What is agentic commerce?

You open your AI assistant and type: "I need trail running shoes under $150 that work for wide feet and wet conditions." Within seconds, the agent queries product catalogs from a dozen brands, cross-references real-time inventory and pricing, filters by your foot profile from a previous conversation, and presents three curated options. You reply, "Show me that second pair in dark green." The agent confirms your size is in stock and estimates delivery by Thursday.

You say, "Buy it." The agent initiates checkout through a structured commerce API, passes a tokenized payment credential to the merchant's payment processor -- never exposing your raw card number -- and confirms the order. You close the conversation and move on with your day.

Two days later, a winter storm delays shipments across the Midwest. Your agent notices the tracking update and sends you a message: "Your shoes are delayed until Saturday due to weather. I can reroute to a nearby pickup location for Friday delivery, or keep the original address." You choose Friday pickup. The agent coordinates the change with the fulfillment network.

No tabs. No search results pages. No checkout forms. No tracking number lookups.

This is agentic commerce -- the shift from human-driven online shopping to AI commerce where agents handle purchasing on a consumer's behalf. Instead of browsing pages, comparing prices, and managing checkouts manually, consumers set intent and guardrails, and an AI agent handles discovery, comparison, and purchase autonomously. The critical distinction from chatbots is that agentic commerce systems are goal-oriented, executing multi-step workflows across multiple merchants and services. They are not just answering questions.

The numbers suggest this is not a theoretical exercise. According to recent industry surveys, 73% of consumers are already using AI in their shopping journeys. During Cyber Week 2025, Salesforce reported that 20% of all global orders were influenced by AI agents or shopping assistants. Morgan Stanley projects that agentic shoppers could represent $190 billion to $385 billion in U.S. e-commerce spending by 2030. The global agentic commerce market, valued at $5.71 billion in 2025, is projected to reach $65.47 billion by 2033.

But a compelling consumer experience is only possible if there is robust infrastructure underneath it. Agents cannot query product catalogs that do not exist in machine-readable formats. They cannot complete checkout through APIs that have not been built. That is why the real story of agentic commerce is not just about AI. It is about the protocols, payment rails, and data standards that make it work.

The protocols powering agentic commerce

For AI agents to buy and sell across thousands of merchants, there needs to be a shared language: a set of open standards defining how product data is structured, how checkout works, how orders are managed, and how fulfillment is coordinated. Several protocols have emerged to fill this role, each addressing a different layer of the problem.

Universal Commerce Protocol (UCP)

The Universal Commerce Protocol is the most broadly endorsed commerce standard in the space. Co-developed by Google and Shopify, with contributions from Etsy, Wayfair, Target, and Walmart, UCP has attracted 40+ endorsers including Stripe, PayPal, Mastercard, Visa, Klarna, Salesforce, Best Buy, Gap, Sephora, and Macy's.

UCP describes itself as "the common language for platforms, agents and businesses." Its initial launch covers three core capabilities:

• Checkout: Complex cart logic, dynamic pricing, and tax calculations via unified checkout sessions
• Identity linking: OAuth 2.0-based secure relationships between agents and merchants
• Order management: Real-time webhooks for status updates, shipment tracking, and return processing

UCP is surface-agnostic, meaning it works whether the consumer interacts through chat, voice, or a visual interface. Its architecture defines four roles: the platform or agent (consumer-facing AI), the business (merchant of record), the credential provider (managing user data and payment tokens), and the payment service provider (processing transactions).

A key design principle is that UCP is business-centric. Retailers retain control of their business rules and remain the merchant of record. This is not a platform play that disintermediates brands. It is a standard that lets them participate in agent-driven commerce on their own terms.

Agentic Commerce Protocol (ACP)

The Agentic Commerce Protocol, created collaboratively by Stripe and OpenAI, takes a more focused approach. ACP is "an open standard for programmatic commerce flows between buyers, AI agents, and businesses," with a particular emphasis on enabling transactions to occur directly within AI interfaces.

ACP supports both REST and MCP implementations, works with existing commerce infrastructure, and handles PCI-compliant payment processing through Stripe's Shared Payment Token -- securely passing payment credentials without exposing underlying card data. It covers physical goods, digital goods, subscriptions, and asynchronous purchases.

