The Agentic Commerce Landscape
Understanding the protocols, players, and architecture reshaping how commerce gets done
The Shift We Are Living Through
Commerce has been through three major platform transitions in the past 25 years. The move from physical stores to websites. The move from desktop to mobile. The move from owned storefronts to marketplaces. Each wave opened new channels, created new winners, and forced established businesses to adapt or lose relevance.
Agentic commerce is the fourth wave, and it is different in a fundamental way.
The first three transitions changed where people bought things. Agentic commerce changes how decisions get made. It introduces a new actor, the AI agent, between the buyer and the seller. That agent does not browse. It does not scroll. It does not click through five pages of reviews. It interprets intent, evaluates options, makes recommendations, and in an increasing number of cases, executes purchases.
By Black Friday 2025, AI-driven traffic to US retail sites had surged 805 percent year over year, with agents driving over $22 billion in global online sales. McKinsey projects the global agentic commerce market will reach $3 trillion to $5 trillion in revenue by 2030.
The infrastructure is being built right now, in real time, by the same companies that built the last generation of payments and commerce. Mastercard, Visa, Google, Stripe, OpenAI, Shopify, and dozens of others are shipping protocols, tokens, and standards designed to make agent-mediated transactions safe, scalable, and interoperable.
Key Distinction
Assisted AI helps you do things faster. A chatbot that answers your question about return policy is assisted AI.
Conversational commerce lets you buy through a conversation. Ordering through a voice assistant is conversational commerce.
Agentic commerce delegates the decision. The AI agent shops, compares, negotiates, and buys on your behalf, often interacting with other agents along the way.
This is not a theoretical future. The infrastructure is being built right now, by the same companies that built the last generation of payments and commerce. If you work in payments, product, commerce, or fintech, this is your landscape for the next decade.
The Protocol Map
The single most important thing to understand about agentic commerce in 2026 is that it is a protocol war. Multiple competing and complementary standards are being built simultaneously by different coalitions of companies. Understanding who is building what, and why, is the foundation for every strategic decision you will make in this space.
The Agentic Commerce Protocol (ACP)
OpenAI and Stripe co-developed the Agentic Commerce Protocol, which launched with instant checkout capabilities in ChatGPT in September 2025. By March 2026, OpenAI shifted its approach. Instant Checkout is being moved to ChatGPT Apps, rather than happening directly in the chat interface. Merchants like Target, DoorDash, Instacart, and The Knot have built ChatGPT apps that handle checkout within their own controlled environments.
The strategic logic is clear: the checkout itself, subscriptions, inventory, shipping, taxes, and merchandising options all constitute a protective moat. Handing all of that to a third-party chat interface was always going to create friction with merchants who wanted to control the experience.
Agent Payments Protocol (AP2) and Universal Commerce Protocol (UCP)
Google launched two complementary protocols on January 11, 2026. AP2 handles the payment layer. UCP handles the broader commerce layer, establishing open-source technical standards for AI agents to execute purchases across different retail platforms without requiring custom integrations for each merchant.
AP2 separates transactions into three mandate layers, each of which can be monitored and audited independently:
- Intent Mandate: The consumer instructs the AI agent with specific parameters ("buy one winter jacket, maximum budget $200"). This is recorded as a verifiable digital contract.
- Cart Mandate: The agent searches, evaluates, and selects products that match the stated intent, adding qualifying items to a cart.
- Payment Mandate: The agent executes the purchase within the authorised parameters, using the consumer's approved payment method.
This three-layer separation is architecturally significant. It means each stage of the transaction can be independently verified, audited, and disputed. Klarna has expanded its partnership with Google to support AP2, and has also launched its own Agentic Product Protocol. Affirm also supports AP2 and UCP.
Mastercard Agent Pay and Verifiable Intent
Mastercard launched Agent Pay at NRF on January 11, 2026, and followed it on March 5 with Verifiable Intent, a cryptographic trust standard co-developed with Google and endorsed by IBM, Worldpay, Fiserv, Getnet, Checkout.com, Basis Theory, and Adyen.
Verifiable Intent addresses a specific and critical question: when an AI agent initiates a purchase on behalf of a consumer, how does anyone know it is actually doing what it was told to do? The specification creates a tamper-resistant record linking consumer identity, specific purchase instructions, and merchant transaction data into a single privacy-preserving authorisation record.
Santander and Mastercard completed Europe's first live end-to-end payment executed by an AI agent in early March 2026, processing a real transaction through Santander's live payments infrastructure to validate the operational and control framework.
