In May 2025, Stefan Hamann - CEO of Shopware, one of Europe's leading open-source eCommerce platforms - published what he called a "cheat sheet" for eCommerce leaders preparing for agentic commerce. The document is significant not because it is revolutionary - it is not - but because it represents the first comprehensive public statement from a major platform vendor on what merchants need to do to prepare for the shift from human-driven to agent-driven commerce. It is a readiness playbook written from the infrastructure layer. And like all infrastructure-layer perspectives, it sees certain things with extraordinary clarity while remaining structurally blind to others.
The AXD Institute has spent eighteen months building the conceptual architecture of Agentic Experience Design - the discipline concerned with how humans delegate, calibrate, observe, interrupt, and recover trust in autonomous AI systems. From this vantage point, the Shopware cheat sheet is a fascinating document. It confirms that the operational layer of the agentic commerce stack is maturing rapidly. It demonstrates that platform vendors are beginning to think seriously about what their merchants need. And it reveals, with inadvertent precision, the structural gap between operational readiness and the trust architecture that determines whether humans will actually delegate to agents in the first place.
This essay is not a critique of Shopware's work. It is an analysis - an attempt to read the platform vendor's playbook through the AXD lens and to identify where the merchant's readiness framework maps to the Five Pillars of AXD Readiness, where it excels, and where it requires the structural completion that only trust architecture can provide.
01 - The Cheat Sheet and the Discipline
The Shopware Agentic Commerce Cheat Sheet opens with a definition that would be familiar to any reader of the AXD Observatory: agentic commerce is "online shopping tasks performed by AI agents on behalf of users - from product discovery all the way to checkout." The document correctly identifies the fundamental shift: this is not another iteration of online shopping. It is a change in who performs key actions. The customer no longer navigates, compares, and purchases. The agent does.
Hamann's framing is instructive. He positions agentic commerce as "a leap similar to the shift from physical to online shopping, or from online to mobile, but now from manual to autonomous." This is the correct magnitude of comparison. What the AXD Institute would add is that the previous shifts - physical to online, online to mobile - changed the channel through which the same human actor made decisions. The shift to agentic commerce changes the actor. The human is no longer the decision-maker at the point of transaction. The machine customer is. This distinction - channel shift versus actor shift - is the reason that operational readiness, while necessary, is not sufficient.
The cheat sheet marshals compelling evidence of momentum. Perplexity AI's partnership with PayPal enables direct purchases within conversational interfaces. Visa's "Intelligent Commerce" rollout introduces AI-ready payment credentials and agent-specific spending controls. Enterprise pilots surged from 37% to 65% in a single quarter of 2025. Two-thirds of consumers expressed interest in using AI agents for limited-stock purchases and price-triggered auto-buying. These are not speculative projections. They are operational realities that confirm the timeline the AXD Institute has been tracking since its founding in September 2024.
Where the cheat sheet and the discipline diverge is not in their assessment of what is happening, but in their analysis of what it requires. The platform vendor sees infrastructure. The AXD practitioner sees infrastructure and architecture. The difference is not semantic. It is structural.
02 - What the Platform Vendor Sees Clearly
Credit where it is due: the Shopware cheat sheet sees the operational layer with remarkable clarity. Its core insight - that agentic commerce requires merchants to rethink "how we present and syndicate product information, and how we build trust in AI-driven decisions" - is precisely correct. The document identifies three operational imperatives that align directly with the AXD Institute's analysis.
First, data quality becomes existential. When a human shops, they can compensate for missing product information through visual inspection, reviews, or intuition. When an agent shops, missing data means the product does not exist. An agent searching for "a red running shoe, size 11, under $100" will not select a product whose listing lacks colour or size attributes. The cheat sheet's emphasis on "Data Fill Rate" - the completeness of product information for AI consumption - is the operational expression of what the AXD Institute calls Signal Clarity: the quality of the information environment in which agents operate.
