Agentic commerce is in the midst of a defining moment. Instead of a customer navigating a checkout flow, AI shopping agents can now autonomously purchase goods, renew subscriptions, or restock supplies, executing payments entirely on the customer’s behalf through agentic payments protocols.
It’s no wonder that ChatGPT fields 2.5 billion prompts per day, with “purchasable products” accounting for roughly 2.1% (about 53 million) shopping queries daily.
This is a pretty dramatic change from how digital commerce has operated over the past two decades. It also raises a critical question:
How can financial transactions remain secure when AI starts spending for us?
This blog explores how agentic payments (transactions initiated and completed by AI agents) are changing the very core of digital commerce. Two major protocols are emerging to define this new frontier: OpenAI’s Agentic Commerce Protocol (ACP) and Google’s Agent Payments Protocol (AP2).
Agentic payments protocols on the rise
The payments infrastructure we rely on today is excellent at moving money efficiently and securely between people and businesses. However, agentic payments introduce an entirely new playing field, one that requires clear protocols defining what an AI agent can do, when it can act, and under whose authority it operates.
Merchants want access to these new agentic commerce channels, yet maintaining custom integrations for every ecosystem is unsustainable. A thriving AI economy depends on a shared payment foundation that avoids the fragmentation that once slowed the growth of web and mobile commerce. Three essential principles define the rules for this new foundation:
- Authorization: Proving that a customer explicitly granted the agent permission to make a purchase.
- Authenticity: Ensuring merchants can trust that an agent’s request accurately reflects the customer’s intent.
- Accountability: Defining who is responsible if a transaction is fraudulent or incorrect.
Two tech giants are racing to refine this new system: OpenAI’s ACP and Google’s AP2, each offering a distinct model for how AI agents securely authorize, execute, and record payments.
Skip the read and discuss a PoC straight away
OpenAI’s ACP prioritizes convenience, speed, and secure in-chat commerce
OpenAI’s ACP, developed with Stripe, is a community-designed, Apache 2.0 licensed framework that enables secure, frictionless transactions inside conversational platforms such as ChatGPT. It turns chat into a commerce channel, letting shoppers browse, compare, and buy without ever leaving the conversation.
With ACP, businesses maintain their customer relationships as the merchant of record, retaining control over which products can be sold, how they’re presented, and how orders are fulfilled. They use their own ACP-compatible payment service providers (PSPs) such as Stripe, so transactions flow through their existing payment systems, just like any other e-commerce checkout.
How shared payment tokens keep in-chat shopping secure
When an AI agent such as ChatGPT helps a customer make a purchase, it uses a delegated payment specification, such as a shared payment token with Stripe, to securely transfer payment details between the AI agent and the seller without exposing sensitive credentials.
Here is how the checkout flow works:
| Buyer | The buyer discovers a product or service through an AI interface, selects what they want to purchase, chooses a saved payment credential (or adds a new one), and grants permission for the AI agent to initiate checkout using a shared payment token. |
| AI agent | The AI agent manages the shopping experience, from discovery to checkout. Once the buyer confirms, the agent securely creates a shared payment token containing the authorization needed to complete the transaction and sends a checkout request to the business. |
| Business | The business receives the checkout request and shared payment token from the AI agent. It can accept or decline the transaction using its standard payment and fraud detection systems, being the merchant of record. |
| Payment provider | The AI agent communicates the shared payment token with the business’s PSP, such as Stripe, allowing it to charge the buyer securely. Each token has usage limits and expiry windows, ensuring controlled, auditable use. The PSP processes the payment like any other transaction, with no changes required to the business’s back-end systems. |
Buy in ChatGPT with Instant Checkout
With Instant Checkout, powered by ACP, ChatGPT users in the US can now buy directly from Etsy sellers, with support for Shopify merchants, including Glossier, Vuori, Spanx, and SKIMS coming soon. Shoppers can request, compare, and purchase items all inside ChatGPT. No redirects, no checkout forms, no friction.
Currently, ACP works within ChatGPT, but according to OpenAI’s developer guidance, businesses can use it to “transact with any AI agent and payment processor,” making it a scalable foundation for agentic commerce.
The protocol supports physical and digital goods, subscriptions, and asynchronous purchases, allowing businesses to publish their checkout configurations via standard APIs or through the Model Context Protocol (MCP).
Learn more about MCP and agentic commerce in our latest e-book
Agentic AI in composable commerceHow ACP supports agentic commerce that works for everyone
ACP creates a controlled and trusted commerce environment, setting the stage for next-level AI-driven payment automation, where customers, merchandisers, and payment providers win:
- Shoppers experience instant convenience from chat to checkout in seconds, without switching apps or filling out forms.
- Merchants reach millions of high-intent buyers by making their products and services available for purchase through AI agents, all while using their existing commerce infrastructure.
- Agentic AI platforms can embed commerce directly into applications, allowing customers to discover and transact directly with businesses, without being the merchant of record.
- Payment providers can grow their transaction volume by processing agentic transactions through secure payment tokens between buyers and businesses through AI agents.
Google’s AP2 prioritizes trust, verification, and interoperability
Google’s AP2 is an open, enterprise-grade framework designed to enable secure, interoperable transactions between AI agents, customers, merchants, and global payment networks. It establishes a trusted, auditable foundation for agentic payments, regardless of whether the underlying rail is a card, bank transfer, or crypto asset.
More than 60 organizations are already collaborating with Google to bring AP2 to life, including Adyen, American Express, Ant International, Coinbase, Etsy, Forter, Intuit, JCB, Mastercard, Mysten Labs, PayPal, Revolut, Salesforce, ServiceNow, UnionPay International, and Worldpay.
