Agentic commerce is the layer that makes a site safely transactable by an AI agent: machine-readable pricing, declared checkout paths, exposed payment options, and discoverable refund and authorisation terms. It’s about the signals a site exposes about its commerce stack, not about replacing Stripe or building on-chain rails.
Why this matters
AI agents are starting to act on behalf of users: comparing products, requesting quotes, subscribing to APIs, booking services, paying for digital actions. The web they navigate was built for humans clicking links, not for programs evaluating SKUs and triggering payments. Today, most sites fail in three ways:
- Pricing is human-only. Tier cards lock prices behind tooltips, JavaScript, or “Contact sales”.
- Checkout is opaque. A button labelled “Get started →” tells an agent nothing about which URL, which auth, or how to confirm.
- Payments are undeclared. No metadata says whether the site accepts cards, ACH, stablecoins, or HTTP-level payment signals.
Five capabilities, declared
- Pricing as data. JSON-LD
Offer/Product, an/api/pricingJSON, or an MCP tool an agent can call. - Checkout as a path. A documented sequence: this URL → this auth → this payload → this confirmation.
- Terms as content. Refund / cancellation reachable as parseable HTML, not buried in a PDF.
- Payment options as metadata. Cards, ACH, wire, stablecoins, x402, declared on
acceptedPaymentMethod. - Identity as proof. Wallets, contracts, or business-identity endpoints when relevant. Optional. Not crypto- mandatory.
What AgentSpeed will measure
AgentSpeed publishes a research-track rubric proposing 12 commerce checks across pricing, checkout, terms, API access, payments, and MCP. It is not yet scored. Crypto- specific checks are scoped n/a by default, so sites that don’t accept on-chain payments aren’t penalised.