Agentic commerce: making a store legible to buyers that are not people.
Agentic commerce is the shift from a person browsing your storefront to an AI agent shopping on their behalf — comparing, deciding and increasingly checking out. Every article about it explains what it is. This page is about the part a merchant has to act on: whether your product data, checkout and bot policy survive contact with a buyer that never renders your CSS.
The short definition. Agentic commerce is commerce mediated by an autonomous AI agent: the shopper delegates a task — find, compare, buy, reorder, return — and software carries it out across storefronts. It differs from a chatbot on your site in one decisive way: the agent is the customer's, not yours. It has no loyalty to your brand, it does not see your merchandising, and it will happily prefer the competitor whose data it can read more easily.
That last sentence is the entire commercial argument, and it is why this is an engineering problem rather than a trend to monitor. When the buyer is a program, the things that win are machine-legible truth about your catalogue, a checkout that does not depend on a human eye, and being reachable at all — a blanket bot block is a decision to be invisible to this channel, and plenty of stores have made it by accident.
What is actually settled, and what is not. The Agentic Commerce Protocol published by Stripe and OpenAI is the first serious attempt at a standard for agent-initiated checkout, and the major platforms are moving. It is also early, and anyone telling you they have a mature, battle-tested ACP integration is telling you a story. We will not claim shipped ACP deployments. What we do claim is twenty years of doing the underlying work — product data, feeds, checkout, integration — which is most of the readiness cost regardless of which standard wins.
So the honest version of this offer: a readiness assessment that tells you where your store stands today, followed by the work that pays off under any protocol — clean structured data, agent-accessible product truth, a checkout that can accept a delegated purchase, and an access policy you chose deliberately. Nothing in that list is wasted if the standards shift, which is exactly why we start there.
The groundwork is already visible on this site: DTN publishes a machine-readable llms.txt and structured data across every commercial page, because we would rather demonstrate the practice than describe it.
Six things that decide whether an agent can transact with you.
Agent-readable product truth
Complete, current schema.org Product markup, clean feeds, and resolvable answers to price, stock, variant, shipping and returns. An agent that cannot establish a fact treats it as absent — and picks a store where it is not.
A bot policy you actually chose
Most stores block or allow AI crawlers by accident, through a WAF rule or a default. We audit what reaches you today across robots.txt, CDN and bot management, then set a policy per agent with the commercial trade-off written down.
Checkout that survives a delegated buyer
Guest and delegated purchase paths, idempotent order creation, programmatic tax and shipping resolution, and failure modes that return a usable error instead of a rendered page. ACP-aware, without pretending the standard has settled.
Platform work on Magento, Shopify Plus and Medusa
The readiness changes land in a real codebase. We are an Adobe Commerce partner, a Shopify Plus partner and a Hyvä Silver Supplier, so this is the same stack we already maintain for clients.
Answer-engine visibility
The same structured, machine-legible substrate decides whether you are cited by AI assistants at the research stage, long before anything is bought. One body of work, two payoffs.
Measurement, not vibes
Agent and AI-crawler traffic separated in your analytics and logs, so you can see this channel arriving rather than infer it. Most stores currently cannot tell you the number.
How the readiness assessment runs.
Crawl and probe, as an agent would
We fetch your store the way an automated buyer does — no JavaScript rendering, no cookies, no goodwill — and record what is establishable about your catalogue and what is not.
Access and policy audit
What your robots.txt, CDN rules and bot management currently do to each major AI crawler and agent, versus what you thought they did.
Checkout path review
Whether a delegated purchase can complete, and where it breaks — auth, tax, shipping, stock reservation, error surfaces.
Ranked findings and a costed plan
Ordered by commercial exposure, with effort against each. Yours to keep, and specific enough to hand to another supplier if you prefer.
What you get, stated as deliverables rather than outcomes.
No conversion or revenue figures on this page on purpose. Agentic commerce is too new for anyone to have credible longitudinal numbers, and a supplier quoting them is quoting someone's projection.
The stack.
Common questions.
Is agentic commerce real yet, or is this a land grab?
Both, honestly. Agent-mediated buying is small today and growing, and the protocol layer is unsettled. What is already real is the research stage: AI assistants are answering product questions about your catalogue now, from whatever data they can reach. The readiness work pays for itself there first, which is why we recommend starting even if you think transactional agents are two years out.
Have you shipped an Agentic Commerce Protocol integration?
Not a production ACP checkout, no — and you should be sceptical of any agency that claims a mature one this early. What we have shipped for two decades is the substrate it depends on: structured product data, feeds, custom checkout, tax and shipping resolution, and integration into ERP-owned price and stock. That is where the readiness cost actually sits, and it is protocol-independent.
Should we block AI crawlers or let them in?
It depends on whether you sell something an agent could recommend, and the answer differs per crawler — training, search indexing and live agent fetching are three different bargains. The failure we see most is not a wrong choice, it is no choice: a default WAF rule quietly excluding you. We will show you what currently reaches you and make the trade-off explicit either way.
Does this replace SEO?
No, it extends it. Structured data, crawlability, clean information architecture and factual product content are what both classical search and AI agents run on. If someone sells you "agentic commerce" as an entirely new discipline unrelated to your existing search work, be careful.
What is the smallest sensible first step?
The assessment. It is deliberately fixed-scope and does not commit you to a build, and its most common outcome is a short list of unglamorous data and checkout fixes rather than an AI project.
Which platform handles this best?
Less than you would hope — the constraint is usually your data quality and checkout customisation, not the platform. Shopify Plus moves fastest on native agent features; Magento and Medusa give you more control to implement ahead of the platform. We will tell you which applies to your catalogue rather than which we prefer to sell.
Where to next.
Find out what an agent can currently see.
We will fetch your store the way an automated buyer does and tell you what is establishable and what is not. Most stores are surprised by the answer in at least one direction.