OpenAI is the first AI platform implementing ACP with ChatGPT, and more than 1 million Shopify merchants have opted into OpenAI's commerce features. The experience is still evolving. OpenAI's initial ChatGPT Instant Checkout (launched September 2025) showed that while users enthusiastically researched products in the platform, most still completed purchases on merchant storefronts. The approach has since shifted to surface products within conversations while directing final checkout to the merchant's site.

ACP and UCP are sometimes framed as competitors, but they operate at slightly different scopes. ACP is more narrowly focused on the Stripe and OpenAI ecosystem, and UCP has broader industry endorsement and a wider architectural vision. Merchants may need to support both.

Order Network eXchange (ONX)

If UCP and ACP define how agents discover products and initiate checkout, ONX addresses what happens after the buy button is clicked. Created by the Commerce Operations Foundation (with Pipe17 as a key driving force), ONX is an open specification creating "a common language for order data exchange across commerce ecosystems."

ONX tackles the operational layer beneath checkout: orders, inventory, shipments, returns, and fulfillment. As the Foundation's CEO Tom Schmitt put it, "AI transforms order starts, but fulfillment delivers on that promise."

The protocol covers 15 core operations from order capture to shipment tracking and works across ERPs, order management systems, warehouse management systems, third-party logistics providers, and commerce platforms. It has attracted 50+ members representing more than $1 trillion in GMV, including Manhattan Associates, IBM Sterling, SPS Commerce, Ryder, and commercetools.

Basis Theory Open Agentic Commerce API

Basis Theory has published an open specification with a distinct architectural approach. Its key innovation is the Commerce Manifest -- a .well-known/commerce-manifest endpoint that merchants expose to advertise their capabilities, serving as a single source of truth for endpoints, schemas, payment methods, and authentication requirements. The specification uses Schema.org vocabulary for product semantics and supports payment rails from tokenized cards to stablecoins.

The common thread

These protocols share a fundamental premise: AI agents should interact with merchants through structured, machine-readable APIs, not by scraping web pages or automating browser clicks. The protocols are largely complementary, addressing different layers of the commerce stack, and most are released under the Apache 2.0 open-source license.

How AI agents communicate

Commerce protocols define what gets exchanged -- product data, cart contents, and order status. But agents also need standards for how they connect to systems and to each other. Three agent communication protocols form this infrastructure layer.

Model Context Protocol (MCP)

The Model Context Protocol, created by Anthropic, is an open-source standard for connecting AI applications to external systems. The simplest analogy is MCP is like USB-C for AI -- a standardized way to connect any AI application to any data source, tool, or workflow.

MCP handles vertical integration -- the connection between an agent and the tools and data it needs to do its job. In a commerce context, MCP is the layer that lets an AI agent talk to a merchant's commerce API, query a product database, or trigger a fulfillment workflow.

Critically, MCP is a transport mechanism, not a commerce schema. It does not define what product data looks like or how checkout works -- that is the job of UCP, ACP, and ONX. But those commerce protocols can be delivered through MCP servers, making MCP commerce-capable without being commerce-specific. ONX is explicitly "built on MCP." Visa's Acceptance Agent Toolkit runs on top of an MCP Server. MCP is the glue layer.

MCP has broad ecosystem support with Claude, ChatGPT, Visual Studio Code, Cursor, and many other platforms supporting it.

Agent-to-Agent Protocol (A2A)

While MCP connects agents to tools, the Agent-to-Agent Protocol connects agents to each other. Originally developed by Google and now maintained by the Linux Foundation, A2A enables communication and collaboration between AI agents regardless of their underlying framework.

A2A handles horizontal integration. In an agentic commerce scenario, a shopping agent discovers products and delegates payment processing to a payment agent, which delegates shipping coordination to a logistics agent. Each agent publishes an Agent Card -- self-describing metadata advertising its capabilities -- so other agents can discover and interact with it without sharing internal memory or proprietary logic.

A2A and MCP are explicitly complementary. MCP defines how one agent accesses data and APIs (vertical) while A2A defines how agents talk to each other (horizontal). Together, they form the full communication stack for multi-agent commerce.

