Agentic commerce in practice: modern stores serve people and AI

What is agentic commerce?

Agentic commerce is a purchasing model in which artificial intelligence, acting as an autonomous agent, completes successive stages of the buying journey on behalf of a user. The agent does not merely suggest products. It understands the intent behind a query, searches offers from multiple retailers, compares them against defined criteria, explains the recommendation, and then performs a transactional action: adding to cart, going through checkout, and initiating payment.

The shopping interface is no longer only a website or a mobile app. It becomes a conversation. A customer types or says, "I need waterproof trekking boots under 400 PLN, delivered before the weekend," and the agent does the rest. This is a shift comparable to moving from printed catalogues to search engines and then to voice assistants, except this time the effect is direct and transactional.

!CALLOUT

The key change: Your online store now has two equally important audiences: the human browsing your site and the AI agent reading your data through an API. If your products are not machine-readable, the agent will recommend a competitor whose data is better structured.

The market is consolidating around hubs

The agentic commerce ecosystem is forming around two types of platforms. The first are surface or agent hubs: places where AI agents talk to users and make purchase decisions. Today the dominant platforms are OpenAI with ChatGPT, Google with Gemini and Shopping Graph, and Perplexity with conversational search and native shopping flows. The second type are payment hubs: transaction infrastructure for AI-initiated purchases, led by Visa and Stripe.

More than 30 public partnerships announced in the first half of 2026 show the direction of travel. The market is consolidating around a handful of hubs that will handle a large share of agentic traffic. Shopify's direct integration with OpenAI gives Shopify merchants privileged access to the ChatGPT channel without every merchant building a custom integration.

Platforms that move slower enter through intermediate layers: protocol adapters, catalogue aggregators and API gateways. That works, but adds latency and data synchronisation risk. A Shopify merchant starts from a stronger position, but only if product data and store policies are ready for machine consumption.

The data layer AI needs

A shopping agent is only as good as the data it operates on. If your store does not provide the right information, the agent will skip your products or recommend them with incorrect details, creating orders you cannot fulfil. The data layer for AI has three parts.

A) Product data

• SKUs and variants as separate objects with their own identifiers, not values buried in descriptions.
• Comparable attributes such as material, dimensions, certificates, waterproof rating or ingredients.
• Real-time prices and availability so the agent knows the current state at the moment of the query.
• Shipping constraints including countries, regions, carriers, dimensions and delivery limitations.

B) Store truth

• Delivery policy: fulfilment times, costs, methods and free-shipping thresholds.
• Returns policy: time window, conditions, process and cost responsibility.
• Warranty and support: duration, scope and support contact details.
• Geographic limitations: supported countries, customs and regulatory constraints.

C) Execution interface

• Cart endpoint for creating carts and updating quantities.
• Shipping and tax calculation before the agent confirms the amount.
• Final amount confirmation before checkout.
• Order initiation and confirmation endpoint.

!CALLOUT

The most common mistake: Merchants assume that because they use Shopify, the data is already ready. In practice the issue is quality: variants written into product names, prices without tax context, policies hidden in PDFs. Agents do not read PDFs reliably.

UCP and ACP: why they exist and how they differ

As more AI agents entered e-commerce, a language problem appeared: how should an agent from one provider communicate with any store's commerce system? Open protocols standardise the interface between agent and merchant. In 2025 and 2026 two leading standards emerged.

UCP, Universal Commerce Protocol, is centred around the Google ecosystem. It is open, compatible with AP2, and Shopify co-creates it with Google as a key technology partner. UCP defines how an agent queries product catalogues, retrieves policy information and initiates transactions. Its strength is search: Gemini, Google Shopping Graph and Google Search increasingly use it as a native interface to external stores.

ACP, Agentic Commerce Protocol, is built by OpenAI and Stripe and available as Apache 2.0 open source. ACP is optimised for ChatGPT and the OpenAI ecosystem. It powers Instant Checkout in ChatGPT, with Stripe implementing the payment layer.

Both protocols solve the same problem: a standard language for catalogue, checkout and payment. UCP is Google-centric and stronger in search. ACP is ChatGPT-first and stronger in conversational commerce. A well-prepared store should be ready for both.

From question to order

The full transaction flow in agentic commerce can be divided into four stages.

1. Query to the agent: the user expresses purchase intent in natural language, from a precise request to a broad gift idea.
2. Fetching and normalising offers: the agent queries connected stores through UCP or ACP and normalises product data, prices and availability.
3. Shortlist with rationale: the agent filters results and explains why each product matches the criteria.
4. Checkout in AI or redirect to store: the user approves the selection and the agent initiates checkout through the protocol.

After the order is placed, fulfilment, returns and support remain with the merchant exactly as with traditional orders. The acquisition channel changes, not store-side operations.

!CALLOUT

Analytics consequence: If checkout happens inside an AI interface, GA4 sessions no longer reflect real acquisition. Attribution must track protocol source and order identifiers, not only browser sessions.

What to implement today: the minimum plan

Readiness for agentic commerce is mostly disciplined data work that is valuable regardless of channel. Implement it in three levels.

Level 1: clean product data

• Use structured product attributes instead of burying parameters in descriptions.
• Model variants as separate objects with their own SKUs, prices and stock levels.
• Keep prices current and clearly mark VAT or net/gross context.
• Synchronise availability in real time.
• Represent shipping limits as data, not prose.

Level 2: store truth

• Delivery times and costs as structured method-time-cost-threshold data.
• Returns policy with clear windows and conditions.
• Warranty information assigned to categories or products.
• Supported countries as a data array.

Level 3: protocol readiness

• Shopify native: monitor and activate Shopify Agentic Storefronts, Shopify Catalog and Knowledge Base.
• Custom or other platforms: build or buy an API layer for cart and checkout compatible with ACP or UCP.
• Prioritise based on whether OpenAI or Google can drive more traffic in your category.

Where Shopify fits in

Shopify has a privileged position in agentic commerce. It is one of the key partners building UCP with Google and one of the first major e-commerce platforms with native OpenAI integration through ACP.

Shopify's agentic stack includes:

• Shopify Agentic Storefronts: an interface layer that lets AI agents handle shopping sessions, carts and checkout through APIs.
• Shopify Catalog: structured product feeds optimised for AI consumption.
• Shopify Knowledge Base: machine-readable store policies such as delivery, returns and warranty.

Some features are currently most available to US and Canadian merchants. Europe will receive rollout waves with delay. That does not mean European merchants should wait. A store prepared today, with clean data and thoughtful policies, will enter the new wave ready instead of catching up later.

Risks: what can go wrong

Agentic commerce introduces a new category of operational risks.

• Hallucinations from missing data: if the agent cannot find structured information, it may infer or invent it.
• Price and availability drift: stale feeds can create orders for products that are unavailable or no longer discounted.
• Analytics outside the website: traditional analytics tools track browser sessions, while AI checkout may not create one.
• Geography and rollout: ACP and UCP features are not available everywhere at once, so delivery and checkout promises must be explicit.

Agentic commerce is not distant futurology. It is happening now. Stores that treat it as something to watch may soon lose discovery traffic to competitors with better structured data. If you want to assess your readiness, talk to us.