Agentic Commerce Explained: How AI Agents Shop, Negotiate and Pay for You
Wasim · 7 Oct 2026What Is Agentic Commerce?
Agentic commerce is a model of buying and selling in which AI agents act on behalf of consumers or businesses. They research products, compare options, negotiate terms and complete the transaction, with little or no human help at each step.
Traditional ecommerce asks a human to do all the work: search, open ten tabs, read reviews, compare prices, fill in a cart and type card details. In agentic commerce, you state a goal instead:
"Find me waterproof trail-running shoes in size 10, under $150, that can arrive before Saturday, and buy the best-reviewed pair."
An AI agent takes it from there. It searches catalogs, checks stock and delivery dates, weighs reviews, applies your preferences and budget, and checks out using a payment credential you've authorized. You get a confirmation, or a quick approval prompt if your rules require one.
It's one of the fastest-rising topics in commerce. "Agentic commerce" was first flagged as a trending search topic on June 5, 2026, and already sees about 14.8K monthly searches, up a striking +8200%. That growth tracks a year in which AI platforms, payment networks and retailers have all released the infrastructure that agent-driven shopping needs.
From Search to Delegation: How We Got Here
| Era | How people buy | Who does the work |
|---|---|---|
| Ecommerce | Browse websites, click "Buy" | The human, entirely |
| Mobile and social commerce | Apps, one-tap checkout, shoppable posts | The human, with less friction |
| Conversational/AI-assisted | Ask a chatbot for recommendations | AI suggests, the human acts |
| Agentic commerce | Delegate a goal to an agent | The AI agent acts, and the human sets rules and approves |
The key shift is from AI as an advisor to AI as an actor. An agent doesn't just recommend; it can take action, including spending money within limits you've set.
How Agentic Commerce Works
A typical agentic purchase moves through five stages:
1. Intent "Reorder coffee beans, same brand, under $25"
│
2. Discovery Agent queries merchant catalogs / product feeds
│
3. Evaluation Compares price, reviews, delivery, return policy
│
4. Negotiation Applies coupons, loyalty, bundles, B2B terms
│
5. Transaction Checks out with a tokenized, user-authorized
payment credential → order confirmation
1. Intent and Mandate
The user (or a business) defines the goal and the guardrails: budget, preferred brands, delivery deadline, and whether the agent may buy on its own or must ask first.
2. Discovery
The agent needs machine-readable access to products: structured catalogs, APIs and product feeds instead of pages designed for human eyes. This is where headless, API-first commerce becomes a big advantage.
3. Evaluation
The agent compares options against the user's criteria: price, specs, ratings, shipping speed, sustainability, return policy and past preferences.
4. Negotiation
In consumer shopping, this might mean finding the best coupon or bundle. In B2B, agents can negotiate volume pricing, delivery schedules and contract terms. In some setups, the other side is also an agent, which leads to agent-to-agent (A2A) commerce.
5. Transaction and Fulfillment
The agent checks out using a payment token tied to the user's permissions, not a raw card number. The merchant fulfills the order, and the agent tracks shipping, handles returns or reorders when supplies run low.
The Protocols Powering Agentic Commerce
For agents to buy from millions of merchants, they need shared standards, just as the web needed HTTP. Several open protocols have emerged since 2025. They are mostly complementary layers, not rivals:
| Protocol | Backed By | What It Handles |
|---|---|---|
| ACP (Agentic Commerce Protocol) | OpenAI and Stripe | Agent-to-merchant checkout flows |
| AP2 (Agent Payments Protocol) | Google, now being standardized through FIDO Alliance working groups | Verifiable user authorization for agent payments |
| UCP (Universal Commerce Protocol) | Co-developed by Google, Shopify and major retailers | Catalog, cart, checkout, identity and order management across AI surfaces |
| Visa Intelligent Commerce | Visa | Agent-specific payment tokens, authentication and payment instructions |
| Mastercard Agent Pay | Mastercard | "Agentic tokens" tied to a specific agent and its limits |
ACP: Agentic Commerce Protocol
Developed by OpenAI and Stripe, ACP is "an open standard for programmatic commerce flows between buyers, AI agents, and businesses." Merchants implement the spec so any compatible agent can start a checkout. Merchants remain the merchant of record and keep control over which products are sold and how orders are fulfilled. Buyers can pass payment credentials to agents without exposing them, through mechanisms like Stripe's Shared Payment Token. ChatGPT was the first AI platform to implement it. See OpenAI's commerce docs and Stripe's agentic commerce docs.
