Overview
Open Standards for AI Commerce

Agentic Commerce Protocol: How Headless E-Commerce Becomes AI-Enabled

Commerce orchestrated by AI

What Lies Behind the Agentic Commerce Protocol

Headless paved the way: data and logic are decoupled, and frontends communicate with commerce and payment systems via APIs.

With the Agentic Commerce Protocol (ACP), a new layer emerges on top of this—a new digital touchpoint in e-commerce. AI agents don’t just conduct dialogues; they also orchestrate structured commerce flows all the way to payment. OpenAI defines ACP as an open standard that enables a conversation between customers, the AI agent, and the merchant up to the final purchase.

At its core is the clearly defined state space. It describes the complete checkout state and is returned by the merchant with every response. The agent does not rely on its own intermediate memory; instead, it displays exactly the state that the backend calculates. The source of truth lies with the merchant.

The ACP Stack at a Glance

ACP consists of three building blocks, each defined in separate specifications.

  • Product Feed Spec.
    Merchants provide structured product data, including prices, availability, media, and flags indicating search and checkout capability. Feeds are updated regularly.

  • Agentic Checkout Spec.
    ChatGPT acts as the AI agent and renders the checkout. The merchant provides REST endpoints to create, update, and complete a session, plus webhooks for order events. The full checkout state is returned with every response.

  • Delegated Payment Spec.
    Payment data is delegated to the merchant or their payment service provider as a single-use, scoped token. OpenAI is not the merchant of record; processing, refunds, and chargebacks remain with the merchant.

OpenAI and Stripe maintain the standard, which is published as open source under Apache 2.0. The current specification is publicly available on GitHub.

FAST INSIGHTS. REAL IMPACT.

Four Dialogues That Drive Sales

Guided Buying: From Need to Order

 

Example: “I’m looking for a jacket for my upcoming ski weekend, size M, budget 150 euros.” The agent asks for temperature range and water column requirements, presents suitable options, explains the differences, and checks availability. It then places the selected jacket in the cart and completes the checkout through dialogue. The result: fewer drop-offs and a higher cart value.

Service to Sales Instead of Returns

 

For size or color issues, the agent recommends a suitable alternative, checks inventory, and immediately reserves the new item. At the same time, it creates the return label and clarifies any payment differences within the conversation. This turns a return into a quick exchange. The result: more new purchases after returns and less effort for customers.

Making Complex Products Understandable

 

For variants and compatibility questions, the agent asks targeted follow-up questions, justifies recommendations using product data, and calculates concrete values — for example range, size, or consumption. This reduces uncertainty. The result: fewer mispurchases, less advisory effort, and clearer decision-making within the dialogue.

New Visibility for Assortments

 

Through the product feed, specialized or niche offerings also become discoverable—items that often get overlooked in a traditional shop. The agent prioritizes relevance instead of rigid search logic and presents products that match need, budget, and context precisely. The result: more targeted product presentation and better visibility for relevant assortments.

You describe the goal - the agent takes care of the rest. Understanding the context, selecting the product, booking the shipment, confirming the payment. ACP translates customer wishes into clean API calls to the headless commerce systems.

Julius Rabe
Lead Developer / Solution Architect

Why ACP Is Relevant for a Headless-First Approach

Anyone who has implemented a headless-first approach consistently is already prepared. ACP uses the same domain logic - only now, the consumer of the API is no longer a web or app frontend, but an agent. This changes the experience, not the core systems. For typical headless architectures, for example with commercetools and microservices for pricing, inventory, cart, and order management, as well as an integrated PSP, the following remains unchanged: catalog, cart, order, and payment. What’s new is the orchestration by the agent - an additional channel within the omnichannel setup.

FAST INSIGHTS. REAL IMPACT.

3 Essentials for Conversational Checkout

Catalog‑Governance

 

 

 

Feeds must be accurate, complete, and up to date. The interaction between the PIM, pricing logic, inventory, and content is crucial to ensure that the AI agent always receives reliable and current data.

 

 

 

Fresh, Complete, Consistent.

 

 

 

 

Risk and Fraud

 

 

 

Risks shift more heavily toward the checkout API. This requires clearly defined error states, signatures, and secure repeatability of all steps. Structured monitoring and appropriate playbooks should be part of every release definition.

 

 

 

Secure by Design.

 

 

 

 

Legal and CX

 

 

 

Legal requirements and customer experience flow directly into the conversational checkout. Terms and conditions, data protection, withdrawal, and returns must be correctly linked within the state and cleanly represented in the post-purchase flow.

 

 

 

Compliance Meets Conversation.

 

 

 

 

From Headless to Agentic

ACP feels like Headless 2.0. The coupling remains loose, only the consumer of the APIs changes from humans to agents. For existing API-first stacks, such as those built with commercetools, the path is manageable.

  • Set up the feed pipeline and cleanly export variants, prices, and availability.

  • Implement the three checkout endpoints plus webhooks, directly on carts, orders, and payments in commercetools.

  • Establish the PSP token flow and harden observability.

After that, customers can shop through dialogue without ever leaving the context. Headless laid the foundation; ACP turns it into an agentic experience.

Outro

Boundries and Roadmap

  • Availability is initially limited to the United States, starting with Etsy. Support for Shopify has been announced. International expansion and more complex shopping carts are planned.

  • The standard is developed openly and versioned. The specification is publicly available on GitHub and maintained as open source (Apache 2.0). Individual specifications and RFCs are currently marked as drafts.

  • Stripe is the first production implementation of the delegated payment spec. Additional PSPs have been announced, and PayPal has officially declared a partnership.

RELIABLE. SCALABLE. PROVEN.

Partnerships with Leading Platforms and Technologies

pixelart supports you across the entire value chain: data and PIM, headless and microservice architectures, Agentic Commerce Protocol, and conversational checkout.
We work with all relevant technology partners including commercetools, AWS, Pimcore, Magnolia, Stripe, and other providers that are essential for your setup.
This results in a solution that is technically robust, future-proof, and measurably effective.

 

Contact

We look forward to hearing from you.

Let’s talk about use cases, architecture, and a pragmatic starting point with your existing headless stack.

 

Julius Rabe

Julius Rabe, Lead Developer / Solution Architect

Mike Glas