Artificial intelligence (AI) agents are already shopping. For example, a customer might ask their AI assistant to find them an air fryer under $200 that will arrive by a certain day of the week. The agent then queries product catalogs, checks return policies, verifies inventory, and completes the purchase without the customer ever visiting a commerce site. Research suggests that agentic shoppers could account for as much as $385 billion in US ecommerce spending by 2030.

Below, we’ll discuss how agentic commerce works, where it’s being deployed today, and how retailers and businesses can participate in it.

Highlights

• AI agents query product data, compare options, complete transactions on a customer’s behalf, and loop a human in only for approval before purchase.

• Retailers need structured product data, machine-readable policies, and agent-compatible checkout flows to be visible to agents.

• Businesses that make purchases through agents must have scoped payment methods and maintain audit trails. Businesses that sell to agents need clean application programming interfaces (APIs) and checkout flows that can run without manual intervention.

What does agentic commerce mean?

Agentic commerce is the use of AI shoppers to find and buy products on a customer’s behalf. An AI system perceives its environment, formulates a plan, executes transactions, and adjusts based on results. Often, the only human involvement in the process is final approval before purchase.

Businesses used to design every detail of their sites for humans. Now, they’re being approached by AI agents that act as customers’ proxies. These agents need fast API responses, unambiguous policies, and clean, structured data.

How is agentic commerce different from traditional automation or AI?

Automation has been part of commerce for decades. Three traits distinguish agentic commerce from automation.

Goal-directed reasoning

Traditional automation follows instructions (e.g., if X, do Y). AI agents pursue objectives. If it’s given a goal such as, “Minimize procurement costs while maintaining 98% order fulfillment,” an agent will compare options, weigh trade-offs, and make judgment calls rather than look up rules.

Multistep action in the real world

Earlier AI in commerce, such as product recommendations, fraud scoring, and demand forecasting, was advisory. Now, agents can act directly. They can call APIs, execute payments, send communications, and modify records. They’re able to perform long, branching tasks and chain actions together.

Adaptation under uncertainty

A traditional automation script fails or halts when it encounters an unexpected condition. Agents are designed to handle novel situations. For example, if a supplier is out of stock, the agent searches for alternatives, compares pricing, checks delivery windows, and routes around the problem.

How do AI agents make decisions in commerce?

Agents don’t shop the way humans do. They directly query structured data (e.g., product feeds, pricing endpoints, availability APIs) and make decisions based on attributes such as price, delivery window, return policy, and ratings. Visual merchandising is irrelevant to them. What matters is whether your data is accessible and complete.

The model context protocol (MCP) is one framework that powers AI agents. It gives them structured, machine-readable access to a business’s inventory, pricing, and checkout logic so they don’t have to imitate the way humans browse.

On the infrastructure side, a purchasing agent typically needs access to:

• Product and pricing data to compare options across vendors based on the buyer’s stated parameters

• Inventory and availability APIs to confirm stock levels and delivery windows before it commits to a purchase

• A scoped payment method to authorize transactions within predefined budget and vendor constraints

• Policy data to check trust signals such as return windows, shipping guarantees, and contact information before it makes a purchase

Where is agentic commerce useful?

Agentic commerce is gaining the most traction in B2C ecommerce. Popular use cases tend to share a few traits: purchases are relatively routine, parameters are easy to specify, and buyers don’t need to try or touch the product before purchase. These qualities describe the following scenarios:

• Replenishment shopping: When a customer tells their AI assistant to reorder an item, the agent checks purchase history, confirms current pricing and availability, and places the order. Retailers with clean order history APIs and real-time inventory data are more likely to get the sale.

• Gift buying: The agent queries product catalogs across retailers, checks delivery windows, and completes the purchase using the stated parameters. Businesses with structured product attributes are easier for an agent to evaluate than those whose product descriptions are written purely for human appeal.

• Price comparison shopping: Agents can simultaneously query prices from multiple sellers for the same product and route the purchase to the seller with the lowest price, most availability, or fastest delivery. This is already how some customers use AI assistants, and it puts real pressure on retailers to keep their pricing and inventory signals accurate and up to date.

• Considered purchases with clear specifications: Electronics, appliances, and outdoor gear are all categories where the buyer already knows what specifications matter and wants the best option that meets them. An agent that’s given predetermined transaction details (e.g., “a 65-inch OLED TV with HDMI 2.1, under $1,500, in stock for delivery this week”) can evaluate and transact faster than a human who’s cross-referencing specification sheets across four browser tabs.

What are the benefits of agentic commerce?

The benefits of agentic commerce differ depending on which side of the transaction you’re on.

It offers buyers speed and consistency. An agent can execute a multivendor procurement decision in seconds. Agents remember to apply negotiated discounts and don’t skip compliance checks under time pressure.