The protocol can be used as an extension of the Agent2Agent (A2A) protocol and Model Context Protocol (MCP). With the A2A x402 extension, developed in partnership with Coinbase, Ethereum Foundation, and MetaMask, AP2 expands beyond traditional card payments into the world of crypto and tokenized assets, paving the way for agent-based Web3 transactions.
At its core, AP2 is built to:
- Prove explicit customer consent for each purchase, not just general permission to spend
- Bind the agent’s request to the shopper’s true intent, minimizing the risk of “agent drift”
- Establish clear accountability for disputed or fraudulent transactions, providing the verifiable records needed for resolution
How AP2 establishes trust through mandates
AP2 uses mandates, tamper-proof, cryptographically signed digital contracts that define exactly what an agent can do on a customer’s behalf. For banks and regulators, this approach hits the right notes as it aligns with existing standards for digital contracts, signatures, and identity, extending proven frameworks into the world of autonomous transactions.
For example, a mandate might specify: “Book an all-inclusive vacation to Los Cabos under $2,000, only through these five travel partners, and within the next three months.”
Once authorized by the customer, the agent follows the mandate to complete the transaction through any compatible network or provider. Every action is logged, creating a full audit trail that documents precisely what the agent was permitted to do, building trust through verifiable accountability.
Mandates cover two primary interaction types:
| 1. Real-time purchases (human present) | When a customer says, “Find me a flight to Austin this Friday evening under $300,” the agent captures that request as an Intent Mandate, creating an auditable context for the interaction. After presenting flight options that meet those criteria, the customer’s approval signs a Cart Mandate, recording the exact airline, fare, and price, ensuring what’s approved is precisely what’s paid for. |
| 2. Delegated tasks (human not present) | For instructions like “Order my usual groceries every Sunday morning, but skip items that have increased more than 10% in price,” the customer signs an Intent Mandate upfront with detailed conditions. Once those parameters are met, the agent automatically generates a Cart Mandate and completes the transaction securely. |
This sequence “intent→cart→payment” forms a tamper-proof audit trail that anchors accountability and trust across every autonomous transaction.
How merchants and customers benefit from AP2
If AP2 sees broad adoption, it could normalize “hands-off” purchasing where an AI agent finds the right item, compares options, checks out securely, and completes the transaction without the customer needing to be present. This shift prioritizes convenience for customers and opens new engagement channels for merchants.
- Smarter, real-time shopping: Agents can continuously monitor inventory and execute purchases the moment a desired product becomes available within approved parameters. For instance, a restaurant’s procurement agent could automatically order sundried tomatoes the instant they drop below a target price.
- Context-aware promotions: Merchant agents can respond dynamically to customers’ intent with personalized offers and bundles. If a customer is shopping for a Halloween costume, for example, the agent can surface relevant promotions or time-sensitive deals before the event.
- Coordinated multi-merchant experiences: Agents can orchestrate entire purchase workflows across different vendors, such as booking a flight and hotel within a single budget. Once the best combination is found, the agent executes both cryptographically signed bookings simultaneously, ensuring accuracy and auditability.
ACP vs AP2: Where should enterprises place their bets?
OpenAI and Stripe are adapting existing digital commerce technology for transactions managed by AI agents, while Google and its partners are developing new infrastructure to support agentic payments at a broader scale.
Here’s how both the payment protocols compare:
| Feature | ACP (OpenAI + Stripe) | AP2 (Google + Partners) |
|---|---|---|
| Standard type | Open and community-driven (Apache 2.0) | Open, enterprise-grade framework for global payment networks |
| Payment method | Card, bank, or any ACP-compatible PSP | Card, bank transfer, crypto, and stablecoins |
| Agent control | Merchant-of-record model with customizable control | Cryptographically signed mandates for intent and cart authorization |
| Launch partners | Stripe, commercetools, Shopify, Etsy | Mastercard, AmEx, PayPal, Adyen, Coinbase, MetaMask |
| Ecosystem discovery | Open registry under active development | Backed by 60+ global partners, active rollout across industries |
| Compliance & security | Secure tokenization, traceable audit trails | Strong cryptographic mandates, full auditability, and regulatory alignment |
| Agent integration channel | ChatGPT Instant Checkout and other conversational platforms | Google ecosystem and enterprise partner integrations |
| Interoperability | Payment-agnostic and platform-flexible | Supports multiple payment rails, including blockchain and Web3 assets |
Enterprises may not have the luxury of choosing sides when trillion-dollar companies back both ecosystems. Merchants will need to integrate with both protocols, and AI agents will need to operate fluently across them. Interoperability will quickly become the new goal for competitive agentic commerce.
With deep expertise in AI orchestration, identity governance, and data observability, you can move from experimental pilots to production-ready agentic AI applications that you can pause, roll back, or retire mid‑workflow, control to always know what systems agents can access and under what permissions, and trace why an agent made a specific decision. We’ve already written the strategy guide for it, too:
To all enterprise tech leaders: Struggling to make agentic AI production-ready?
Start hereLooking ahead
Agentic commerce doesn’t end at the “buy now” button. Once trust protocols and payment standards are in place, agents will begin managing entire transaction cycles, including:
- Monitoring inventory and triggering restocks automatically
- Comparing supplier pricing and executing B2B purchases within budget limits
- Building personalized bundles and negotiating custom offers in real time
- Handling multi-step logistics, from booking shipments to managing returns
- Using analytics and context to optimize purchasing decisions continuously
Grid Dynamics is actively helping enterprises prepare for this next phase of digital commerce. Whether it’s ACP-style instant checkout or AP2-driven procurement, you get the strategy, architecture, and engineering talent to make it happen safely, securely, and ready to scale.
Book a discovery session with us to find the best-fit solution for your business requirements.
You might also like
