competing options. This market is still up for grabs.
Key Takeaways
x402 lets AI agents pay for digital resources through HTTP.
Keyrock tracked over 176 million x402 payments, with USDC handling 98.6%.
USDT can compete once facilitators make network fees predictable for micropayments.
Stablecoin demand will depend on pre-funded balances, wallet limits and settlement liquidity.
What Are AI Agent Payments And How Do They Work?
AI agent payments are transactions initiated by software within rules set by a person or business. The agent selects a paid resource, sends the request to a wallet and receives access after payment approval.
An agent may need one dataset, one paid tool or a quick burst of compute to finish a task. Buying a monthly plan for that makes no sense. With agentic payments, the software pays on demand and keeps working.
A typical payment involves five steps:
The agent requests a paid resource.
The service returns a price and payment terms.
A wallet checks the amount, recipient and available budget.
The wallet signs and processes the approved transaction.
The service verifies the payment and releases the resource.
x402 handles payment terms inside the web request flow. A service responds with 402 Payment Required, telling the agent what it needs to pay and where to send the funds. The agent then repeats the request with payment proof attached.
Amazon Bedrock AgentCore Payments entered preview in May 2026 with x402 support and session-level wallet controls. Cloudflare’s Monetization Gateway is designed to verify payments before releasing a web page, dataset or MCP tool. The product remains on a waitlist, so there is no public adoption data yet.
How Stablecoins Work For AI Agent Payments
Why Does USDC Lead x402 AI Agent Payments?
USDC leads x402 AI agent payments because Coinbase launched the protocol with an early EVM flow based on EIP-3009, which USDC supports. Low fees on Base and gas sponsorship also reduce the cost of micropayments.
Keyrock’s figures point to high-frequency, low-value activity. The median payment ranges from $0.01 to $0.10, while 76% of transactions cost less than $0.30.
USDT can compete in AI agent payments if facilitators make its network costs predictable enough for micropayments. Tether reported more than $88 billion of USDT on TRON as of May 2026. This liquidity gives agent wallets and merchants a strong reason to support it.
Fees decide whether that reach can translate into micropayments. USDT transfers on TRON consume Bandwidth for transaction data and Energy for smart-contract execution. When the sending account lacks either resource, TRX is burned to cover the shortfall. At a $1 settlement cost, a $0.02 API call becomes 50 times more expensive than the resource itself.
x402 supports multiple assets and networks, although its current documentation does not list TRON. USDT TRC-20 would require a TRON-specific scheme and a compatible facilitator.
Do AI Agent Payments Create Stablecoin Demand?
AI agent payments can increase demand for working balances and settlement liquidity. Transaction volume alone does not represent fresh capital entering stablecoins.
The same $10 of USDC can pay for hundreds of API calls. New demand comes from pre-funded agent wallets, merchant settlement, cross-network liquidity, asset conversion and transaction fees.
How To Secure And Audit Autonomous AI Agent Payments
The wallet draws a hard line around the agent’s spending. If a research task has a $5 budget, that ceiling stays in place even when the agent retries a call, loses context or follows a bad payment request. Before signing, the wallet checks where the money is going and how much will leave the balance.
Google AP2 puts those limits into signed mandates that link each payment to a merchant, spending ceiling and user instruction.
Lobstar Wilde had no effective transaction cap. In February 2026, an X user asked it for 4 SOL. The agent sent 52.4 million LOBSTAR instead, roughly 5% of the token supply. Developer Nik Pash valued the transfer at about $450,000.
A new session had lost context and treated the wallet balance as the amount to send. No transaction cap stopped it. Once signed, the mistake went straight onchain.
Auditing requires more than an onchain transaction hash. A useful record must show:
which agent initiated the transaction;
what the user asked it to do;
which resource it purchased;
which policy approved the payment;
whether the seller delivered the result.
An onchain record proves that funds moved. The surrounding data proves whether the agent had permission to move them.
