K5 2026, Day 2: The Field Agrees Agentic Is Coming. Who Controls the Experience Layer?

K5 in Berlin is over. Two days on the show floor, and from the stages, booth conversations and threads a clear picture reads out: the field has settled on a direction. Agentic commerce is coming, and hardly anyone is still arguing the point. What became clear on the final day is not the question of whether autonomous discovery and AI agents will change commerce, but how far the industry still is from the on-stage vision once you listen closely.

This is our wrap from what we took away on the show floor and in the K5 conversations. Three sober observations instead of hype, and one open question that no stage really answered.

Observation 1: Backend Autonomy Is Maturing Faster Than Experience Control

Most of the agentic substance shown at K5 sits in the backend: pricing agents, promotion logic, fulfillment automation, agentic discovery over structured product data. That is real, and it is maturing fast. commercetools, for instance, marked where the backend side is heading with its agentic jumpstart focus at the booth.

Experience control, meaning what the customer actually sees and operates, lags behind that backend maturity. It is easier to let a pricing agent compute autonomously than to define how autonomous decisions arrive cleanly, on-brand and controllable in the storefront. That gap was tangible on the show floor.

Observation 2: Agent-Ready Means Clean Data First for Merchants

When a merchant asks what agent-ready concretely means for them, the honest answer is unspectacular: are your product data structured enough for an agent or an AI answer engine to read them reliably at all? Before the question of autonomous frontends comes the question of valid Schema.org markup, deterministic rendering and consistent product feeds, the basis for being cited by AI answer engines at all. How a storefront sets itself up for that is on our SEO and GEO page.

These basics are not a stage topic, but they are the prerequisite for everything else. A storefront that does not meet them simply will not be cited by AI systems, no matter how agentic the backend behind it is. We laid out what that looks like on the frontend layer in our parallel post on the composable stack in the DACH mid-market.

Observation 3: The Experience Layer Stays the Open Governance Question

When agents take over parts of discovery and personalization, a new question emerges that became visible at K5 but stayed unanswered: who controls the layer where human and agent work together? Marketing wants brand consistency and control, the agent wants autonomy, and both ultimately operate the same component library in the storefront. That is neither a pure backend problem nor a pure AI question, but a question of experience governance.

This is exactly where the idea of a Frontend Management Platform (FMP) comes in: a control layer where human designers and AI agents operate the same component library, with clear rules about who is allowed to do what. We raised this observation before the event in our frontend questions for K5, and the final day confirmed it rather than refuted it.

What We Take Away from K5

K5 marked the direction, it did not reach the destination. Agentic commerce is consensus, but the implementation is at the beginning, and it gets decided not in the backend but at the experience layer. Whoever thinks ahead now takes care of clean data, deterministic rendering and a frontend layer built for the collaboration of human and agent. We drew the broader industry arc on this once before after Shoptalk Europe, in our post on agentic commerce and the composable frontend.

We will keep following the K5 topics over the coming days. If you want to read along on the discussion around agentic commerce and the experience layer, you will find more on our approach on the Agentic Frontend Management Platform.