Agentic commerce means an AI agent, not a person, performs part or all of a purchase: it searches, compares, negotiates and in some cases checks out. In 2026 this stopped being a demo. Google introduced the Universal Commerce Protocol at NRF, an open standard built with Shopify, Etsy, Walmart and Target and endorsed by more than twenty companies including Visa and American Express. Its agentic checkout, branded ‘Buy for Me’, is live inside AI Mode with launch partners such as Wayfair, Chewy and Quince.
Meta moved in parallel. Meta One for Business bundles insights, competitor benchmarking, scheduling and an AI Business Agent, and Meta Muse is a shopping agent that works with Shopify stores and can negotiate prices on the shopper’s behalf. OpenAI, Perplexity and Amazon are building comparable flows. The shopper is increasingly a delegate, and the storefront the agent sees is a data feed, not a web page.
How an agentic purchase actually works
- The user states a goal in natural language: “find a quiet dishwasher under €700 that fits a 45 cm gap, delivered this week.”
- The agent queries structured catalogues (merchant feeds, UCP endpoints, marketplaces) rather than browsing pages.
- It filters on hard constraints (size, price, delivery), then ranks on soft signals (reviews, return policy, brand trust).
- It presents one to three options; on approval it completes checkout through a protocol such as UCP, with payment handled by a tokenised wallet.
Why this changes retail marketing
Three assumptions break. First, persuasion on the product page is bypassed — the agent never sees your hero image. Second, ranking is decided by data completeness and policy clarity rather than ad spend alone. Third, price becomes negotiable in real time, because agents like Muse are explicitly built to bargain. Retailers that treat the feed as an afterthought will be invisible to the fastest-growing purchase channel.
The retailer checklist for 2026
1. Make the product feed the primary storefront
Every attribute an agent could filter on must be present and accurate: dimensions, weight, materials, compatibility, energy class, availability by location, delivery windows, return window, warranty. Google’s feed in Merchant Center, Meta’s catalogue and your UCP endpoint should be generated from one source of truth and validated daily.
2. Publish machine-readable policies
Agents weigh return policy, shipping cost and delivery speed as ranking factors. Encode them in schema (Offer, shippingDetails, hasMerchantReturnPolicy) and in the feed; a policy buried in a PDF does not count.
3. Decide your negotiation stance
If agents can ask for a better price, you need rules: floor prices, bundle offers you are willing to make, and which products are fixed. Treat it as dynamic pricing with guardrails rather than a surprise.
4. Earn structured trust signals
Verified reviews, ratings with volume, merchant certifications and consistent brand entities across Google, Meta and marketplaces feed the soft-ranking step. Review velocity is now a merchandising lever.
5. Keep ads in the loop
Shopping ads already serve inside AI Mode from existing Shopping and Performance Max campaigns. Sponsored placements inside agent recommendations are the next inventory; budgets should be ready for it.
Measurement in an agent world
Expect fewer sessions and more direct conversions with little or no on-site behaviour. Attribution models built on page views will undercount agent-driven sales. Track orders by channel of origin (UCP, Meta, marketplace API), monitor feed impression and selection rates where platforms expose them, and compare margin per channel — negotiated orders may cost more to win.
What to do this quarter
- Audit the product feed for completeness against the attributes above.
- Add Offer, shipping and return schema to every product page.
- Join the Universal Commerce Protocol via your platform (Shopify and major carts expose it) and test ‘Buy for Me’ eligibility.
- Set negotiation guardrails and document them.
- Create a reporting line for agent-originated orders.
The retailers winning early are not the ones with the prettiest sites. They are the ones whose data an agent can trust.