Agentic Commerce
for E-Commerce Merchants

AI agents are taking over product discovery, advice, and checkout. What this means concretely for online shops — and what you should be preparing now.

Agentic Commerce — AI agent between customer and online shop
Why now

Three movements reshaping the sales channel — verified from public platform roadmaps and our own ongoing build.

March 2026
Shopify ACP endpoint live
Shopify ChatGPT Integration · Public Announcement
June 2026
Shopware agent features announced
Shopware Roadmap 2026
Q1 2026
MEMOTECH Agentic Commerce platform in build
memotech.ch Blog · Open-Source Commerce-AI

The shopping context is shifting faster than most Shopware merchants notice day-to-day. Today customers still type "Which hiking boots keep water out at freezing temperatures?" into a search engine and click. Tomorrow they ask Claude, ChatGPT or an assistant embedded directly in the storefront — and it decides which shop delivers the answer. This is not the next search evolution. This is Agentic Commerce.

What is Agentic Commerce?

Agentic Commerce describes shopping interactions where an AI agent — an LLM with tool access — sits between the customer and the shop. The agent understands natural-language queries, searches product catalogs semantically (vector embeddings instead of keyword match), aggregates offers across shops and, if needed, executes the checkout itself. Behind it stand open protocols such as the Agentic Commerce Protocol (ACP) — driven by OpenAI and Stripe — and UCP for checkout handoff. Shopify has had an ACP endpoint since March 2026. Shopware has announced comparable features for June 2026.

Why does this hit e-commerce merchants now?

Short version: because visibility is tipping. Three shifts are accelerating it:

  • LLM traffic is growing exponentially. ChatGPT Search, Perplexity, Claude Projects — consumers are increasingly delegating product research to agents.
  • Shopping widgets are becoming default UX. Embedded on the product page, in the header, in chat — AI answers service questions and sells within the same dialogue.
  • Conversion data is shifting. Semantic search and contextual recommendation are replacing facet navigation. Merchants without structured product data lose to merchants with it.

For online shops this means: the product catalog has to be machine-readable (clean taxonomy, structured metadata), the service FAQ has to be indexable, and a customer-facing AI layer belongs in the storefront — not in a separate support channel decoupled from conversion.

Where MEMOTECH stands

MEMOTECH has been building an Agentic Commerce platform since Q1 2026 — Swiss-based, Shopware-first, with Magento and Shopify connectors in preparation (Q4 2026). Apache 2.0 platform foundation, LLM stack via NVIDIA Connect Program . Technical details and measurements published openly on the MEMOTECH Blog (English-language deep dives). If you want to understand the shift before it reaches your own conversion, the material lives there.

What an Agentic Commerce platform has to do

Four architecture principles against which every commerce-AI solution should be measured.

Multi-agent instead of monolith

Specialized agents with clearly scoped roles (e.g. search, recommendation, service FAQ, post-purchase, demand forecast) perform more reliably than a single chat assistant attempting everything at once.

Open protocols

ACP (Agentic Commerce Protocol) and UCP are the emerging standards. Platforms that support both stay connectable to external agents — proprietary SaaS APIs do not.

Platform-native, not parallel

Product catalog sync, storefront widget, and FAQ indexing dock cleanly onto the existing shop infrastructure — instead of creating duplicate data and broken checkouts. Shopware 6 is live today; Magento and Shopify connectors are in preparation (Q4 2026).

Tenant isolation as architecture

Customer data from different shops must never touch inside the vector index. Physical separation at the DB level rules out cross-shop leaks — filter logic at the application layer is too fragile.

In practice

Six use cases that matter in the next twelve months

Concrete agent capabilities that decide whether e-commerce merchants stay visible — or get filtered out of the next generation of search.

Semantic product search

"Hiking boots that keep water out at freezing temperatures" finds the right product — even without exact keywords. Vector embeddings instead of facet navigation.

Recommendation agent

Contextual recommendation in checkout instead of static "also bought" boxes. Considers cart, season and availability.

Service FAQ agent

Shipping, returns, size charts answered directly inside the storefront widget — no support ticket needed. Indexes the existing FAQ.

Post-purchase agent

Order status, tracking, returns as a conversational flow — relieves the service team and keeps the customer in the brand context.

Demand forecast from queries

Query patterns become signals for sourcing and bundle pricing. Visibility into what customers are looking for — before they buy.

