What Is Agentic Commerce?

By the Agentic Commerce Report editorial team — covering the AI-to-checkout layer since September 2024.

McKinsey projects that AI agents will influence $3–5 trillion in global commerce annually by 2030. During Cyber Week 2025, 20% of purchases were already agent-influenced, according to industry data tracked in our Issue #23. What was theoretical 18 months ago is now a live revenue channel — and most retailers are not ready for it.

Agentic commerce is the shift from humans browsing and buying to AI agents researching, negotiating, and completing purchases on their behalf. It is not chatbots. It is not product recommendations. It is software that opens your accounts, reads your preferences, talks to merchant APIs, and closes transactions — without you clicking anything.

This guide explains what agentic commerce is, how the transaction flow works, which protocols power it, who the major players are, and what retailers need to do now to stay visible and sellable in an agent-driven market.

We cover:

• The formal definition and how it differs from AI-assisted shopping
• The five-step transaction flow, from intent to fulfillment
• The three protocol camps (UCP, ACP, MCP) and which is gaining ground
• Market size projections and growth data through 2030
• Who is building the infrastructure — from OpenAI and Google to Stripe and Visa
• Real-world examples with measured outcomes
• A practical readiness checklist for retailers
• Security and identity risks, including the emerging Know Your Agent (KYA) standard

If you are a retailer, a brand, a platform engineer, or a commerce operator, this is the foundational read.

What Is Agentic Commerce?

Agentic commerce is the use of AI agents to initiate, negotiate, and complete commercial transactions autonomously, operating within pre-authorized budgets and preference constraints set by the user.

The key word is autonomously. Traditional ecommerce requires a human at every decision point — search, compare, select, pay. AI-assisted shopping, which most retailers already offer in some form, adds recommendations and answers to that same human-driven process. Agentic commerce removes the human from most or all of those steps. The agent acts on the user's behalf with the user's credentials, money, and intent — but without requiring the user to participate in each step.

Here is the distinction in plain terms:

AI-assisted shopping: You search Amazon. The system recommends similar products. You decide.

Agentic commerce: You tell your agent "get me the best protein powder under $40, auto-ship monthly." The agent evaluates options across merchants, finds the best price against your stated constraints, initiates the purchase, and confirms it — possibly without you seeing a single product page.

This distinction matters because the entire retail funnel changes. Merchants are no longer selling to human shoppers first — they are selling to agents that filter, rank, and transact before any human sees the result.

The transaction chain in agentic commerce involves five actors:

1. User — sets the budget, constraints, and preferences upfront
2. AI agent — executes the shopping mission autonomously
3. Merchant — exposes products and pricing in a machine-readable format
4. Payment processor — authorizes agent-initiated transactions with scoped credentials
5. Protocol layer — provides the shared language for agent-merchant communication

Each actor must be ready for this model. Most are still catching up.

The agentic commerce definition also includes what it is not. It is not a smarter search bar. It is not a recommendation engine with extra steps. It is not a voice assistant that reads you product descriptions. Agentic commerce is transactional autonomy: the agent has the authority, the credentials, and the decision-making capacity to close a purchase without returning to the user for each step.

That transactional autonomy is what makes agentic commerce structurally different from every prior wave of ecommerce technology — and why it requires merchants, payment networks, and identity providers to rebuild assumptions about who is on the other side of a transaction.

How Does Agentic Commerce Work?

The transaction flow in agentic commerce follows five steps. Understanding this flow is essential for any retailer or platform that wants to remain visible and sellable in an agent-driven market. For a detailed technical walkthrough of each step — including how agents query catalogs, scope payment tokens, and handle offer resolution — see How AI Shopping Agents Work.

Step 1: Intent Capture

The user expresses a goal in natural language: "Order camping gear for a three-day trip, budget $500." The agent extracts structured intent: category (camping gear), budget ($500), timeline (implied: soon), constraints (as needed from prior preferences — brand loyalties, size preferences, shipping requirements). This intent is then translated into a machine-readable query that the agent uses to drive the discovery phase.