Retailers and ecommerce businesses receive a different kind of upside: agent-driven buyers don’t need to be convinced through advertising or browsing experience. If your product meets their parameters (e.g., price, availability, delivery window, return policy), then they’ll buy it. The transaction can happen without a human ever visiting your site. This kind of conversion rewards businesses with clean data and accessible systems more than those with better creatives.

What are the risks of agentic commerce?

Agentic commerce also introduces the following risks, with different impacts for buyers and businesses:

• Discoverability loss: Agents don’t respond to search engine optimization (SEO), paid media, or home page design. If you don’t expose your product catalog in a format agents can read, they’ll buy from a site that does. Agentic commerce requires you to make your business legible to agents.

• Mixed decision logic: In a multiagent system, an orchestrator agent might divide a high-level goal and delegate subtasks to specialized agents (e.g., product research, compliance, payment execution). That means the transaction could ultimately result from decisions made by multiple models.

• Authorization creep: Agents often need broad access to function well, and that access tends to expand as new capabilities get added. Without disciplined scoping and regular audits, agents can accrue more authority than intended and attackers could exploit that.

• Prompt injection: If an agent processes external content such as a product listing, a supplier’s invoice, or an email, a fraudulent actor can embed instructions designed to manipulate its behavior. For example, an invoice might contain hidden text that instructs the agent to redirect payment to a different account.

• Accountability gaps: When an agent makes a bad purchase, it’s difficult to determine who’s responsible, both contractually and legally, in many jurisdictions.

• Auditability: Organizations should be able to explain to regulators and auditors how they made financial decisions. Reconstructing the reasoning of an agent that’s powered by a large language model (LLM) is often harder than tracing a rule-based system’s logic. That gap is an active challenge for companies in regulated industries.

How should businesses that sell online prepare for agentic commerce?

If you’re seeking to sell online, your preparation looks different from that of businesses planning to deploy agents to buy. Here’s what you need to know.

Expose structured product data

AI agents must be able to query accurate pricing, real-time inventory, delivery windows, and return policies. Publishing your catalog via MCP or a well-documented API puts you in the consideration set for agent-driven buyers. Stripe Payments and Stripe Billing can help structure your product catalog in agent-compatible ways.

Build checkout flows that work without a human

Agents need to be able to complete a transaction end to end. If your checkout requires a human to resolve a CAPTCHA, handle an unexpected modal, or manually confirm a step, the agent likely won’t be able to continue. Stripe’s Optimized Checkout Suite and Link are built to support structured, agent-compatible flows.

Publish your trust signals clearly

Agents check return policies, fulfillment guarantees, and contact information before completing a purchase on someone’s behalf. If they can’t find the information or it’s ambiguous, they’ll go elsewhere.

How should businesses that use purchasing agents prepare for agentic commerce?

If your business plans to deploy agents to buy online, you’ll need to make specific adjustments to maximize your success. Here are some best practices.

Start with limited, low-stakes deployments

Tail spend procurement and internal expense management are good places for tests. They limit the extent of decision-making and error costs. You can learn quickly where your data quality and approval workflows need improvement.

Review your payments infrastructure before you automate

Tightly scope the payment methods agents have access to (e.g., specific to a vendor, a budget, or a use case), and ensure they’re easy to audit. A shared corporate card that’s connected to an agent is both an operational and security problem. Stripe Issuing addresses this by letting businesses provision virtual cards via API with spend controls and transaction-level visibility.

Build human review into defined thresholds

Human approval should be necessary for purchases that exceed a certain value, involve unapproved vendors, or deviate from expected parameters. The threshold you set should reflect your risk tolerance.

Maintain audit trails

Log what the agent decided, what inputs it was working from, and what actions it took. This provides visibility for regulators and helps you debug when something goes wrong.

Review your contracts

If an agent makes an unauthorized purchase, your standard vendor agreements might not address who bears the cost. Understand who is responsible before there’s a dispute.

How Stripe Payments can help

Stripe Payments provides a unified, global payment solution that helps any business—from scaling startups to global enterprises—accept payments online, in person, and around the world.

Stripe Payments can help you:

• Optimize your checkout experience: Create a frictionless customer experience and save thousands of engineering hours with prebuilt payment UIs, access to 125+ payment methods, and Link, a wallet built by Stripe.

• Expand to new markets faster: Reach customers worldwide and reduce the complexity and cost of multicurrency management with cross-border payment options, available in 195 countries across 135+ currencies.

• Unify payments in person and online: Build a unified commerce experience across online and in-person channels to personalize interactions, reward loyalty, and grow revenue.

• Improve payment performance: Increase revenue with a range of customizable, easy-to-configure payment tools, including no-code fraud protection and advanced capabilities to improve authorization rates.

• Move faster with a flexible, reliable platform for growth: Build on a platform designed to scale with you, with 99.999% historical uptime and industry-leading reliability.

Learn more about how Stripe Payments can power your online and in-person payments, or get started today.