AP2: Agent Payments Protocol
Announced by Google in September 2025, AP2 answers the hardest question in agentic payments: how does a merchant or bank know the user actually authorized this purchase? It uses cryptographically signed mandates, verifiable digital credentials that record what the user approved, from the original constraints (budget, allowed instruments) to the final cart. Chained together, they form an auditable trail. Its guiding principle is "Verifiable Intent, Not Inferred Action." AP2 extends Google's Agent2Agent (A2A) protocol and works with MCP.
UCP: Universal Commerce Protocol
UCP calls itself "the common language for platforms, agents, and businesses." It covers catalog search, cart building, checkout sessions, identity linking via OAuth 2.0 and order webhooks. It uses AP2 for payments and supports MCP and A2A. Its co-developers include Google, Shopify, Etsy, Wayfair, Target and Walmart, with endorsers such as Visa, Mastercard, PayPal, Adyen and American Express.
Card Networks: Visa and Mastercard
The card networks are adapting their rails for agents. Visa Intelligent Commerce provides agent-specific payment tokens, step-up authentication with passkeys, and "payment instructions" that check that every charge matches what the user asked for. Mastercard Agent Pay issues agentic tokens tied to a particular agent and its spending limits. Both let issuers and merchants tell a trusted agent apart from a bot, and give clear data when disputes come up.
Practical takeaway: real deployments often combine layers. For example, ACP or UCP for the merchant flow, AP2 for proof of consent, and Visa or Mastercard tokens for the payment itself.
How Big Is Agentic Commerce?
Forecasts vary a lot because firms define "agentic" differently. Some count only purchases agents complete themselves; others include purchases agents merely influence. Still, they agree on direction:
- Bain & Company: the US agentic commerce market "could reach $300 to $500 billion by 2030", roughly 15–25% of US ecommerce. About 30–45% of US consumers already use generative AI for product research.
- Morgan Stanley: agentic shoppers could account for roughly 10–20% of US ecommerce by 2030 (about $190B base case, $385B bull case).
- Juniper Research projects global agentic commerce spend of about $1.5 trillion in 2030. Higher-end consulting estimates reach several trillion dollars globally.
When narrowed to US online retail, most estimates land in the same range: agents handling roughly a tenth to a quarter of ecommerce by 2030.
Real-World Use Cases
For Consumers
- Routine replenishment: groceries, pet food, coffee, household supplies, reordered automatically.
- Deal hunting: watching prices and buying when a product drops below a target.
- Travel booking: finding and booking flights and hotels within budget and loyalty preferences.
- Gift shopping: "a $50 gift for a 10-year-old who loves science, delivered by Friday."
- Subscriptions management: cancelling unused services and switching to better plans.
For Businesses (B2B)
- Procurement: agents that source suppliers, request quotes, compare bids and issue purchase orders within policy.
- Inventory restocking: automatic reorders when stock hits a threshold.
- Contract negotiation: agents negotiating volume discounts or delivery terms, sometimes with a supplier's agent.
- Expense and travel management: booking compliant travel automatically.
For Merchants
- Selling inside AI assistants: being discoverable and purchasable where shoppers now ask questions.
- Sales agents: merchant-side agents that answer, recommend, bundle and close.
- Dynamic offers: responding to agent requests with tailored prices or bundles.
Benefits of Agentic Commerce
- Efficiency: tasks that take a human 30 minutes of comparison happen in seconds.