Visa Intelligent Commerce (VIC) and Trusted Agent Protocol
Visa launched Visa Intelligent Commerce as a global initiative, and in October 2025, introduced the Trusted Agent Protocol (TAP) with more than 10 partners. TAP is an open framework designed on existing web infrastructure that enables safe agent-driven checkout by helping merchants distinguish between malicious bots and legitimate AI agents acting on behalf of consumers.
By late 2025, hundreds of secure agent-initiated transactions had been completed in collaboration with ecosystem partners. Visa is now working with over 100 partners worldwide, with over 30 actively building in the VIC sandbox and over 20 agents and agent enablers integrating directly.
Stripe Shared Payment Tokens (SPTs)
Stripe built Shared Payment Tokens in late 2025 specifically for agentic commerce. SPTs let agents initiate payments using a buyer's permission and preferred payment method, without exposing the underlying credentials. They are designed to be programmable, reusable, and interoperable.
In March 2026, Stripe expanded SPT support to include Visa and Mastercard agentic network tokens, as well as BNPL from Affirm and Klarna. This makes Stripe the first and only provider supporting both agentic network tokens and BNPL tokens through a single primitive. For merchants already processing payments through Stripe, these new payment methods are automatically available in agentic flows without additional integration.
Protocol Comparison
| Protocol | Lead Players | Primary Function | Key Differentiator |
|---|---|---|---|
| ACP | OpenAI, Stripe | Agent-to-merchant checkout | Largest consumer AI surface |
| AP2 / UCP | Google, Klarna, Affirm | Three-mandate payment + commerce | Auditable mandate layers |
| Agent Pay + Verifiable Intent | Mastercard, Google, IBM | Trust + authorisation proof | Cryptographic intent verification |
| VIC + TAP | Visa, 100+ partners | Agent authentication + discovery | Bot/agent distinction at scale |
| SPTs | Stripe, Visa, MC, Klarna, Affirm | Permissioned payment tokens | Single primitive, multi-rail |
The critical insight: these protocols are not all competing for the same layer. ACP and AP2/UCP handle the commerce flow. Verifiable Intent and TAP handle the trust layer. SPTs handle the payment execution layer. The winners will be the ones that interoperate across layers, not the ones that try to own the entire stack.
The Four-Layer Agentic Commerce Stack
To make sense of the protocol landscape, we need a mental model. The four-layer stack describes every agentic commerce transaction from beginning to end:
Every player in agentic commerce operates in one or more of these layers. Your strategic position depends on which layers you own, which you participate in, and which you depend on others for.
Layer 4: Discovery
In agentic commerce, the consumer tells an agent what they want, and the agent goes looking. The agent does not see your website design. It reads structured data, product attributes, availability signals, reviews, ratings, and price comparisons. If your product data is not machine-readable, you do not exist to the agent.
Layer 3: Decision
The decision layer is where the conversation happens. When a consumer says "find me a sustainable gift under fifty dollars for a coworker who loves cooking, and it has to arrive by Friday," the agent is operating in the decision layer. It interprets intent, evaluates options against constraints, and narrows the field.
The smartest operators will do both: optimise for third-party agent discovery while building owned conversational experiences that create durable competitive advantage.
Layer 2: Trust
Trust is the layer that makes or breaks the entire stack. Without it, agents are just code with access to your payment credentials. With it, they are accountable, auditable, and governable.
Only 16 percent of US consumers currently trust AI to make payments on their behalf. Only 29 percent of UK consumers would trust AI to make even small automated payments. Closing that gap requires infrastructure, not reassurance.
Layer 1: Payment
The payment layer is where the transaction actually settles. The core challenge is that trust cannot be inferred from credential possession. In agentic commerce, the agent has the credentials but that does not mean it is authorised to use them for this specific transaction, at this specific amount, with this specific merchant. Trust has to be explicitly granted, scoped, and enforced in code.
Where MCP Fits
Model Context Protocol provides a connectivity layer. It provides a standardised way for AI agents to connect to tools, data sources, and services. An agent using MCP can query an inventory database, check a shipping API, or pull customer data from a CRM. MCP tells the agent how to talk to systems.
The commerce protocols (ACP, AP2, UCP, Agent Pay, VIC) tell the agent how to conduct commerce. They handle product discovery, intent verification, payment authorisation, and settlement. These are specialised commerce flows that sit on top of the general connectivity that MCP provides.
MCP: "Here is how you connect to Shopify's inventory API and retrieve product data."