Step 2: Discovery

The agent queries merchant catalogs and comparison sources. This is where the first major difference from human shopping appears: the agent does not browse web pages. It reads structured product feeds, API endpoints, and machine-readable data formats. A product that lacks structured data — clear schema markup, accurate inventory signals, agent-accessible pricing — is effectively invisible at this stage. Discovery happens in milliseconds, across dozens of catalogs simultaneously. The agent does not read marketing copy. It reads metadata.

Step 3: Offer Resolution

The agent compares options against the user's stated constraints. It may negotiate in real time with merchant APIs that support dynamic pricing, check for applicable loyalty rewards, verify return policies, and cross-reference shipping lead times. This is not keyword matching — it is structured reasoning over product data. An agent operating under a "$500 camping gear" mandate will evaluate a sleeping bag at $89, a tent at $240, and a stove at $65, summing to $394 and leaving $106 for incidentals — and may add a headlamp without prompting if prior preference data indicates frequent night hiking.

Step 4: Authorization

The agent initiates payment using pre-authorized credentials with scoped permissions. This is not the user's full card or account — it is a bounded token that limits the agent to specific transaction types, maximum amounts, and approved merchants. Visa, Mastercard, and Stripe have each announced infrastructure updates to support this model. Stripe's agent-aware payment infrastructure distinguishes bot-originated transactions from human-originated ones, applying different scoring models — a development we covered in detail in Issue #18.

Step 5: Fulfillment

The transaction completes. The agent confirms the order, logs the purchase against the user's stated goal, and may schedule follow-on actions — for example, monitoring shipment tracking and initiating a return if the item arrives damaged. The user receives a summary of what was purchased, at what price, from which merchant, with what delivery estimate.

The Role of Protocols

Agents cannot communicate with merchant APIs without a shared protocol layer. Three protocol camps are competing to become that standard: UCP (Google), ACP (OpenAI), and MCP (Anthropic). They define how agents identify themselves, how they query catalogs, how offers are structured, and how payment tokens are scoped and transmitted.

Merchants that implement none of these protocols rely on web scraping and general-purpose AI browsing — a slower, less reliable path that is increasingly filtered out by agentic systems that prefer clean API access.

Why Structured Data Is Non-Negotiable

An agent reading a merchant's product catalog does not parse visual layouts or marketing copy. It reads metadata. Products with incomplete schema markup, missing pricing signals, or inconsistent inventory data are ranked lower or excluded from agent-initiated discovery entirely. The practical implication: a product that is perfectly optimized for Google Shopping may still be invisible to an agentic system if it lacks structured offer data, pricing validity windows, or machine-readable availability signals.

The Three Protocol Camps: UCP, ACP, and MCP

Three major technology companies are each promoting a protocol for agent-merchant communication. They are not fully interoperable. Merchants face a real decision about which to implement first — and the wrong choice could mean being invisible to the largest agent ecosystems.

Google Universal Commerce Protocol (UCP)

Google's Universal Commerce Protocol is an open coalition standard with 20+ partners, announced through Google's AI Mode and "Buy for me" infrastructure. UCP is designed to work across retailers without proprietary lock-in. Google's pitch is interoperability: a single protocol implementation that works across Google's AI Mode search results, the "Buy for me" purchasing assistant, and any third-party agent that adopts the UCP standard.

Because Google controls the largest share of commerce-adjacent search traffic, UCP adoption is well-positioned among mid-to-large retailers who depend on search visibility for customer acquisition. UCP prioritizes structured product feeds, pricing APIs, and agent-readable inventory signals that align closely with Google Merchant Center data formats many retailers already maintain.

OpenAI Agent Communication Protocol (ACP)

OpenAI's Agent Communication Protocol powers ChatGPT Shopping, which launched in February 2026 and reached 50 million shopping queries per day within its first month. ACP is the ChatGPT ecosystem standard. Integration complexity is low — OpenAI claims a 2–4 hour integration time for merchants who have existing product APIs — and the merchant directory (initially populated with Etsy, Glossier, SKIMS, and others) is growing rapidly.

ACP's strength is the size of the ChatGPT user base and the speed of merchant onboarding. Its risk: ACP is OpenAI-controlled. Merchants integrating only ACP are dependent on OpenAI's terms, fee structures, and product policies, which have not been fully disclosed.