- Better decisions: agents can weigh far more options and data points than a person realistically would.
- Less friction: no forms, no abandoned carts, no repeated checkout steps.
- Personalization: agents remember preferences, sizes, budgets and brand loyalties.
- Always on: agents can monitor prices and stock 24/7 and act at the right moment.
- Lower operating costs for businesses: automated procurement and reordering reduce manual work.
- New sales channels for merchants: reaching customers inside AI platforms.
Risks and Challenges
Agentic commerce also raises serious questions:
- Trust: Bain and others identify consumer trust as the main barrier. Most shoppers aren't yet comfortable letting AI handle an entire purchase.
- Authorization and liability: if an agent buys the wrong thing, who's responsible: the user, the agent platform, the merchant or the bank? Protocols like AP2 exist mainly to answer this.
- Fraud and security: malicious agents, prompt injection on product pages and impersonation of legitimate agents. Agent tokens and signed agent identities are the defenses being built.
- Privacy: agents need access to personal data, payment methods and purchase history.
- Errors and hallucinations: an agent misreading a spec or a size chart has real costs.
- Merchant disintermediation: when an agent sits between brand and buyer, merchants risk losing the customer relationship, brand loyalty and first-party data.
- Bias and transparency: how do agents rank products, and could they be influenced by paid placement?
- Fragmented standards: several overlapping protocols are still maturing.
How Businesses Can Get Agent-Ready
You don't have to wait for the standards to settle. Here's how merchants can prepare now:
1. Make Your Catalog Machine-Readable
Agents rely on structured data. Add complete, accurate schema.org Product markup (Google's product structured data guide is a good reference), and keep product feeds rich with specs, availability, pricing and shipping details.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Product",
"name": "TrailRunner Waterproof Running Shoe",
"sku": "TR-WP-10",
"brand": { "@type": "Brand", "name": "TrailRunner" },
"description": "Waterproof trail-running shoe with Vibram outsole.",
"offers": {
"@type": "Offer",
"price": "139.00",
"priceCurrency": "USD",
"availability": "https://schema.org/InStock"
},
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": "4.7",
"reviewCount": "312"
}
}
</script>
Tip: validate JSON-LD and API payloads with the JSON Prettifier or Fix JSON. A single malformed bracket can hide your product from machines.
2. Go API-First
Agents work through APIs, not page layouts. A headless commerce architecture that exposes catalog, cart and checkout as APIs puts you in a strong position to support ACP or UCP.
3. Adopt Agentic Checkout Protocols
Check whether your commerce platform or payment provider already supports ACP, UCP or AP2. Shopify, Stripe and the major PSPs have been adding support, so it may be a configuration step rather than a rebuild.
4. Set Agent Policies
Decide which agents you accept, which products are agent-purchasable, and how you'll verify agent identity and handle disputes.
5. Optimize for AI Discovery
Clear product titles, honest specifications, real reviews and transparent return policies matter even more when an algorithm, not a person, is skimming your page.
6. Keep the Relationship
Use loyalty programs, post-purchase communication and your own merchant-side agents so you don't become an anonymous supplier behind someone else's AI.
The Future of Agentic Commerce
The direction is clear: shopping is moving from "search and click" to "describe and delegate." In the near term, expect agents to win first in routine, spec-driven purchases: replenishment, commodities and clear-criteria buys. Trust will then grow toward considered purchases like apparel, electronics and travel. As protocols mature and payment networks build agent identity into their rails, agent-to-agent commerce between buyer and seller agents will become normal, particularly in B2B.
For consumers, that means less time spent shopping and better-informed purchases. For businesses, the question is no longer whether AI agents will become customers, but whether your store will be ready when they arrive.
Related Reading
- Headless Commerce Explained: A Complete Guide to Decoupled Ecommerce
- Liferay Commerce Explained: B2B and B2C on One Platform
- REST APIs Auto-Generated by Liferay Objects
- More practical guides in Web dev, or browse the free developer tools.
