Agentic commerce 2026: how AI agents shop in your online store

Agentic commerce 2026: how AI agents shop in your online store

TL;DR — the essentials in 30 seconds

• Agentic commerce means an autonomous AI agent researches, compares and buys on a customer's behalf, without the person clicking through the shop themselves.
• Four protocols shape the market in 2026: the Agentic Commerce Protocol (ACP) from OpenAI, Stripe and Meta, Google's Agent Payments Protocol (AP2) (now with the FIDO Alliance), Visa Trusted Agent Protocol (TAP) and Mastercard Agent Pay.
• With Instant Checkout, ChatGPT users in the US can already buy directly in the chat from Etsy sellers and a growing number of Shopify stores.
• Since its Community Day 2026, Shopware has explicitly positioned itself as open commerce infrastructure for the “agentic era” and is turning its Copilot into an assistant that takes action.
• Visibility to AI agents requires structured product data based on Schema.org, machine-readable feeds and a checkout that can connect to open protocols.

Definition:
Agentic commerce refers to purchases that an autonomous AI agent researches, compares and completes on a customer's behalf, instead of the customer navigating an online shop themselves. Open protocols such as ACP and AP2 govern how product data, purchase intent and payment authorization are exchanged between agent, merchant and payment network.

A customer types into ChatGPT: “Find me running shoes for marathon training, budget €150, delivered by Friday.” The agent compares products, builds a cart and completes the purchase right in the chat, without the customer ever visiting the shop. That's agentic commerce: the shift from a person clicking through a shop to an agent acting on that person's behalf. It's based on the concept of agentic AI, in which an AI system plans and carries out multi-step tasks on its own instead of just giving individual answers. For online retailers, this shifts a core assumption of the last 25 years of e-commerce: visibility no longer comes only from your own shop interface but also from machine-readable data that an agent understands without ever rendering the page.

How does a purchase by an AI agent work technically?

An agentic commerce purchase follows a clearly defined five-step process: discovery via structured product data, authorization through an intent mandate, cart and checkout, payment authorization via a payment network and, finally, fulfillment confirmation via webhook.

In the first step, the agent doesn't read product data from the visual shop interface but from feeds and Schema.org markup: price, availability, variants, return conditions. The agent then records the user's purchase intent in a signed mandate, such as budget and selection criteria. Only then does it build the cart and start the checkout session directly with the merchant's system. Payment authorization runs through a payment network that verifies the agent's identity before releasing a payment token. Finally, order webhooks send shipping and delivery status back to the agent and the customer.

‍

‍

For merchants, this means the key touchpoints are shifting from the product page to API endpoints and structured feeds. If you don't provide this data cleanly, you simply don't exist for an AI agent, no matter how well your website is designed.

Which protocols govern agentic commerce in 2026?

Four protocols share the market, and none of them covers the entire process alone. The Agentic Commerce Protocol (ACP) was developed by OpenAI and Stripe and is now also backed by Meta. It governs cart, checkout session and payment handover and is the technical basis for Instant Checkout in ChatGPT, which, according to Stripe, already lets US users buy from Etsy sellers and a growing number of Shopify stores such as Glossier or Spanx.

Google's Agent Payments Protocol (AP2) takes a broader approach: it represents every agent purchase through three signed mandates – an intent mandate for the purchase request, a cart mandate for the assembled cart and a payment mandate for payment authorization. Over 60 partners, including Mastercard, PayPal, Adyen and Coinbase, already support AP2. Since May 2026, its further development has been with the FIDO Alliance, which has set up its own working groups for agent authentication and payment standards.

Visa and Mastercard have built their own network-specific additions. The Visa Trusted Agent Protocol (TAP), developed with Cloudflare, solves a specific problem: it helps merchants distinguish genuine purchase-intent traffic from an agent from malicious bot traffic, so legitimate agent requests aren't wrongly blocked. Mastercard Agent Pay takes a similar approach on its own card network infrastructure, including in a pilot with PayPal.

‍

‍

In practice, this means most merchants won't integrate any of these protocols directly themselves. The work happens at the level of payment providers like Adyen, Stripe or Fiserv, which embed TAP, AP2 and ACP into their existing integrations. What matters for you as a merchant is whether your payment provider and shop system already support these protocols – not whether you implement them yourself.

What does this mean for your online shop?