Cross-shop bundle

Multi-storefront setups (e.g. B2B + B2C) sharing one agent layer. Consistent answers across every brand touchpoint.

What we build on

Open standards, documented architecture

Platform foundation on Apache 2.0. We rely on open protocols (ACP, MCP) and tool-neutral components — no SaaS black-box APIs, no proprietary data formats.

NVIDIA Connect Program
LLM inference stack
OpenAI ACP
Agentic Commerce Protocol
Stripe
UCP / Payment Layer
Shopware 6
Storefront integration (live)
Magento / Shopify
Connector roadmap Q4 2026
MCP
Model Context Protocol
Apache 2.0
Open-source foundation

Comparison of the six protocols relevant to agentic commerce — spec and repo linked directly.

As of: May 2026 — snapshot, refreshed annually.

ProtocolInitiatorPurposeCore capabilitiesStatusShopware relevanceSpec / Repo
AXPAgentic Commerce Lab (Hamann-led)Agent-experience layer for storefrontsProduct Data · Quality Signals · Experience EmbeddingDraft / RFCHigh — Shopware-native
A2AGoogle (a2aproject)Agent-to-agent communicationAgent Discovery · Message Routing · OSSBeta, OSSMedium — multi-agent setups
ACPOpenAI + StripeAgent-to-merchant transactionCatalog Exposure · Cart API · Checkout HandoffStable, Shopify live since March 2026High — Shopware roadmap
UCPUniversal Commerce Protocol Working GroupUniversal checkout handoffCheckout Sessions · Identity Linking · Order ManagementBetaHigh — direct checkout
AP2Google CloudAgent payment authorisationPayment Auth · Tokenisation · Google-Pay flowBeta (Cloud blog March 2026)Medium — Google-Pay flows
StoreSyncPayPalAgentic shopping via PayPal Commerce PlatformProduct Discovery · Cart Management · Checkout CompletionProprietary, PayPal Commerce PlatformSelective — for PayPal-driven stores

Adjacent: NANDA (MIT — agent discovery, early research) and x402 (Coinbase — Web3 payments) sit outside the Shopware mainstream and are not listed here.

Decision frame

Which tier fits your shop?

The decisive question for 2026 is not „which vendor" but „which tier". Three paths, depending on revenue, IT maturity and strategic requirements.

Tier 1

SaaS-First

CHF ~1'000–10'000/month

Shops with CHF 500k–10M revenue, standard assortment, no differentiation argument via AI depth. Time-to-market beats vendor lock-in. Vendors: Algolia, Klevu, Nosto, Bloomreach, Shopware AI Copilot.

2–6 weeks integration + monthly SaaS fees

Request SaaS recommendation
Tier 2

Audit + strategic decision

One-time CHF 980–7'200

Merchants unsure whether Tier 1 is enough or Tier 3 is needed. Before six-figure investments: structured analysis of data, conversion paths and use-case fit — platform-agnostic.

Written tier recommendation with sources, methodology and risk annex

Request slot or audit
Tier 3

Custom agentic platform

Initial range scope-dependent + inference costs

Merchants from CHF 10M revenue with in-house IT, sovereignty requirements (compliance, brand risk) or genuine differentiation via AI depth (B2B configuration, regulated industries).

3–6 months initial build, multi-agent stack on ACP/UCP/MCP, Apache 2.0 foundation, sovereign inference

Request architecture call

Unsure which tier? The free PoC below is the free-tier variant of the Tier 2 audit — deliverable today for Shopware 6 merchants.

Free 2-week PoC

Request a slot — free PoC for Shopware 6

4–6 slots per quarter on our NVIDIA Connect stack. We bring 20–25h of senior engineering per PoC; you bring a catalog snapshot and 2–3 hours of onboarding time. 48-hour reply with slot confirmation, waitlist or honest rejection with reasoning.

Request slot or waitlist

Magento, Shopify and custom platforms: connector roadmap Q4 2026 — waitlist entry available in the same form.

MEMOTECH Newsletter

Agentic Commerce updates straight to your inbox

New articles and measurements from commerce-AI practice. GDPR-compliant, unsubscribe any time.

Looking for the full archive? All articles on the MEMOTECH Blog →

Agentic Commerce for E-Commerce Merchants | MEMOTECH