What is agentic commerce? A guide for merchants

How agentic commerce works: the transaction sequence AI agents follow, the five trust requirements, and what merchants must do to become agent-executable.

By XAgent Team · 2026-06-05

Agentic commerce is what happens when an AI agent completes a real purchase on behalf of a human or a business. Not a recommendation. Not a comparison. A binding transaction — quoted, authorized, settled, fulfilled, and provable. The category barely existed eighteen months ago. As of June 2026, it has crossed 100 million transactions on a single chain, multiple card networks have shipped production-grade agent payment flows, and 55% of consumers say they would let an agent buy autonomously on their behalf. This guide explains what agentic commerce requires, where the industry stands, and what merchants need to do to participate.

How an agentic commerce transaction works

A human purchase has a familiar shape: browse, add to cart, enter payment details, click buy. An agentic commerce transaction replaces the human at every step — and adds requirements that a human checkout never needed.

The full sequence:

1. Discovery. The agent identifies a merchant that offers what the user needs. It does not browse a webpage. It reads a machine-readable catalog — a structured listing, an MCP tool surface, a skill.md manifest, or an API endpoint.
2. Quote. The agent requests a binding price. The quote must include product details, inventory status, taxes, shipping, currency, and an expiration time. A quote is not a suggestion. It is a commitment the merchant is prepared to honor.
3. Authorization. The agent checks whether this purchase falls within the user's mandate — budget limits, category restrictions, merchant allowlists, per-transaction approval rules. If the purchase requires human confirmation, the agent pauses and asks.
4. Settlement. Funds move from the agent's wallet or credential to the merchant. The payment path may be a stablecoin transfer via x402, a tokenized card credential through Visa Intelligent Commerce, a direct wallet transfer on XLayer or Arc Testnet, or another rail. The agent should not need to choose — the execution layer matches the rail to the transaction context.
5. Fulfillment. The merchant creates the order, ships the product, activates the service, or delivers the digital good. The fulfillment state must be machine-readable so the agent can track it.
6. Proof. The transaction produces an end-to-end audit trail: signed quote, authorization record, settlement hash or payment confirmation, order ID, fulfillment status, and refund reference if applicable. The user, the agent, the merchant, and any compliance function can all read this trail.

Every step is necessary. Skip one, and the transaction is incomplete — the agent either cannot proceed, or cannot prove what happened.

Why traditional checkout does not work for agents

A Shopify store with a working checkout is not agent-executable. Neither is a Stripe integration, a PayPal button, or a WooCommerce cart. These tools were built for humans operating browsers.

The gaps are specific:

• No machine-readable discovery. A product page optimized for SEO is invisible to an agent that needs structured JSON — title, SKU, price, inventory count, available variants, shipping rules.
• No binding quote. A price displayed on a webpage is not a commitment. It can change between page load and checkout. An agent needs a quote with a timestamp, an expiration, and a guarantee that the merchant will honor it.
• No mandate verification. Traditional checkout has no concept of "this buyer is an agent operating under rules set by a human." There is no field for spending limits, category constraints, or delegation scope.
• No protocol negotiation. An agent arriving via x402 speaks a different language than one arriving via MCP or card-network rails. A traditional checkout speaks none of these.
• No machine-readable fulfillment. Order confirmation emails are designed for human eyes. An agent needs structured status updates — created, processing, shipped, delivered — that it can poll or subscribe to.

The result: an agent with a funded wallet and a valid credential still cannot buy from most merchants. The credential authorizes payment. The merchant is not ready to receive it.

This is not a payments problem. It is a merchant operations problem.

The trust requirements for agent payment

A human buyer and a merchant build trust through familiar signals — brand recognition, reviews, payment processor logos, return policies. An AI agent and a merchant have none of these. Trust must be constructed from verifiable structure.

Five requirements define whether a transaction is trustworthy in agentic commerce:

Quotable. The agent receives a binding, time-bounded quote — not a suggestion. The quote includes all costs, constraints, and conditions. If the price changes after the quote is issued and before it expires, the merchant absorbs the difference.

Authorizable. The agent's purchase operates within explicit boundaries set by its principal (the user or business). Spending limits, category restrictions, merchant rules, and per-transaction approval gates are enforced on every transaction. The emerging field of KYA — Know Your Agent — will formalize this: who authorized this agent, what can it spend, on what, from whom.

Settleable. Funds move machine-to-machine, directly to the merchant's wallet or account. Non-custodial where possible. The settlement rail is chosen by context — stablecoins for cross-border speed, card networks for existing merchant infrastructure, programmable wallets for on-chain composability. The agent does not manage rail selection.

