Agentic commerce: when AI agents start buying on your online shop

From browsing to delegating: a new kind of shopperFor two decades, running an online shop has meant designing for a human being: someone who lands on...

From browsing to delegating: a new kind of shopper

For two decades, running an online shop has meant designing for a human being: someone who lands on a product page, compares options, hesitates at the checkout and occasionally abandons the cart. That assumption is starting to crack. Large retailers are already testing ways to let AI assistants complete purchases on a customer's behalf, with spending limits, per-order approvals and revocable permissions. The shopper no longer clicks; the shopper instructs.

This shift is usually called agentic commerce, and it matters well beyond the big players. If your store sells in Barcelona, Lleida, Tarragona or Girona — or ships across the EU — your next visitor may not be a person at all, but a software agent acting on someone's instructions.

What an AI agent actually does in a shop

Today, most agents work the way a person would: they open a browser, scroll, fill in forms and click through the checkout. It works, but it is slow, fragile and prone to errors when a page changes. The more mature approach is to expose a structured interface that agents can query directly, so they can search a catalogue, check stock and place an order without replaying every human step.

Alongside that, retailers are experimenting with permission models. Think of it as adding an authorised user to a payment card: the agent gets a defined budget, a scope of action and the ability to be switched off at any moment. Customers can choose between approving each order or letting the assistant operate autonomously within a cap.

Scoped permissions: the agent only sees what it needs to complete the task, not the full order history.
Spending limits: a ceiling in euros that the agent cannot exceed without fresh approval.
No access to payment credentials: agents should never read or modify stored card data.
Revocable access: the customer can cut the agent off instantly.
Same prices and offers: an agent should not unlock different conditions than a human buyer.

Why this is a conversion question, not a sci-fi one

Strip away the novelty and agentic commerce is the next layer of the same problem every ecommerce project faces: removing friction between intent and completed order. A store that is slow, has inconsistent stock data or hides shipping costs will fail an AI agent just as it fails a human. In fact, it will fail faster, because an agent does not persevere out of brand loyalty.

That is why the groundwork is familiar. Clean, well-structured product data. Accurate availability. Transparent pricing and delivery times. A checkout that behaves predictably. Stores built on PrestaShop or WooCommerce are well placed here, provided the catalogue and the logic behind it are properly maintained rather than patched together over the years.

Preparing your store in Catalonia for agent traffic

You do not need to wait for a formal standard to arrive in your sector. Several practical steps make sense now, and most of them improve the experience for human customers too.

Audit your product feed. Attributes, variants, dimensions and compatibility data should be complete and consistent across the catalogue.
Expose structured endpoints. A documented API for search, stock and orders gives agents a reliable path that does not depend on scraping your theme.
Keep stock truthful. Sync inventory with your ERP or warehouse so an agent never commits to a product you cannot ship.
Design for partial disclosure. Let an assistant retrieve what it needs — a favourite brand, a repeat purchase — without handing over the customer's entire history.
Log agent activity separately. Distinguish orders placed by people from those placed by assistants, so support and analytics stay readable.
Keep your returns policy intact. Whatever channel the order came through, consumer rights under Spanish and EU rules do not change.

GDPR and trust: the part nobody should skip

Delegating a purchase means sharing personal data with a third-party assistant, and that raises questions your privacy documentation should already answer. Which data does the agent receive, for how long, and on what legal basis? If a customer revokes access, is the data deleted? Under the GDPR, minimisation is not optional: an agent that only needs to know a milk brand should not receive the full purchase history.

Trust will be the real currency of agentic commerce. Customers will delegate to stores they believe will not overspend on their behalf, will honour returns and will be transparent about what the assistant did. A clear permissions screen and an honest activity log do more for conversion than any badge.

The opportunity for smaller merchants

It is tempting to read this as a story about large chains with engineering teams to spare. In practice, the merchants who move first tend to be mid-sized brands with a modern, well-maintained ecommerce stack and the flexibility to test. A custom online shop built with conversion in mind, with a solid API layer and clean data, is exactly the kind of environment where an agent integration can be piloted without rebuilding everything.

The realistic timeline is gradual: agents will handle repeat purchases and simple replenishment first, while complex or high-value orders stay with people. But the stores that treat their catalogue, their checkout and their permissions as infrastructure — rather than decoration — will be the ones an assistant can actually buy from.

If you are planning your next ecommerce project in Lleida or anywhere in Catalonia, the question is no longer whether to optimise for humans or for machines. It is whether your shop is clear enough for both.

Put these ideas into practice

Talk to ALMC about a solution for your business. Explore your options or contact our team.

Results without leaving the page.