Anthropic Model Context Protocol (MCP)

Anthropic's Model Context Protocol operates at the infrastructure layer. MCP is vendor-neutral — it does not tie a merchant to Anthropic's own agent products. Instead, it provides a common interface that any MCP-compatible agent can use to read product catalogs, pricing, and inventory. MCP has been adopted by a growing set of third-party agent frameworks precisely because it does not create a dependency on a single AI platform.

Integration complexity for MCP is higher than ACP, but the long-term flexibility is greater. A merchant that builds to MCP can serve any MCP-compatible agent — not just Claude, not just ChatGPT, but any agent runtime that adopts the standard.

Protocol Comparison

No single protocol has won. The defensible path for most retailers is to ensure clean structured product data that can serve all three, rather than betting exclusively on one.

Agentic Commerce Market Size and Growth Projections

The market data on agentic commerce is moving faster than most analysts anticipated. The primary research sources tell a consistent story: the agentic commerce layer is growing faster than broader ecommerce, and it is concentrating among platforms that have built the protocol infrastructure to support agent-to-merchant communication.

Grand View Research pegs the agentic commerce market at $7.71 billion in 2026. This measures the addressable market for agent-powered transaction infrastructure — the protocols, identity layers, payment APIs, and agent runtime services that make agentic transactions possible — not including the broader commerce volume those systems influence.

Mordor Intelligence projects the agentic AI in retail market will reach $60.43 billion by 2030, driven by adoption across product discovery, dynamic pricing, supply chain optimization, and autonomous purchasing.

McKinsey Digital projects that AI agents will influence $3–5 trillion in annual commerce by 2030. This represents agent-influenced volume — transactions that are either initiated or meaningfully shaped by AI agents. It includes both direct agent-completed purchases and hybrid flows where agents do the research and the human confirms.

Bain & Company narrows the US-specific figure: $300–500 billion in agentic commerce in the US market by 2030, representing 15–25% of total US ecommerce at projected growth rates. Bain's analysis focuses on consumer retail rather than B2B procurement, making it the most directly applicable benchmark for direct-to-consumer brands assessing their agentic commerce exposure.

Adobe Analytics reported that AI agents influenced $262 billion in global online sales during the 2025 holiday season. This is the clearest evidence that agentic commerce is not a 2030 event — it is happening now, at scale.

During Cyber Week 2025, 20% of purchases were agent-influenced, according to industry data we covered in Issue #23. One in five purchases during the highest-volume shopping period of the year was shaped by an AI agent.

For B2B, the trajectory is even steeper. Gartner projects $15 trillion in B2B purchases will be made via AI agents by 2028. This reflects the automation of procurement workflows that are already heavily digitized in enterprise purchasing.

Who's Building Agentic Commerce?

The infrastructure for agentic commerce is being built by a converging set of players: AI platform companies, major retailers, payment networks, and identity providers. Here is who matters and what they have shipped.

OpenAI

ChatGPT Shopping launched in February 2026 with a curated set of retail partners. By launch, ChatGPT was processing 50 million shopping queries per day. OpenAI's ACP protocol powers the agent-to-merchant communication layer. OpenAI's positioning is as the agent platform with the largest active user base — not the payment processor or the identity provider — which means its value to merchants is distribution reach, not infrastructure.

Google

Google is building on three fronts simultaneously: AI Mode (the successor to standard search results, which surfaces agent-ready product data at the top of the SERP), the "Buy for me" feature in Google Assistant, and Project Mariner (an autonomous web agent capable of completing multi-step commercial transactions across any website). UCP is Google's protocol bid, designed to route traffic from all three products through a standardized merchant integration that reuses Merchant Center data structures.

Amazon

Amazon's "Buy for Me" feature allows agents to complete purchases from non-Amazon retailers — an unusual move for a marketplace that historically preferred to keep transactions on its own platform. We covered the launch details in Issue #12. The Rufus AI shopping assistant has reached 300 million users, making it the largest deployed AI shopping assistant by user count among any single platform.