The big shop systems are already responding visibly. At the Shopware Community Day 2026 in Cologne, Shopware explicitly positioned itself as “Open Commerce Infrastructure for the Agentic Era” in front of more than 1,700 merchants, partners and developers. The Shopware Copilot is being developed from an advisory assistant into an agentic system that carries out tasks in the administration on its own. At the same time, the Shopware merchant survey 2026, which polled around 360 German online and omnichannel retailers between February and March 2026, shows that 44 percent plan targeted investments in artificial intelligence. Which levels of automation are already natively available for Shopware stores today is summarized on the Shopware integration page.

In concrete terms, this shifts the priority list for IT managers in mid-sized companies. Until now, search engine optimization for keywords came first. Now a second level is being added: whether an AI agent can find your products at all, understand them and add them to a cart. That concerns structured product data, stable product APIs and a checkout that doesn't necessarily require a graphical interface.

Agentic commerce and shop agents: where's the line?

Agentic commerce and a shop agent in your own store solve different problems and aren't mutually exclusive. With agentic commerce, the customer never even enters your shop, because an external agent like ChatGPT or Gemini handles the purchase entirely outside your platform. A shop agent, on the other hand, like the one melibo offers for online shops, sits directly in your own frontend and advises customers while they're still with you: it answers product questions, recommends suitable items and guides the customer all the way to your own checkout.

These two levels complement each other. A shop agent keeps customers in your own buying process longer and provides exactly the structured, well-maintained product data that external shopping agents also need for discovery. So if you're investing in conversational product advice anyway, you're also building the data foundation needed for visibility in ACP- or AP2-based purchase processes. How AI agents fundamentally differ from classic chatbots and how to get started in your company is explained in AI agents in service: how they change processes – and how to get started. You'll find more on how AI agents work in customer service and sales in the melibo platform overview on AI agents.

How do you prepare your shop for agentic commerce?

The first and most important step is structured product data. Price, availability, variants, shipping costs and return conditions must be available as Schema.org markup, not just as running text on the product page. Without this foundation, no agent – whatever the protocol – can reliably read your product range.

The second step concerns your technical infrastructure: check with your payment provider and shop system whether and when support for ACP, AP2 or TAP is planned. These integrations are already underway at Shopware and Shopify; for smaller or custom-built shop solutions, it's worth asking your payment provider early.

The third step is strategic: build out your own conversational advice in the shop in parallel, for example with a shop agent like melibo's, which actively keeps customers in your own buying process instead of losing them to an external agent. If you combine both levels – cleanly structured data for external agents and a strong advisory experience of your own – you secure visibility on two fronts at once.

The fourth step is monitoring: track how often and in what position your shop appears in answers from ChatGPT, Gemini or Perplexity to relevant purchase prompts. AI search tracking tools like Peec AI make this visibility measurable, much like classic SEO tools do for Google rankings.

What risks and open questions remain?

The protocol landscape in 2026 is deliberately fragmented, and that will stay the case for now. Mastercard has publicly confirmed it is working on practically all relevant protocols at the same time, which shows that the industry itself expects a multi-protocol environment rather than a single winning standard. For merchants, this means extra coordination across several payment partners.

Liability questions are also unresolved: who is responsible if an agent buys a wrong or unwanted product within its mandate? AP2's signed mandates and ACP's checkout sessions are meant to create exactly this traceability, but practical disputes and case law on this are still pending in the DACH region.

For German merchants, there's also the familiar compliance dimension. As soon as an agent processes personal data, the GDPR applies under Art. 5 and 28 GDPR, as with any other system – something the article How to use chatbots in your company in a GDPR-compliant way covers in detail. And as soon as an AI system interacts with a person, the transparency obligation under Art. 50 EU AI Act (Regulation 2024/1689) applies: users must be able to recognize that they're dealing with an AI system. This affects both your own shop agent and the question of how transparently external shopping agents identify themselves to your shop system.

Conclusion

In 2026, agentic commerce is no longer a distant vision but a technical stack that's already running: ACP, AP2, Visa TAP and Mastercard Agent Pay are live, Shopify catalogs with millions of products are connected to Instant Checkout, and Shopware is actively turning its Copilot into an assistant that takes action. For online retailers in the German mid-market, this doesn't mean integrating all four protocols yourself right away. It means doing the homework now that counts regardless of which protocol wins: clean, structured product data, a checkout that can be made protocol-ready, and strong advice in your own shop that keeps customers before they delegate the purchase to an external agent. If you bring both together, you're positioned for both paths to the customer at once.

‍

‍