Learn what agentic commerce is, how AI agents discover and buy services, and why programmable stablecoins and spending controls help manage purchasing risk.
Learn what agentic commerce is, how AI agents complete purchases, how agentic checkout differs from traditional ecommerce, and why programmable stablecoins, spending controls, and reliable delivery matter.
AI agents are increasingly asked to complete tasks such as testing software or booking travel. Agentic commerce lets them also pay for what those tasks require: discovering, buying, and using goods or services on behalf of people and businesses, within limits their operators set.
Consider a coding agent testing a new application. Partway through, it needs a browser service it cannot run locally. With permission to spend, it checks the price, pays for a session, runs the tests, and returns the results to its operator with a record of the expense.
This post takes a closer look at agentic commerce: how it differs from agentic payments, how agent-led purchases work, why agents need a different kind of checkout, why programmable stablecoins suit that checkout, and which controls make the whole process trustworthy.
AI agents as buyers: What agentic commerce covers
Agentic commerce is the exchange of goods or services initiated or executed by AI agents on behalf of people or businesses. An AI agent is software that uses models and tools to pursue a goal. It can discover an offering, evaluate its suitability and price, and complete a purchase within delegated permissions. Depending on those permissions, it may act independently or seek human approval before committing funds.
Agentic commerce vs agentic payments
Agentic commerce is the transactional layer of the broader agentic economy, which also includes activities such as managing funds and coordinating agreements. It is often discussed alongside agentic payments, but the two terms cover different scopes. “Agentic payments” describes the movement of money within a transaction. Agentic commerce covers the purchase as a whole: what is being bought, why it is useful, how it is paid for, and whether it is delivered.
How agentic commerce works: A step-by-step example
To see how the pieces fit together, consider an agent authorized to manage a company’s travel bookings with a capped budget. A purchase within that workflow could look something like this:
- Receive the task: The operator defines the destination, travel dates, spending limit, permitted booking providers, and transactions requiring approval.
- Discover a paid service: The agent identifies a flight or hotel provider offering options that meet the trip requirements, with pricing and terms it can read.
- Evaluate cost and value: It compares available options, checks whether the cancellation and usage terms fit the task, and confirms that the total cost stays within budget.
- Execute payment: The payment system verifies the agent’s authority and enforces spending limits before submitting the transaction.
- Receive the service: The provider delivers booking confirmations and itinerary details. The workflow checks delivery and records the payment confirmation.
- Continue the work: The agent adds the itinerary to the travel plan, shares the relevant details, and connects the expense to the completed booking.
This agent-led sequence makes procurement responsive to the task. A business can authorize access to approved resources without selecting every purchase in advance. The operator retains control over the objective and financial boundaries.
Each step above assumes the agent can read prices, authorize payment, and confirm delivery without a person clicking through screens. That assumption is where agentic checkout departs from the online shopping experience most people know.
Agentic checkout vs traditional ecommerce checkout
Human ecommerce typically involves a login, product page, shopping cart, and checkout form. Those interfaces help people browse, compare, and confirm decisions. For an agent buying a single response from a paid data service, navigating them adds unnecessary steps.
A machine-oriented purchase relies on structured descriptions, explicit prices, digital authorization, and receipts that software can process. In an x402 flow, for example, a service returns an HTTP 402 payment request; the agent signs the payment and returns proof of it with the request, and the service verifies that payment before granting access.
The commercial questions remain familiar: Who authorized the purchase? What does it cost? What will be delivered? Agentic checkout makes those answers available inside the workflow, so an agent can pay for a resource without stopping to wait for human approval when its delegated permissions allow.
A checkout built for software also needs a payment method that works the way software does: available around the clock, governed by programmable rules, and clear about when a payment is complete. That is where stablecoins come in.
Why stablecoins are well suited to agentic commerce payments
Stablecoins are digital assets designed to track a reference value, such as a fiat currency like the US dollar. Combined with suitable blockchain networks, wallets, and smart contracts, they offer several capabilities that map to the purchase steps described above.
24/7 settlement
Blockchains let value move outside banking hours. On suitable networks, 24/7 transfers can settle in seconds or less, allowing a provider to confirm payment and grant access promptly. Actual settlement speeds depend on the network. Stablecoin-to-fiat conversions and payouts to bank accounts can add further steps.
Programmable payment logic
Programmable rules can connect a payment to defined conditions. A contract can hold funds until delivery is verified or divide a payment among service providers, turning a purchasing agreement into rules software can execute. A contract can't see whether a service was delivered, though; it acts on whatever confirmation it receives, so that confirmation needs to be trustworthy.
Payment finality and dispute resolution
Sellers need clarity about completed transfers. Direct, onchain payments generally do not have card-style chargebacks. However, refunds and dispute resolution may still be desirable in some instances. Some stablecoin payment providers and solutions support refunds and dispute resolution mechanisms.
Software-based authorization
Authorized agents can authenticate with credentials and use preconfigured permissions, reducing repeated manual interactions. Applicable identity verification, business verification, and sanctions requirements still apply to the relevant participants and services. An agent acts under delegated authority; automation does not remove those obligations.
Nanopayments
Agent purchases can be small, such as a single response from a paid data service. Stablecoins can support small, frequent purchases where network and service fees make the economics viable.
How well stablecoins deliver these capabilities depends on their acceptance, liquidity, issuer reliability, crosschain support, and the cost of money movement.
Agentic commerce examples: AWS and Mastercard
Public implementations show how these purchases work today. They take different approaches, one built into a cloud agent platform and one extending an established card network, which shows that agentic commerce can run on more than one payment method.
Amazon Bedrock AgentCore Payments supports agent access to paid APIs, and content, with configurable spending limits and visibility into transactions. It brings purchasing controls into the environment where an agent already performs its work.
Mastercard Agent Pay addresses purchases through established merchant networks. Its acceptance framework verifies agents and uses tokenized credentials and purchase-intent information to help merchants recognize authorized transactions.
Though their approaches differ, both follow the same pattern as the coding agent at the start of this post: acquire what a task needs, pay for it within set limits, then continue working. Whether the purchase is a browser session, a dataset, or a hotel booking, it happens inside the workflow rather than at a separate checkout.
Building trust in agentic commerce: Controls and safeguards
Amazon Bedrock AgentCore Payments and Mastercard Agent Pay both pair purchasing capability with explicit controls: spending limits in the first, agent verification in the second. These pairings reflect a broader principle: reliable execution requires controls outside the model’s reasoning. Spending limits, approved suppliers, expiring permissions, and escalation rules define what an agent can commit to. Receipts should connect each purchase to its authorization, cost, and delivery outcome.
Failures also need explicit handling. A timeout should trigger a status check before another payment. An unusable result needs a refund or dispute path. These details determine whether automated purchasing saves effort or creates reconciliation work.
As these capabilities mature, businesses can make specialized services available at the moment a task needs them. Agentic commerce makes that opportunity concrete: software that can find, purchase, and use resources under clear human authority, with an accountable record of the exchange.
Circle Nanopayments are offered by Circle Technology Services, LLC (“CTS”). CTS is a software provider and does not provide financial, advisory, or marketplace services. Circle Nanopayments enable developers and users to initiate and receive transactions programmatically, including through automated workflows or artificial intelligence-enabled applications. Transactions may occur without real-time human review and are executed based on user-defined configurations and permissions. Users are solely responsible for their use of these tools, reviewing and validating transaction behavior, and evaluating any associated security, financial, or operational risks. For additional details, please see the Circle Developer Terms of Service available at console.circle.com/legal/developer-terms.