Shopify

Shopify's Agentic Storefronts initiative is building native agent-accessibility into the Shopify platform. The goal: any Shopify merchant that opts in gets automatic agent-readiness without custom API development. This is the clearest retailer infrastructure play in the ecosystem — a direct response to the risk that non-Shopify-native merchants will outpace Shopify's base in agentic accessibility.

Payment Infrastructure: Stripe, Visa, Mastercard

None of the above works without payment infrastructure that supports agent-initiated transactions. Stripe has updated Stripe Radar to score agent-initiated transactions separately from human-initiated ones, recognizing that the risk profile, behavioral patterns, and authorization requirements are materially different. Visa and Mastercard have each announced agentic commerce frameworks — scoped credentials and pre-authorization models that allow agents to transact without full account access.

Identity Layer: Experian and Prove

The identity verification layer is the piece most retailers have not considered yet. Experian and Prove are building Know Your Agent (KYA) infrastructure — the ability to verify the identity, authorization scope, and trustworthiness of an AI agent making a purchase on a consumer's behalf. Without KYA, a merchant accepting an agent-initiated transaction has no way to verify that the agent is legitimate and authorized.

Real-World Agentic Commerce Examples

The clearest evidence that agentic commerce works comes from live pilots with measured outcomes.

Copilot.com in-chat checkout produced a 194% higher purchase likelihood compared to standard web browsing for the same products. When an agent guides a shopper through discovery and initiates checkout within the conversation, the conversion rate nearly triples. This is the most cited outcome in the agentic commerce pilot literature and is consistent with the intent-to-purchase compression we have tracked across our coverage.

Stripe Radar bot scoring is now active for agent-initiated transactions. Stripe's system distinguishes between human and bot payment flows, applying different scoring models and approval logic to each. The practical effect: legitimate agent-initiated purchases get smoother processing; fraudulent bot traffic gets flagged separately from human fraud patterns. This separation matters because it means the payment layer is now agentic-aware, not just ecommerce-aware — a threshold we identified as critical in our Issue #18 protocol infrastructure coverage.

ChatGPT Shopping B2C pilots: Etsy, Glossier, and SKIMS were among the retailers in the initial ChatGPT Shopping launch cohort. Early pilot data suggests agent-initiated discovery converts at higher rates than paid social for these brands for high-intent queries, though merchant-reported data is still limited to the first quarter post-launch.

B2B autonomous procurement: Gartner's $15 trillion B2B projection by 2028 is rooted in the acceleration of autonomous procurement workflows. In enterprise B2B, purchasing agents already operate with pre-authorized spending authority and vendor relationships embedded in ERP systems. Agentic AI extends this further — agents can query vendor catalogs, compare contracts against pre-negotiated terms, initiate purchase orders, and route approvals with minimal human involvement.

The pattern across these examples is consistent: agentic commerce works best when product data is clean, transaction scope is clear, and payment authorization is pre-established. All three conditions are within a merchant's control.

How to Prepare for Agentic Commerce as a Retailer

Retailers who act now will capture agent traffic as the ecosystem scales. Retailers who wait will find themselves invisible to agents managing a growing share of their customers' purchases. The readiness checklist is five items — and none of them require a protocol decision to implement.

1. Structured Product Data

Every product needs complete, accurate, machine-readable data: title, description, category, attributes, pricing (including sale price and expiration date), inventory status, SKU, and shipping parameters. Inconsistent or incomplete data is the most common reason agents exclude a merchant from discovery at the offer resolution stage.

A practical audit: identify the 20% of SKUs with missing or outdated attributes in your product feed. Those SKUs are already invisible to agents, regardless of which protocol you implement. Fix the data layer first.

2. Product Schema Markup

Schema.org Product markup in JSON-LD format tells agents and AI crawlers the exact properties of each product without requiring them to parse your page layout or marketing copy. At minimum, implement Product, Offer, AggregateRating, and BreadcrumbList schema. For priority SKUs, add ItemAvailability and PriceValidUntil to enable reliable offer resolution.

