Key Highlights
- Coinbase Chief Policy Officer Faryar Shirzad said AI agents are changing how online content and services are accessed, creating a need for machine-to-machine payments.
- A new Coinbase Institute paper argues that traditional payment rails are poorly suited to frequent transactions worth fractions of a cent.
- The paper identifies stablecoins on low-cost networks as a potential payment rail for high-frequency, low-value agent transactions.
Coinbase Chief Policy Officer Faryar Shirzad said the rise of AI agents could create a new payment layer for the internet as software increasingly retrieves content and services on behalf of users.
In an October 7 post on X, Shirzad highlighted a new Coinbase Institute paper titled “Machine-to-machine payments in the AiFi era.”
The paper examines how AI agents could pay for data, content and digital services on a per-use basis and why existing payment systems may not be efficient for very small transactions.
AI agents create a new revenue gap for publishers
The Coinbase Institute paper describes AI agents as a “second customer” for the internet.
Unlike human users, agents do not watch advertisements, browse websites in the same way or necessarily create accounts with every service they use. They can instead retrieve information, compare services and execute tasks on behalf of a person.
The paper argues that this creates a potential revenue gap for publishers, developers and data providers that have traditionally relied on advertising, subscriptions or human-driven traffic.
If an AI system consumes content or calls an API without generating a human visit, the paper says providers need another way to charge for that use.
One proposed model is pay-per-use, where an agent could pay for an individual API request, document, data query or other service rather than creating an account or entering into a longer-term billing relationship.
“That raises a real question for creators, publishers and developers: how can they get paid when an agent uses their work, sometimes for a fraction of a cent?” Shirzad said.
Traditional payment rails struggle with micropayments
The Coinbase Institute paper argues that cards, bank transfers and other conventional payment systems were designed around larger transactions and established relationships between buyers and sellers.
A major issue is the fixed cost attached to individual payments.
The paper notes that fixed payment-processing fees can make traditional payment rails impractical for very small transactions. A $0.30 fee, for example, would represent a 30,000% transaction cost on a $0.001 API request.
The paper estimates that the minimum viable transaction for conventional payment methods can range from several dollars to more than $10 when transaction costs are kept below 5% of the payment value.
The cost problem becomes more relevant when an AI agent makes hundreds or thousands of requests during a single task.
Stablecoins proposed as a payment rail for agents
The paper identifies stablecoins on low-cost networks as a potential rail for high-frequency, low-value machine-to-machine payments.
It points to characteristics including low network costs, rapid settlement, programmability and the ability to send payments across borders.
Shirzad said the paper examines “why existing payment rails often struggle with high-velocity microtransactions—and why stablecoins on low-cost networks are a strong fit today.”
The paper does not propose replacing every existing payment method.
Instead, it distinguishes between larger purchases involving established counterparties and the smaller, more frequent transactions it expects AI agents to make.
For example, a conventional card could remain suitable for an annual software subscription, while a stablecoin payment could be used for a single data request worth a fraction of a cent.
Coinbase’s x402 uses an HTTP payment request
The paper uses x402, a payment standard developed by Coinbase, as an example of how machine-to-machine payments could work.
The flow begins when an AI agent requests a resource from a server.
If payment is required, the server returns an HTTP 402 Payment Required response containing machine-readable information about the price, payment asset, recipient and supported network.
The agent can then authorize the payment and send the request again with payment information attached.
A facilitator can verify the payment authorization and settle the transaction onchain, after which the server provides the requested resource.
The example included in the paper uses USDC on Base and shows a payment of 0.01 USDC for access to market data. The agent receives the payment request, signs an EIP-3009 transferWithAuthorization transaction and returns the payment proof to the server.
The facilitator then verifies and settles the payment on Base before the server returns the requested data.
The example describes the payment as settling in about two seconds, with a network fee of less than $0.001. The example illustrates a payment flow that can operate without a separate account or traditional card-based checkout.
x402 does not require one stablecoin or network
The paper distinguishes the x402 payment protocol from the assets and networks used to settle payments.
x402 uses the HTTP 402 status code to communicate payment requirements, but it does not require a particular stablecoin, blockchain or payment provider.
A server can specify which payment schemes it accepts, while facilitators can verify and settle supported payments.
According to the paper, the protocol can accommodate different payment providers and settlement networks.
Publishers could charge AI agents for access
The paper proposes per-use payments as one way for publishers to monetize automated access by AI agents.
A publisher could set a price for an article, database or API and allow an agent to pay automatically when it requests the resource.
Payment is only one part of agentic commerce
The Coinbase Institute paper also identifies agent identity and authorization as issues that need to be addressed.
An AI agent may act on behalf of a person or company, making it important for a service to determine who authorized a transaction and what the agent is allowed to spend.
The paper argues that this could be handled through cryptographic credentials and spending limits rather than requiring the agent to create a separate human-style account with every service.
For example, an agent could have a predefined spending limit or be restricted to certain types of services.
The paper also calls for portable identity credentials so an agent does not have to repeat the same verification process with every provider.
Coinbase Institute calls for open payment standards
The policy recommendations in the paper focus on keeping machine payments open and interoperable.
The authors argue that policymakers should preserve the economics of very small payments, build compliance around identity and authorization, and avoid rules that force AI-agent payments through closed marketplaces.
Shirzad made a similar point in his X post:
“The policy goal should be an open, interoperable payment layer, with safeguards that match the risks—not closed marketplaces that decide who can participate.”
Coinbase has previously linked stablecoins to AI payments
The latest paper follows earlier Coinbase initiatives involving stablecoin payments for businesses and AI agents.
On September 28, Coinbase expanded its partnership with Citi to allow corporate customers to accept stablecoin payments. Under the arrangement, Coinbase would convert incoming stablecoins into dollars, while Citi would remain the bank of record for corporate merchant payments.
Coinbase CEO Brian Armstrong had also discussed the potential role of stablecoins in AI payments in August.
In an August 21 report, The Crypto Times covered Armstrong’s view that AI agents could eventually make cross-border payments for goods and services, areas where he said blockchain-based stablecoins could be useful.
AI payments remain an emerging use case
The paper comes as companies across AI and payments continue to test machine-to-machine transaction systems.
It points to experiments involving companies such as Cloudflare and AWS, while developers have tested x402 for API and agent-based services.
The infrastructure remains at an early stage. Wider adoption will depend on whether publishers, developers, payment providers and AI platforms adopt compatible payment and identity standards.
Also Read: Ethereum Researcher Drake Flags ECDSA Risk for Crypto Holders