Second, speed becomes structural. In a world where AI agents scan dozens of merchants simultaneously, the latency between a price change in your backend and that change reaching the agent's decision environment is not a technical detail - it is a competitive variable. The cheat sheet's "Update Speed" metric captures this reality. A merchant whose inventory feed updates hourly will lose to a merchant whose feed updates in real time, not because the second merchant's products are better, but because the agent never sees the first merchant's current offer.
Third, API-first architecture becomes non-negotiable. The cheat sheet repeatedly emphasises that merchants need "open, API-ready" systems - platforms that can expose real-time inventory, pricing, and ordering functions to external AI services. This is the infrastructure prerequisite for agentic AI protocols. Without machine-readable, real-time APIs, a merchant is invisible to the agent economy. Our guide to machine-readable commerce details the practical steps for achieving this readiness, and the AXD Institute's analysis of the merchant's stack maps the complete technology architecture that merchants must build to serve autonomous agents. The platform vendor sees this clearly because it is, fundamentally, a platform problem.
03 - The Four KPIs and What They Measure
The cheat sheet proposes four core KPIs for measuring agentic commerce readiness. Each is well-defined and operationally sound. Each also reveals, through what it measures and what it does not, the boundaries of the platform vendor's perspective.
Data Fill Rate measures the completeness of product information available for AI consumption - descriptions, specifications, pricing, inventory, images, metadata. A 95%+ fill rate means an agent can confidently understand and compare your products. In AXD terms, this maps directly to Signal Clarity - the first pillar of the AXD Readiness Assessment. The mapping is almost exact: both measure whether the information environment is rich enough for autonomous decision-making.
Update Speed measures the latency between a backend change and that change reaching all consumer touchpoints, including agent interfaces. This maps to Reputation via Reliability - the second AXD pillar. An agent that encounters stale data from a merchant does not merely miss a sale; it forms a reliability judgment about that merchant's data infrastructure. Repeated encounters with stale data erode the merchant's reputation in the agent's decision model. Update Speed is, therefore, not just an operational metric - it is a trust signal.
Agent Uptake Rate measures the share of customers or orders leveraging AI agent assistance. This is a pure adoption metric - useful for tracking the pace of transition but silent on the quality of the agent experience. It tells you how many customers are using agents but not why they are using them, whether they trust the agent's decisions, or whether they are satisfied with the outcomes. In AXD terms, uptake without trust calibration is a vanity metric.
Agent Conversion Rate measures how effectively the AI agent turns opportunities into completed sales. This maps partially to Engagement Architecture - the fourth AXD pillar - but with a critical limitation. The cheat sheet frames conversion as a measure of how "persuasive" the agent is. The AXD Institute would frame it differently: conversion is a measure of how well the agent's recommendations align with the human's actual interests. An agent that converts at 80% by exploiting cognitive biases is not a success. An agent that converts at 40% by making genuinely appropriate recommendations is. The metric is the same. The interpretation depends entirely on whether you are measuring from the merchant's perspective or the human's.
04 - The Emerging Metrics: Where Trust Enters
The most revealing section of the cheat sheet is its treatment of "emerging metrics" - indicators that the platform vendor acknowledges are not yet standard but will become strategically important. Two metrics stand out, and both represent the moment where the operational perspective begins to brush against the trust architecture it cannot quite see.
The Uniqueness Index measures how distinct a merchant's catalogue and content are in an AI-driven marketplace. The insight is sharp: AI agents excel at comparison shopping, and if your products are indistinguishable from competitors', the agent will default to price or speed. A high Uniqueness Index means your offerings escape the commodity trap. From the AXD perspective, this metric captures something important about Signal Clarity - not just the completeness of information but its distinctiveness. An agent that cannot differentiate your product from a competitor's has a signal clarity problem, regardless of how complete your data fill rate is.
The Bot Trust Score is where the cheat sheet comes closest to the AXD Institute's core concern. Defined as a measure of "how much your customers trust AI agents to act on their behalf through your services," it can be gauged through surveys or behavioural signals - the ratio of fully automated orders versus those requiring human approval. The cheat sheet correctly identifies that "trust is the currency of agentic commerce" and that low trust means customers "double-check everything or simply opt out."