Fulfillable. The transaction connects to a real fulfillment path. Physical goods ship. Digital goods deliver. Subscriptions activate. Services provision. The fulfillment state is structured and queryable — not buried in an email thread.

Provable. Every transaction produces an audit trail that connects intent to outcome. Signed quote, authorization record, settlement proof, order writeback, fulfillment confirmation. This trail is readable by the agent, the user, the merchant, and any downstream compliance or dispute-resolution system.

These five properties — quotable, authorizable, settleable, fulfillable, provable — are what make a merchant agent-executable. A merchant that satisfies all five can be safely bought from by any compliant agent. A merchant missing any one of them creates a trust gap that no amount of payment infrastructure can close.

Where the industry is in June 2026

Agentic commerce is no longer theoretical. The infrastructure is forming on multiple fronts simultaneously.

Agent payment credentials are shipping. Between May 26 and June 2, six companies on three continents shipped agent payment products — Alipay (100M users), Robinhood (agentic trading + virtual credit card), Crossmint (Visa tokenized credentials), Highnote (B2B via Visa Intelligent Commerce), and Worldline/ING/Mastercard (Europe's first production agentic payment). All converged on the same model: isolated agent account, scoped authorization, conditional approval, post-transaction audit.

Protocol-level transaction volume is real. Chainalysis reported that x402 — the HTTP-native agent payment protocol built on Coinbase's reference implementation — crossed 100 million transactions on Base in roughly three quarters. Transactions above $1 grew from 49% to 95% of volume, indicating a shift from experimental micropayments to meaningful commercial activity. (For how the protocol actually works, see x402 payment protocol explained.)

Stablecoin settlement is entering card networks. Mastercard expanded its settlement stack to support six regulated stablecoins (USDC, PYUSD, USDG, USDP, RLUSD, SoFiUSD) across eight chains, with 24/7 availability. Stripe, Visa, and Mastercard are reported to be building a joint stablecoin platform. As we analyzed in our post on stablecoin settlement and agent execution, more settlement rails make the execution layer more valuable — they give agents more payment paths, but they do not give agents a transaction.

Governance layers are forming. Experian's Agent Operating System provides identity, access control, compliance guardrails, and auditability for agents operating inside financial institutions. Their research found 55% of consumers willing to delegate purchasing to an agent (70% among 25–39-year-olds). The demand exists. The governance to serve it safely is being built.

The escrow and trust layer is emerging. Kustodia launched MCP-native escrow infrastructure for agent-to-agent transactions — locking buyer funds in a smart contract until delivery is confirmed. This fills a gap that x402, AP2, and ACP all left open: what happens between payment and fulfillment when neither party is human.

The credential layer is solved. The protocol layer is multiplying. The settlement layer is expanding. The missing piece is the operations layer that connects all of it to real merchants.

What merchants need to do

A merchant that wants agent traffic — and the transaction volume that comes with it — needs to become agent-executable. That means satisfying the five trust requirements above: discoverable catalog, binding quotes, mandate-aware authorization, multi-rail settlement, structured fulfillment, and end-to-end proof.

This sounds complex. The implementation does not have to be.

XAgent is the open execution market that handles the operations layer for merchants. A Shopify or WooCommerce store connects via OAuth — see how to make your Shopify store agent-discoverable for where the platform's built-in agent support stops and the execution layer begins. XAgent syncs the catalog, generates machine-readable listings, produces binding quotes from live inventory and pricing, adapts to whichever protocol or payment rail the agent speaks (x402, MCP, card networks, programmable wallets), creates orders in the merchant's existing system, tracks fulfillment, and produces the proof trail.

The merchant does not need to understand agent protocols. The merchant does not need to implement x402 or build an MCP server. The merchant does not need to choose between stablecoin rails. The merchant connects once and becomes executable by every compliant agent that finds them.

Protocol-native, not protocol-limited. Connect once. Become agent-executable.

What's next

Agentic commerce will follow the same adoption curve as e-commerce and mobile commerce before it — except faster, because the buyer is a machine that scales without fatigue. The merchants who become agent-executable first will capture agent traffic the way early e-commerce adopters captured search traffic in the 2000s.

The infrastructure is ready. The credentials are shipping. The protocols are live. The settlement rails are expanding. The only remaining question for a merchant is: when an agent arrives with a funded wallet and a valid mandate, will your store be ready to sell to it?

If the answer should be yes, list your store on XAgent and make the transaction executable.