Agentic Commerce and Emerging AI Shopping Protocols

To understand this shift, we need to look at the Digital Concierge concept. In agentic commerce, an AI can basically handle the whole shopping process for you: finding stuff, checking out options, and finishing the purchase with hardly any effort from your side.

So, picture this: you just tell your AI assistant, Book me a nonstop flight to London next week for less than $600, and no overnight flights, please (make sure you say please because you never know🤫). The agent then goes and finds the perfect flight that meets the criteria, buys the ticket using your stored payment info, and sends the confirmation.

We're moving past the old way of e-commerce, where AI mostly just suggested stuff, to a totally proactive system where AI agents basically take the reins, planning, deciding, and making purchases for the user.

However, for this Digital Concierge to work at scale, we face a major technical challenge: How do we get millions of different online stores to speak the same language as these AI agents? The industry is solving this through a new stack of protocols.

If a travel agent had to learn a totally new language for every airline, they'd be completely stuck. That's essentially the problem AI agents face trying to make sense of product data across different websites.

The Universal Commerce Protocol (UCP) is a new open standard designed to enable seamless agentic commerce across different platforms and services. Launched by Google in collaboration with industry partners, UCP establishes a common language and set of functional primitives that enable AI shopping agents, retailers, and payment providers to interoperate more easily.

Instead of each AI agent integrating separately with each merchant or platform, UCP provides a universal bridge that allows all agents to interact easily with all businesses. UCP is open-source and built to work with existing retail infrastructure – it’s compatible with related protocols like Agent2Agent (A2A) for agent-to-agent communication, the Agent Payments Protocol (AP2) for payments, and the Model Context Protocol (MCP) for sharing context.

Practically, how is UCP used? One early application is powering instant checkout on Google’s AI-driven search results. Let’s say you use Google's new AI Mode in Search (or the Gemini app) to check out a product. UCP lets you buy directly from a qualifying retailer without leaving the search screen. You can just hit "Buy" on a product result and check out instantly using stored payment and shipping info (like Google Pay or PayPal). No need to trouble yourself with going to the retailer's website. The retailer is still the seller and can tweak the integration to work with their systems, but they get to snag those high-intent sales immediately and reduce abandoned carts thanks to this super slick, AI-driven checkout process.

Going forward, UCP is expected to expand internationally and enable richer capabilities (such as applying loyalty rewards or suggesting related products) within AI-assisted shopping experiences.

Once your agent can "read" the store (thanks to UCP), the next big issue is trust. If you tell an AI to buy something while you're offline, how does the merchant know that the AI is actually allowed to spend your cash?

This is where Agent Payments Protocol (AP2), an open protocol introduced by Google in 2025, comes to light. AP2 provides a standardized, payment-agnostic framework that allows AI agents to transact on your behalf across different payment methods and platforms. Up until now, in e-commerce, we just assumed that a person hits the Buy button. With AI agents, though, we need solid extra assurances of trust and authorization. AP2 is set to handle things like:

• Authorization: ensuring the AI agent has explicit user permission for a given purchase.
• Authenticity: allowing the merchant to verify that the agent’s request reflects the user’s true intent (no unauthorized or altered orders).
• Accountability: establishing clear responsibility and audit trails in case of fraud or errors in an agent-initiated transaction.

To solve the Trust Gap, AP2 uses cryptographically signed mandates as proof of the user’s instructions. For example, when you instruct an AI agent to buy something, your request is captured in an Intent Mandate (a signed digital contract detailing what you asked for). Once the agent selects the item(s) and you approve, a Cart Mandate is signed, locking in the exact items and price. Together, these create a tamper-proof record linking your intent to the final order, forming a verifiable audit trail from request to payment.

This mechanism supports both real-time purchases (you’re present to approve the cart) and delegated purchases (the agent executes later under pre-set conditions). Notably, AP2 is designed to work across many payment types – from credit cards to bank transfers to digital currencies.

AP2 opens up cool new shopping experiences that just weren't possible before. Imagine you telling the AI agent, I really want this winter jacket in green, and I'm okay with paying up to 20% extra for it. If it's out of stock, the agent will keep an eye on the inventory and price, then automatically seal the deal with a secure purchase the second that green one pops up within the price limit.

This hands-free shopping means you won't miss out on what you want. Plus, it's totally safe thanks to AP2's rules—the agent can't spend more than you have pre-approved. AP2 could also let an agent handle something more complicated, like booking flights and hotels within a set budget, by chatting with different providers and finalizing the deal only after everything lines up.

Finally, we gotta figure out how to link the chat (where the customer is talking) with the merchant's checkout (where the stuff is kept).

This is the role of the Agentic Commerce Protocol (ACP). In 2025, OpenAI (working closely with Stripe) rolled out a cool feature: users can now find a product through ChatGPT and buy it right there in the chat. The whole idea behind ACP is to define the secure, consistent way AI agents, shoppers, and businesses talk to each other to get a purchase done.

While AP2 is all about handling the tough, hands-off autonomous stuff, ACP is currently focused on making the Human-in-the-Loop experience as smooth as possible. For example, you might ask ChatGPT for product suggestions (e.g., "best running shoes under $100"). ChatGPT will show relevant products, and if a product supports Instant Checkout via ACP, a "Buy" button appears. You can click Buy, then confirm the order details, shipping, and payment, all without leaving the chat interface.

In this flow, ChatGPT is essentially a smart middleman: it orchestrates the exchange but does not process payments or fulfill orders itself, so merchants retain control over inventory, payment processing, and customer relationship (they remain the merchant of record).

Merchants don't need to overhaul their backend to participate. If a merchant already uses Stripe, enabling agentic payments can be as simple as adding one line of code. However, a Merchant Reality Check is needed here: While the payment connection is one line of code, the AI still needs clean data (size, color, stock) to recommend the product correctly in the first place.

While protocols like UCP and ACP provide the rules of the road for AI agents, they do not solve the underlying logistics of the transaction. The biggest bottleneck for Agentic Commerce isn't the AI, it's the data.

For an autonomous agent to close the loop (searching, comparing, and buying), it needs to read product details, check real-time stock, and understand complex variations (for example, distinguishing a waterproof shell from an insulated jacket). Legacy monolithic commerce platforms, built primarily for human eyes and visual HTML pages, often trap this data in unstructured blobs that AI agents struggle to parse.

This is where headless commerce and Product Information Management (PIM) platforms like Crystallize become the essential engine of the Agentic economy. Because Crystallize is built for APIs and structured data rather than static pages, it is inherently Agent-Ready.

This isn't just theory. Want to see this in action? Check out our case study on Composable Architecture and AI. It shows exactly how structured data gives AI the context it needs (totally beating the unstructured blob problem) and turns being technically ready into a real-world edge over the competition.

UCP, AP2, and ACP might sound like competing standards (like VHS vs. Beta, do you remember those!?). But they are actually complementary layers of the same future. UCP is compatible with related protocols, such as Agent2Agent (A2A) for communication and the Agent Payments Protocol (AP2) for payments.

Let's be real: shopping is shifting from just browsing to telling an AI exactly what you want. Instant Checkout and ACP are connecting people, AI agents, and businesses to make buying as easy as a quick chat. That's agentic commerce in action, and it keeps the important trust and roles for both merchants and customers.

You'll want to nail the basics, like content modeling, and make sure the platform supporting your commerce initiatives is good to go for agents—you know, like Crystallize😎

SCHEDULE A 1-on-1 DEMO to see how Crystallize can help you with agentic commerce. Or, why not START building for FREE.