This is the right instinct. It is also where the structural limitation of the platform perspective becomes most visible. The Bot Trust Score measures trust as a sentiment - a survey response, a behavioural ratio. It does not measure trust as an architecture. It asks "how much do customers trust?" but not "what mechanisms are in place to form, calibrate, maintain, and recover that trust?" The difference is the difference between measuring a building's occupancy rate and measuring its structural integrity. Both are useful. Only one tells you whether the building will stand.
The AXD Institute's Readiness Assessment includes Trust Architecture as its fifth pillar precisely because trust must be measured not as a feeling but as a designed system. The Bot Trust Score is a leading indicator. Trust Architecture is the structural condition that determines whether that indicator will rise or fall.
05 - The Implementation Roadmap: A Trust Architecture Reading
The cheat sheet proposes a four-phase implementation roadmap: 0-3 months (preparation and assessment), 3-6 months (pilot an agentic experience), 6-12 months (expand and iterate), and beyond 12 months (scale and innovate). The roadmap is sensible, pragmatic, and operationally sound. It is also, from the AXD perspective, structurally incomplete in a way that matters.
Phase 1 (0-3 months) focuses on auditing data quality, evaluating API capabilities, and achieving internal alignment. This is correct and necessary. What is missing is any assessment of the trust conditions under which customers will delegate to agents. The AXD equivalent would include: What is the current trust baseline? What are the consequence levels of the transactions agents will handle? What trust signals does the existing customer relationship provide? Without this trust assessment, the operational audit is building on an unexamined foundation.
Phase 2 (3-6 months) recommends piloting with a specific use case - smart reordering, a personal shopping assistant, or integration with an external AI platform. The advice to "focus on a use case with clear customer value but limited risk" is sound and aligns with the AXD principle of the Autonomy Gradient - beginning with low-stakes delegations before progressing to high-consequence ones. But the cheat sheet's pilot guidance focuses entirely on functional testing: "Does it pick suitable products? Does it follow rules?" The trust architecture questions are absent: How does the agent earn the customer's initial trust? What happens when the agent makes a mistake? How does the customer calibrate their expectations? How does trust recover after a failure?
Phases 3 and 4 focus on measurement, optimisation, gradual rollout, and scaling. The cheat sheet wisely advises maintaining a "customer-centric and merchant-controlled mindset" and building trust "at each step." But the mechanisms for building that trust are left unspecified. The roadmap tells you to build trust. It does not tell you how. This is the gap that trust architecture fills - not as an alternative to the operational roadmap but as its necessary structural complement.
06 - Revenue Levers and the Delegation Problem
The cheat sheet identifies four revenue levers unlocked by agentic commerce: smart reordering and subscription autofill, bot-built bundles and upsells, B2B AI negotiation, and the 24/7 personal shopping concierge powered by AI shopping agents. Each is commercially sound. Each also depends, in ways the cheat sheet does not fully articulate, on a trust architecture that the merchant must design.
Smart reordering - where an agent monitors usage patterns and autonomously reorders products - is the canonical example of delegation design in consumer commerce. The customer delegates the replenishment decision to the agent. But delegation is not a binary state. The customer must specify: Which products? At what price threshold? From which suppliers? With what delivery constraints? What happens if the preferred product is unavailable? Each of these is a delegation parameter that must be designed, communicated, and calibrated. The cheat sheet describes the revenue opportunity. The AXD practitioner sees the delegation architecture that makes the revenue opportunity realisable.
Bot-built bundles - where an agent assembles personalised product combinations - raise the question of intent translation. When a customer says "I want to start a home gym," the agent must translate that vague intent into specific product selections. The quality of that translation depends not on the agent's recommendation algorithm alone but on the agent's understanding of the customer's actual needs, budget, space constraints, fitness level, and aesthetic preferences. Intent translation is a design problem, not a data problem. The cheat sheet's observation that "a well-trained agent will only suggest relevant combinations" is correct but incomplete - relevance is determined by the quality of the intent model, which must be designed and calibrated over time.