Agent Payments Protocol (AP2)

The Agent Payments Protocol, created by Google, addresses a specific gap: current payment systems assume a human user is directly authorizing purchases. When an AI agent initiates a payment, three questions arise that existing infrastructure was not designed to answer:

1. Authorization verification -- is this agent authorized for this specific transaction?
2. Authenticity confirmation -- does the request reflect genuine user intent?
3. Accountability clarity -- who is responsible when something goes wrong?

AP2 solves this through Verifiable Digital Credentials (VDCs) -- cryptographically signed digital contracts called "Mandates." An Intent Mandate captures standing conditions for autonomous purchases ("buy paper towels when they are under $8"). A Cart Mandate records explicit user authorization for specific items and prices, with cryptographic signatures. A Payment Mandate is shared with payment networks and issuers to signal agent involvement and whether the human was present.

AP2 is designed as an open extension for A2A and integrates with Google's Agent Development Kit, MCP, and A2A. UCP explicitly lists AP2 as a supported protocol standard, making AP2 the payment security layer that UCP relies on for agent-initiated transactions.

The layered architecture

It helps to think of these protocols as a stack:

• Consumer layer: ChatGPT, Gemini, Copilot, brand AI assistants -- where intent is expressed
• Agent communication layer: MCP (agent-to-tools) and A2A (agent-to-agent) -- how agents connect
• Commerce protocol layer: UCP, ACP -- product discovery, checkout, order management
• Payment layer: AP2, Visa, Mastercard, x402, Stripe tokens -- how money moves
• Operations and fulfillment layer: ONX -- orders, inventory, shipments, returns

Each layer builds on the ones below it. Commerce protocols define the business logic; agent protocols provide the communication infrastructure; payment protocols secure the financial transactions; and operations protocols ensure that what was sold actually gets delivered.

Agentic payments: completing the loop

An agent can discover products, compare prices, and build a cart, but none of that matters if it cannot securely complete a payment. The world's largest payment networks are building this layer now.

Visa Intelligent Commerce

Visa Intelligent Commerce leverages Visa's infrastructure -- 4.8 billion payment credentials and 150+ million merchant locations -- to enable AI agents to conduct secure transactions on behalf of consumers and businesses.

The centerpiece is the Acceptance Agent Toolkit, built on top of an MCP Server. It provides prebuilt workflows for commerce tasks like invoicing and pay-by-link, accessible through plain-language prompts. Available as a self-hosted npm package, the toolkit bridges AI agents and established payment infrastructure.

Visa has also launched the Trusted Agent Protocol (October 2025), an open framework developed with 10+ partners to help merchants distinguish between malicious bots and legitimate AI agents. Early pilots with Skyfire, Nekuda, PayOS, and Ramp have executed end-to-end consumer and B2B purchases in closed beta.

Mastercard Agent Pay

Mastercard Agent Pay builds on Mastercard's proven tokenization capabilities with purpose-built agentic tokens for agent-initiated transactions. These extend existing mobile contactless and card-on-file tokenization to support programmable payments, recurring expenses, and subscriptions in agentic contexts.

The rollout has moved quickly. After launching in April 2025, all U.S. Mastercard cardholders were enabled by the 2025 holiday season, with global expansion to Australia, New Zealand, Singapore, Malaysia, and India underway in 2026. Mastercard completed Korea's first live agentic transactions and is integrating with Microsoft Copilot Checkout. A partnership with PayPal brings Agent Pay into PayPal's wallet, letting AI agents complete transactions through one of the world's largest digital payment platforms.

x402

x402, created by Coinbase's Developer Platform, takes the most radical approach. It is an open, internet-native payment standard that uses the HTTP 402 (Payment Required) status code, which has existed in the HTTP specification for decades but was never widely implemented.

The flow is simple. A client sends an HTTP request, the server responds with a 402 status if no payment is included, the client pays instantly using stablecoins and retries. Zero protocol fees, no accounts, no personal information required.

For agentic commerce, x402 is purpose-built for agent-to-agent transactions where neither party is a human. Agents can pay for services programmatically via standard HTTP, with no signups, API keys, or approval processes. Early adoption has been concentrated in crypto-native applications, but the protocol's simplicity makes it a candidate for broader use. Stripe, AWS, Alchemy, Vercel, and Cloudflare are listed among its adopters.

