Agentic Commerce Explained

Agentic commerce is commerce where AI agents act on a person's or business's behalf to discover, decide, and pay for goods and services, with the human setting the goals and limits rather than clicking every button. Instead of a person browsing and checking out, an agent does the work and the human stays in control through policy, approvals, and budgets; Amazon Buy for Me is the mainstream example. The hard part is not the shopping. It is paying safely, which is why a layer of payment infrastructure and standards has grown up around it.

This primer defines agentic commerce, explains the standards you keep hearing about, and shows where payment infrastructure fits.

What is agentic commerce?

Traditional e-commerce assumes a human at the keyboard. Agentic commerce removes that assumption. An agent can search across merchants, compare options, fill a cart, and complete payment, all from a goal like "reorder our supplies under 200 USD" or "book the cheapest direct flight that fits my calendar." The human's role shifts from operator to principal: you say what you want and what the limits are, and the agent executes within them.

That shift creates a trust problem. Merchants need to know a payment from an agent is authorized and will not be charged back as fraud. Buyers need to know the agent will not overspend or pay the wrong party. Both needs point to the same answer: verifiable authorization and clear accountability. That is the gap the new standards and the payment infrastructure try to close.

The standards you should know

Four efforts come up most often. They solve different layers of the same problem, and they are largely complementary rather than competing.

AP2: proving authorization

The Agent Payments Protocol focuses on the question every party in agentic commerce wants answered: did a human actually authorize this? AP2 uses verifiable mandates, signed, tamper-evident records that capture what the user authorized. A mandate can express intent ("you may buy a flight under 400 USD this week") or a specific cart ("buy exactly these items at this price"). Because the mandate is cryptographically verifiable, a merchant or payment provider can trust that the agent is acting within real authority. We go deeper in our guide to agent payment mandates.

ACP: standardizing checkout

The Agentic Commerce Protocol, from OpenAI and Stripe, standardizes the checkout step so an agent can complete a purchase at any participating merchant without bespoke integration. Think of it as a common language for "here is the cart, here is the payment, complete the order." It makes the buying half of agentic commerce work at scale, and it is what powers Instant Checkout inside ChatGPT.

x402: paying per call

x402 revives the long-dormant HTTP 402 "Payment Required" status to let agents pay tiny amounts per request, typically in stablecoins, for APIs, data, and services. This matters because agents consume machine resources in ways humans do not, often many small calls, and traditional card fees make sub-dollar payments uneconomic. x402 makes pay-per-call practical for agent-to-service commerce.

MPP and the merchant side

On the merchant side, presentment patterns describe how a store recognizes an agent buyer, presents an offer it can act on, and accepts the resulting payment. This is less a single spec and more a set of conventions emerging as merchants adapt to agent traffic.

Where payment infrastructure fits

Standards define how authorization and checkout are expressed. They do not, by themselves, fund an agent, enforce your spend limits, gate large purchases, or keep a defensible record. That is the job of payment infrastructure, and if you are the one building the agent that checks out it sits in a clear place in the stack:

1. Identity. Each agent has an identity tied to a named human owner. See Know Your Agent.
2. Authorization. Mandates (AP2-style) capture what the human allowed.
3. Funding. A capped agent wallet holds the budget the agent can draw on.
4. Policy and controls. Spend controls and approvals enforce limits and gate big spend before money moves.
5. Settlement. The actual rail, cards, bank, or crypto, moves the money.
6. Record. An audit trail ties every payment back to intent, agent, owner, and policy.

A rail-neutral control plane like Agentspay spans the funding, policy, and record layers and works across whichever settlement rail and checkout protocol a transaction uses. The principle is the same regardless of standard: never move money without policy.

Common questions

Is agentic commerce only for crypto?

No. Most agentic commerce settles on cards today, with crypto rails like x402 well suited to micropayments and agent-to-service calls. A good control plane is rail-neutral so you are not forced to choose.

Do these standards compete?

Mostly they complement each other. AP2 handles authorization, ACP handles checkout, x402 handles micropayments. A real agent transaction might touch more than one. The open question for the industry is how cleanly they interoperate over time.

What does a business need to start?

Identity for your agents, a funded and capped wallet, spend controls and approvals, a settlement rail, and an audit trail. Start in a sandbox and run the full lifecycle before any real money moves. See how it works.

The takeaway

Agentic commerce is agents buying and paying within limits a human sets. The standards, AP2 for mandates, ACP for checkout, x402 for micropayments, are making it interoperable. But standards describe authorization and checkout; they do not enforce your budget or keep your record. That is payment infrastructure, and it is where the trust actually lives. Next, read how payment mandates authorize agent spend or compare agent payment platforms. If you are unclear where the commerce shift ends and the money movement begins, agentic payments explained draws the line.