ACP/AP2: "Here is how you conduct a purchase on behalf of a consumer, including how to discover products, verify intent, handle payment, and confirm settlement."
If you have invested in MCP infrastructure, that investment is not wasted. It is foundational. The commerce protocols will build on top of the connectivity MCP provides. You are further ahead than organisations that have done neither.
Real-World Deployments
Santander and Mastercard: Europe's First Live Agent Payment
In early March 2026, Santander and Mastercard completed the first live end-to-end payment executed by an AI agent in Europe. The transaction was processed through Santander's live payments infrastructure, not a sandbox. The solution enables AI agents to initiate and execute payments on behalf of customers within predefined limits and permissions.
This is not a demo. It is a regulated bank processing a real agent-initiated transaction through production infrastructure. It validates that the operational and control framework works under real conditions.
OpenAI's ChatGPT App Ecosystem
Target, DoorDash, Instacart, and The Knot have all launched ChatGPT apps. Shopify and Etsy were among the first to use ChatGPT's checkout features in 2025. OpenAI's shift from in-chat checkout to app-based checkout reflects the reality that merchants want to control the transaction experience.
The move to apps means agentic commerce is not about one company owning the checkout. It is about building a discovery-to-merchant pipeline where the AI surfaces the right product and the merchant controls the conversion.
Stripe's Multi-Rail Expansion
Stripe's expansion of SPTs to support Visa agentic network tokens, Mastercard agentic network tokens, Klarna BNPL, and Affirm BNPL in early March 2026 is the clearest signal yet that agentic commerce infrastructure is moving to production at scale.
For any merchant already on Stripe, agent-initiated payments with multiple payment methods are now available without additional integration. That frictionless onramp will accelerate adoption faster than any protocol specification.
The Emerging Crypto Layer
On the decentralised side, Ethereum's ERC-8004 standard deployed to mainnet on January 29, 2026, establishing blockchain-native identity infrastructure for AI agents. Combined with Coinbase's x402 crypto-native payment standard, agents can verify counterparties and execute transactions without relying on a central platform. CoinGecko has already integrated x402 into its API, allowing AI agents to access data through pay-per-use without traditional subscriptions.
The crypto layer offers an alternative trust model. Where Mastercard and Visa build trust within controlled ecosystems, crypto builds trust through open protocols and on-chain verification. Both approaches will coexist, serving different use cases and risk profiles.
Implications and Analysis
The checkout is fragmenting, not consolidating. There is no single agentic checkout. OpenAI, Google, Mastercard, Visa, Stripe, and the crypto ecosystem are all building different approaches. The winners in the near term will be the companies that can operate across multiple protocols.
Trust is the bottleneck, not technology. The technology to execute agent-initiated transactions exists today. What does not exist is widespread consumer trust. Only 16 percent of US consumers trust AI to make payments. Closing that gap requires cryptographic proof of intent, auditable mandate chains, and clear liability frameworks. This is not a marketing problem. It is an infrastructure problem.
Product data is the new competitive moat. If your product attributes, descriptions, and availability signals are not structured for machine consumption, you will be invisible to AI agents. This is the most immediate action item for any commerce business.
Payment providers must embed in the protocol layer. Klarna and Affirm's sprint to support SPTs, AP2, and every other emerging protocol is not optional. AI agents default to the simplest available payment method. If your payment option is not programmatically available in the agent's decision environment, it will not be presented to the consumer. Period.
The owned agent advantage. Discovery on third-party platforms creates awareness. Owned conversational experiences create durable advantage. The feedback loop from owned agents, capturing decision context, preference signals, and conversion patterns, feeds every other part of the business. The most sophisticated operators will pursue both strategies simultaneously.
Which layers of the agentic commerce stack does your organisation currently operate in, and which represent your biggest opportunity or risk?
Key Takeaways
- The shift: Agentic commerce changes how decisions get made, not just where transactions happen. AI agents are becoming active participants in the global digital economy.
- The protocols: Five major protocol families are shaping the landscape: ACP, AP2/UCP, Agent Pay + Verifiable Intent, VIC + TAP, and SPTs. They operate at different layers and are complementary, not mutually exclusive.
- The stack: Discovery, Decision, Trust, and Payment form the four layers of every agentic commerce transaction. Your strategic position depends on which layers you own and which you depend on.
- MCP connection: MCP provides the connectivity layer that commerce protocols build on. Your MCP investment is foundational, not redundant.
- The trust gap: Technology is ahead of trust. Only 16 percent of US consumers trust AI to make payments. Closing this gap is the central challenge and central opportunity.