Agentic commerce: how businesses get discovered and paid by AI agents

What changes when the buyer uses an AI agent?

A conventional buyer searches, opens several pages and completes each step themselves. An AI agent can do more of that work on the buyer’s behalf: turn a request into requirements, compare providers, check constraints, prepare a booking or purchase and ask for approval before committing.

That changes the unit of competition. A business is no longer competing only for a click. It is competing to be understood accurately, included in the shortlist and usable when the agent tries to complete the task.

• Discovery: can the agent find the right product, service or provider?
• Understanding: can it identify price, availability, location, policies and fit?
• Selection: is there enough evidence to compare the business with alternatives?
• Action: can it enquire, book, add to cart or check out without avoidable friction?
• Measurement: can the business identify agent-led visits, orders and qualified enquiries?

The three routes into agentic commerce

The first route is the open web. An agent uses a browser to read pages and operate forms much like a person would. This makes accessible navigation, explicit product facts and resilient forms commercially important.

The second route is a catalog or commerce channel. Shopify, for example, distributes structured product information to supported AI channels and can keep price and availability synchronized. The third route is a connector: a reviewed integration that lets an agent use a company’s product or service directly.

These routes are complementary. A connector can unlock a richer workflow, while the public website remains important for explanation, trust, policies and buyers who want to verify the recommendation.

What makes a business agent ready?

Start with commercial clarity. The agent needs to know what is sold, who it is for, where it is available, what it costs and what makes it a suitable choice. Important constraints should be written as facts rather than buried in imagery, vague marketing language or a long support document.

Then test the journey. Product selection, variant choices, booking calendars, account requirements, delivery rules and payment steps can all stop an otherwise qualified buyer. The useful question is not whether a site contains an AI-specific file. It is whether an agent can complete the same valuable task a buyer can.

• Specific product and service descriptions with stable names and identifiers
• Current price, stock, service area, delivery and availability information
• Clear returns, cancellation, privacy and customer-support policies
• Credible evidence such as reviews, case studies and independent references
• Forms and checkout flows that communicate errors and required fields clearly
• A measurable handoff from discovery to enquiry, booking or order

Where Meta Muse fits

Muse is a useful current example because it combines web browsing, connected apps and approved purchases. It can operate across ordinary websites, while its Connector Platform gives businesses a route to place a capability directly inside the product.

For Shopify merchants, Meta is also available as an AI channel through Agentic Storefronts. Products can be distributed through Shopify Catalog, and eligible buyers can complete a Shopify-powered direct checkout on Meta surfaces. Businesses outside Shopify still need to test the actual web journey rather than assume that payment compatibility makes the whole experience reliable.

How should agentic commerce be measured?

Separate discovery, action and commercial outcome. A crawler request shows access. A recommendation shows inclusion in an answer. A completed form or checkout shows that the journey worked. Revenue and qualified pipeline show whether the channel mattered.

Run controlled tests before changing the site, then repeat the same tasks after each intervention. Record what the agent saw, where it hesitated, which facts it got wrong and whether the final transaction matched the buyer’s instruction. This creates a practical improvement backlog instead of a speculative checklist.

• Agent visits and identifiable referrals
• Product or provider inclusion for relevant buyer requests
• Accuracy of descriptions, prices, policies and availability
• Task completion and checkout success rate
• Qualified enquiries, bookings, orders and revenue
• Failures requiring a buyer or staff member to intervene

A sensible first project

Choose one valuable buyer task, such as finding a suitable product, booking a service or starting a trial. Test that task across the most relevant assistants, including Muse where the target buyer has access. Fix the smallest set of information and journey problems that prevent a correct result.

Only then decide whether a catalog integration or connector is justified. Its not a race to add every protocol, its a focused effort to make one commercially important journey work end to end.