The 24/7 personal shopping concierge - an always-on agent that proactively identifies purchase opportunities - is the most trust-intensive of the four levers. A proactive agent that pings the customer with "The tent you liked is now 20% off, shall I grab it for you?" is exercising a form of initiative that requires deep trust. The customer must trust that the agent's proactive suggestions serve their interests rather than the merchant's margin targets. The line between helpful concierge and aggressive salesperson is a trust architecture problem. Without designed trust signals - transparency about why the recommendation was made, clarity about the agent's incentive alignment, and easy mechanisms for the customer to adjust the agent's proactivity level - the concierge becomes an annoyance rather than an asset.
07 - What the Playbook Cannot See
The structural limitation of the platform vendor's perspective is not a failure of intelligence or ambition. It is a consequence of vantage point. Platform vendors see the commerce stack from the infrastructure layer upward. They see data, APIs, integrations, and transactions. What they cannot see - because it is not visible from the infrastructure layer - is the trust architecture that determines whether the human will delegate to the agent in the first place.
The cheat sheet does not address trust formation - the mechanisms by which a customer develops initial confidence in an agent's capabilities. It does not address trust calibration - the ongoing process by which the customer adjusts their trust level based on the agent's actual performance. It does not address trust erosion detection - the systems that identify when trust is degrading before the customer disengages entirely. And it does not address trust recovery - the protocols that repair trust after a failure, which is inevitable in any autonomous system.
These are not optional enhancements. They are structural requirements. The cheat sheet's own data confirms this: only 24% of consumers are currently comfortable letting an AI agent shop for them. That means 76% are not. The question the platform vendor's playbook cannot answer is: what will move that 76%? Better data fill rates will not do it. Faster update speeds will not do it. More API endpoints will not do it. What will move it is a trust architecture that gives humans the confidence to delegate - and the assurance that when things go wrong, the system will recover gracefully.
This is not a criticism of Shopware's work. It is a statement about the boundaries of any single-layer analysis. The platform vendor's playbook is excellent infrastructure guidance. It needs the trust architecture layer to become complete readiness guidance. The two are complementary, not competing.
08 - The AXD Readiness Overlay
The AXD Institute's Readiness Assessment evaluates organisations across five pillars: Signal Clarity, Reputation via Reliability, Intent Translation, Engagement Architecture, and Trust Architecture. The Shopware cheat sheet's KPIs map to the first, second, and fourth of these pillars with reasonable fidelity. The third pillar - Intent Translation - receives partial coverage through the discussion of agent recommendations and personalisation. The fifth pillar - Trust Architecture - is the structural gap.
For merchants reading the Shopware cheat sheet, the AXD overlay would add five questions to each phase of the implementation roadmap:
During assessment (0-3 months): What is the trust baseline with your customers? What consequence levels will agent-mediated transactions involve? What trust signals does your existing brand relationship provide? What are the failure modes that would destroy customer trust in agent-mediated commerce? What recovery mechanisms exist today?
During piloting (3-6 months): How does the agent earn initial trust? What happens when it makes a mistake? How does the customer calibrate their expectations? What observability does the customer have into the agent's decisions? How does the pilot measure trust formation, not just conversion?
During scaling (6-12+ months): How does trust architecture scale across product categories and consequence levels? What trust erosion patterns are emerging in the data? How does the system detect and respond to declining trust before customers disengage? What is the trust recovery protocol when systemic failures occur?
The Shopware cheat sheet provides the operational foundation. The AXD readiness overlay provides the structural architecture. Together, they constitute a complete readiness framework for the age of agentic commerce.
The merchant who follows only the operational playbook will have excellent infrastructure and wonder why customers are not delegating. The merchant who adds the trust architecture will understand that delegation is not a feature to be launched but a relationship to be designed - and that the design of that relationship is the defining challenge of the agentic age.