The agentic payments picture

The payment layer for agentic commerce is not a blocker. It is actively being built by the industry's largest players. Whether through card network tokens (Visa, Mastercard), embedded payment flows (Stripe via ACP), cryptographic mandate proofs (AP2), or stablecoin settlement (x402), multiple rails are converging to give agents the ability to move money securely on behalf of users.

B2B agentic commerce and the enterprise opportunity

The scenario that opens this post -- a consumer buying running shoes through an AI assistant -- is the most intuitive illustration of agentic commerce. But it may not be the most consequential one. The same protocol stack that powers consumer shopping applies directly to business procurement, supply chain management, and B2B transactions, often with higher stakes and greater complexity.

The core interaction pattern is identical in both contexts: an AI agent discovers products or services, negotiates terms, initiates checkout, and manages post-purchase operations via structured APIs. Both B2C and B2B ride the same protocol stack described above.

The differences lie in complexity and scale:

• Transaction structure: A consumer buys one pair of shoes. A procurement agent negotiates pricing, terms, and volume discounts across hundreds of suppliers, manages multi-stakeholder approval chains, and handles contract-based pricing with net terms.
• Payment methods: Consumers use cards and wallets. Businesses use invoicing, purchase orders, ACH transfers, and virtual cards -- each with its own compliance requirements.
• Autonomy and approval: Consumer transactions are moving toward agent autonomy with human confirmation. B2B procurement involves approval hierarchies and spend limits, but routine reorders often have higher autonomy because the guardrails are well-defined.
• Fulfillment: Consumer fulfillment means shipping to an address. B2B fulfillment involves split shipments, multiple warehouses, third-party logistics coordination, and just-in-time ordering -- precisely where ONX's operational backbone becomes critical.

The B2B opportunity may dwarf the consumer side. Gartner projects that 90% of all B2B purchases will be handled by AI agents within three years, channeling more than $15 trillion in spending through automated exchanges. Forrester expects procurement teams to deploy agents capable of negotiating across hundreds of suppliers simultaneously. And according to a CrewAI survey, 100% of enterprises surveyed (500 C-level executives at companies with $100M+ revenue) plan to expand agentic AI use in 2026.

For both B2C and B2B, the prerequisite is the same: structured, machine-readable product data exposed through protocol-compliant APIs. Brands that do not meet this bar will not just lose rankings in a search engine. They will be invisible to the AI agents making purchasing decisions. Data quality is becoming the new SEO.

For a deeper look at what this means for your brand specifically, see our companion post: Why should I care about agentic commerce?

Where CorgiMaps fits in

The protocols are published. The payment rails are being built. The AI platforms are live. But for most brands and manufacturers, a practical gap remains: their product data is not ready.

Product information lives in PIMs designed for human merchandisers. Pricing and inventory sit in ERPs built for internal reporting. Catalog data is scattered across e-commerce platforms, spreadsheets, and legacy systems that were never intended to serve machine-readable APIs. Even brands with sophisticated digital infrastructure typically lack the structured, protocol-compliant endpoints that AI agents need to discover products, check availability, and initiate checkout.

This is the problem CorgiMaps addresses. CorgiMaps is a data integrations layer that connects to the systems brands already use -- PIMs, ERPs, e-commerce platforms, order management systems -- and exposes clean, structured, protocol-ready endpoints. It does not require rebuilding a tech stack or migrating to a new platform. It adds a bridge layer that translates existing product data into the formats that UCP, ACP, ONX, and the broader agentic commerce ecosystem require.

The timing matters. Consumer adoption is growing (23% of Americans made purchases using AI in the past month), enterprise procurement is automating ($15 trillion in B2B spending projected to shift to AI agents per Gartner), and the protocols defining how agents interact with merchants are solidifying now. Brands that expose protocol-compliant product data will be discoverable by AI agents as adoption scales.

We have seen this pattern before. Companies without websites in 2005 and those without mobile-optimized storefronts in 2015 lost visibility in the channels where buying decisions were being made.

The infrastructure layer of agentic commerce is being built in the open by the largest technology and payment companies in the world. For brands, the practical question is whether their product data is structured and accessible enough for AI agents to find, interpret, and transact against.