3. llms.txt

The llms.txt file — analogous to robots.txt in function — tells AI agents which parts of your site are available for agent access, which products are agent-ready, and what your policies are for agent-initiated purchases. Without a llms.txt file, agents must infer permissions from your existing technical signals, which introduces friction and risk in the authorization step.

4. API-First Catalog

Agent systems prefer direct API access over web scraping. A product catalog API — even a simple REST endpoint that returns structured JSON for each SKU with current pricing and availability — dramatically improves your discoverability and transaction success rate in agentic systems. Shopify merchants who activate Agentic Storefronts get this by default. Non-Shopify retailers need to build or configure it independently.

API-first catalog access also enables real-time inventory signaling — agents can confirm availability at the moment of authorization rather than discovering an out-of-stock condition during fulfillment.

5. Agent-Readable Pricing

Dynamic pricing that changes within a session creates problems for agents trying to authorize a purchase at a stated price. Agent-readable pricing means stable price signals during an agent session, clear communication of sale end dates, and machine-readable promotional terms. If your pricing infrastructure cannot produce a reliable price commitment for the duration of an agent-to-merchant negotiation, that is a gap to address before protocol integration.

The retailers who will win agent-driven traffic are those who treat their catalog as an API-first product rather than a web page. The investment is recoverable — clean structured data serves SEO, paid search, comparison shopping engines, and now agentic systems simultaneously.

Agentic Commerce Security: Identity and Fraud Risks

Agentic commerce introduces fraud and identity risks that traditional ecommerce systems are not designed to catch. The core problem: how do you verify that the entity initiating a purchase is a legitimate AI agent acting on behalf of an authorized user — and not a malicious bot, a stolen credential, or an over-scoped agent exceeding its authorization?

Know Your Agent (KYA)

Know Your Agent is the agentic commerce equivalent of Know Your Customer (KYC). KYA is the process by which merchants, payment processors, and identity providers verify the identity, provenance, and authorization scope of an AI agent before processing a transaction.

The question KYA answers: "Is this agent who it claims to be, and is it authorized by the user to make this specific purchase?" Without KYA infrastructure, a merchant has no reliable way to answer that question. With KYA — currently being built by Experian and Prove, among others — merchants can score and trust agent-initiated transactions the same way they currently score human ones using device fingerprinting and behavioral signals.

As of mid-2026, KYA is not yet standardized. Merchants must evaluate each agent's identity claims independently, using a combination of payment token validation, behavioral scoring, and, where available, agent-specific verification tokens issued by the AI platform.

Pre-Authorized Scope

The most effective security mechanism in agentic commerce is pre-authorization with narrow scope. A well-designed agent transaction uses a payment token restricted to: a maximum transaction value, approved merchant categories, a time window, and a specific user account. An agent presenting a scoped token cannot exceed the authorized parameters — the payment processor rejects out-of-scope transactions automatically.

The primary fraud risk is not from legitimate agents exceeding scope. It is from malicious actors who spoof agent credentials, acquire stolen scoped tokens, or exploit over-permissive authorization grants established by users who did not fully understand what they were authorizing.

Conclusion

Agentic commerce is the most significant structural shift in retail since mobile commerce. AI agents that research, negotiate, and complete purchases autonomously are already driving hundreds of billions of dollars in annual commerce volume — and the trajectory points to trillions by 2030, according to McKinsey.

The retailers who are ready share three characteristics: clean, structured product data that agents can read; machine-readable pricing and inventory that agents can trust; and protocol implementation that makes their catalog findable in the major agent ecosystems.

The retailers who are not ready share one characteristic: they are optimizing exclusively for human shoppers in a market where an increasing share of purchase intent is now processed by software before any human sees a product page. During Cyber Week 2025, that share was already 20%. According to Bain, it will be 15–25% of all US ecommerce by 2030.

We have been covering this transition since September 2024 — 29+ issues of primary-source reporting on the protocols, the pilots, the platforms, and the data. The infrastructure is being built now. The winners will be those who build alongside it.

Subscribe to the Agentic Commerce Report at agenticcommerce.report for weekly primary-source coverage of the AI-to-checkout layer.

Frequently Asked Questions

Track the latest developments in agentic commerce weekly: browse the archive or explore by topic on the topics index.