AI Agent Payment Risks And Spending Controls
Autonomous AI agent payments can fail through repeated requests, misconfigured limits, duplicate charges, altered payment details or network fees that exceed the purchase itself. One bad setting can affect thousands of transactions before anyone notices.
Risk
What Can Happen
Required Control
Runaway request loop
20,000 calls at $0.05 create a $1,000 bill
Request cap and automatic stop
Wrong spending limit
A configuration error gives the agent too much money
Platform-level budget and manual approval
Duplicate payment
A retry pays twice for the same resource
Payment ID and replay protection
Changed price or recipient
A service returns new payment details
Price check and recipient allowlist
Fee exceeds purchase
A $0.02 request triggers a larger network fee
Fee cap, gas sponsorship or pre-funded balance
These controls must run before the wallet signs a transaction. The agent should not be able to raise its budget, edit recipient lists or access private keys. Coinbase Agentic Wallet, for example, applies per-call and per-session spending limits outside the agent’s control.
Why Network Fees Matter For AI Agent Payments
USDC’s lead in x402 is partly structural. Coinbase launched the protocol, while its early EVM payment flow relied on EIP-3009, which USDC supports. Low fees on Base also made cent-level payments practical. Circle issues USDC, while Coinbase provides wallet and facilitator infrastructure around x402.
USDT has broader market reach, but its costs depend on the network. On TRON, transfers consume Energy and Bandwidth. Without those resources, TRX is burned to cover the fee, which may cost more than a small API request.
Energy rental, gas sponsorship, batching and pre-funded balances can reduce that cost. NOW Wallet’s GasFree feature already moves the TRX requirement to the service provider.
TRON is not currently supported by x402, so USDT TRC-20 would need another payment rail or a future facilitator integration.
Where ChangeNOW And NOWPayments Fit Into The Payment Stack
ChangeNOW can provide asset conversion and treasury rebalancing for agent wallets while NOWPayments can connect agents to merchant-facing payment actions through its existing infrastructure and MCP server.
Kate Lifshits, Chief Executive Officer of NOWPayments told ChangeNOW team:
“NOWPayments already has an MCP server. x402 is not currently on our roadmap, but we are considering joining this movement. We see significant potential in enabling both people, through a standard checkout, and AI agents to make payments. This is about machine payments.” – Kate Lifshits, Chief Executive Officer of NOWPayments
NOWPayments has not announced an x402 integration or launch date. Its MCP server still gives agents a standardized interface for accessing payment actions.
For ChangeNOW, the opportunity sits one step before settlement.
A paid endpoint may request USDC on one network while the agent’s treasury holds USDT, ETH or another asset elsewhere. The ChangeNOW API can provide the conversion layer needed to rebalance funds across supported assets and networks.
Trying to swap assets during every paid API request would slow a flow designed for quick settlement. A better setup keeps the agent wallet pre-funded with the asset it expects to spend.
The near-term use case is treasury routing:
ChangeNOW converts and rebalances the assets;
the wallet holds the working balance and controls spending;
ChangeNOW’s clearest role is asset routing and wallet rebalancing, while wallet policies handle authorization. If x402 adoption spreads across more assets and networks, agents will need infrastructure that can source the seller’s preferred stablecoin without adding a manual swap to every purchase.
It is usually a restricted card credential or authorization token. It lets an agent make approved payments without exposing the user’s original card number or full wallet access.
Cards can compete through broader merchant acceptance and chargeback protection. Stablecoins fit machine-to-machine micropayments between services with no previous payment relationship.
The agent uses a tokenized version of a stored card. Merchant rules, spending limits and real-time authorization determine whether the payment can proceed.
Merchants can start with one clearly priced API, dataset or digital tool. The flow needs machine-readable payment terms, verification, request limits and protection against duplicate charges.
Visa Intelligent Commerce provides tokenized card credentials, configurable spending conditions and tools for verifying agents and merchants.