LVEN
Start a project

E-commerce

Agentic commerce: preparing your online store for AI buyers

AI agents are starting to search and buy on a user's behalf, and open protocols for it are published. A store can prepare with ordinary work: clean product data, accurate stock and prices, a predictable checkout.

FSM Studio · · 6 min read

A small robot beside a cardboard parcel and a woven shopping basket on a cream background

Protocols for AI agents that buy on a user’s behalf are published by OpenAI and Stripe, Google, Visa and Mastercard. But even a technically ready store needs sound data first. An agent cannot buy what it cannot parse, and it will not trust stock figures that do not match reality.

What is published, and its status

Agentic commerce means software finds a product, compares offers and places an order for a person. That needs a shared language between agent, store and payment provider. Four sources can be checked today.

  • Agentic Commerce Protocol (ACP). Open standard from OpenAI and Stripe, published 29 September 2025, behind Instant Checkout in ChatGPT. Apache 2.0, marked beta on GitHub, latest spec 2026-04-17. The merchant stays the merchant of record and accepts or declines each transaction.
  • Universal Commerce Protocol (UCP). Open standard under Apache 2.0, co-developed by Google, Shopify and others. It covers catalogue, cart, checkout, account linking and order status. Trade press dates the announcement to 11 January 2026 and says Google’s implementation needs a Merchant Center account and is open to US merchants for now. That is a secondary source, so confirm before you decide anything.
  • Agent Payments Protocol (AP2). Announced by Google on 16 September 2025. It rests on signed “mandates”: one records what the user allowed the agent to do, the other confirms the exact cart and price.
  • Visa and Mastercard. Visa Intelligent Commerce launched on 30 April 2025, Mastercard Agent Pay with agentic tokens on 29 April. Both replace card details with tokenised credentials and let the user set spending limits.

We found no primary source showing that any of these channels already works for Latvian or Baltic stores. OpenAI’s documentation says product feed onboarding is limited to approved partners for now. So this is not an urgent rollout. It is a good reason to fix the basics.

What an agent reads from your store

An agent has no eyes. It reads fields: name, price, currency, stock, delivery terms, return policy. Missing or contradictory fields mean it skips the product.

  1. Structured data on the page. Add schema.org Product and Offer markup with price, currency, availability, shipping and returns. Google’s documentation recommends sharing availability, shipping costs and return information.
  2. Unique identifiers. SKU, manufacturer part number or GTIN where one exists. This is how an agent matches the same product across stores.
  3. Attributes, not just prose. Size, material, compatibility, pack size in separate fields. A long marketing text is worse for an agent than a precise spec table.
  4. A product feed. ACP and Google Merchant Center work from feeds. OpenAI recommends uploading a full catalogue snapshot at least once a day, which is not enough if your stock changes faster and you have no incremental updates.

If you already run a Google Shopping or price-comparison feed, part of the work is done. Fix its errors first.

Stock and price accuracy

A person who sees “out of stock” after ordering writes an email. An agent simply stops choosing your store. Accurate stock and prices matter more than any new protocol.

Where accounting or warehouse software sits apart from the website, stock often reaches the site late. Measure how often it updates and compare with what the site and the feed show. If the gap is hours, fix the integration, not the feed. We do this, for example, with Moneo synchronisation for online stores and Jumis and WooCommerce integration.

The price in the feed, on the page and in the cart must match to the cent, including VAT and delivery. If it does not, an agent has reason to reject the offer.

Checkout: no obstacles, but control

An agent buying for a user will not solve a CAPTCHA, and you should not expect it to. ACP and UCP take another route: the store exposes a separate checkout interface, and the agent submits the cart and a payment token. The payment provider processes the token and the store accepts or declines. Stripe says the token is programmatically controlled, permissioned and logged.

In practice:

  • Ask your payment provider whether it supports agent tokens or plans to. Ask, do not assume.
  • Keep checkout predictable: clear delivery options, terms before payment, an email confirmation.
  • Do not switch CAPTCHA off for everyone. Build a separate path for verified agents and keep normal protection for the rest.

Fraud and liability

Stripe writes that businesses will need to respond to new fraud signals and tell good bots from bad ones. The AP2 authors name three questions: did the user authorise the agent, does the request match their intent, and who is responsible if something goes wrong.

For a store this means one thing: you decide whether to accept an order, and the liability stays with you. Before connecting, ask your payment partner how disputes and returns are handled for agent orders.

What to do this year, and what to leave

Do what pays off anyway: clean product data and markup, accurate stock, matching prices, a reviewed checkout. These also help ordinary search and comparison feeds.

Leave connecting to a specific protocol for later. The standards are still moving (ACP is beta, parts of UCP are marked draft) and we have no data on demand in the Baltics.

Conclusion

If your catalogue is complete and accurate and stock is synchronised with the warehouse, connecting to a protocol will be plain technical work once it reaches your market. If attributes are missing, prices differ or stock lags, start there: it pays off now, and agents are one more reason.

We build and repair online stores and integrations with accounting systems. We can review your catalogue and stock flow and tell you what to fix first. Discuss your project.

Sources

Read also

All e-commerce articles →
E-commerce · 5 min

E-shop accessibility in Latvia: what the law requires since June 2025

E-commerce · 4 min

B2B E-commerce in the Baltics: How Wholesalers Go Online

E-commerce · 4 min

70% of Shoppers Abandon Their Cart: What Actually Helps