Agentic commerce research.
Reporting on the protocols, product shifts, legal fights, and market structure changes shaping agent-led commerce.
A Tool Call Is Not an Authorization
WebMCP gives a browser agent a clean way to ask a merchant page for a checkout. It says nothing about who is asking or whether they can spend the money. Here is how we split that into a read-only preflight and a server-side decision — and the three pieces that are not finished.
Read articleAuthorization Needs an Outcome Record
A valid agent credential proves that a request carried an authorization claim. It does not prove what the merchant decided or how the transaction ended. KYA’s Phase 0 design records those facts as one issuer-neutral, append-only evidence chain.
Read articleMandate Matching and Evidence Receipts for Agent Checkout
How to check an agent-initiated purchase against the mandate its principal granted: normalize the mandate, match it field by field, treat "not specified" as not allowed, decide accept/review/decline/step-up, and write a signed evidence receipt your dispute tooling can replay. Written for merchant architecture teams, with the API contract, policy config, and schemas.
Read articleAgent checkout needs verifiable records
KYA ships signed traces, verify recommendations, checkout-session enforcement, and pre-dispute APIs for agent checkout.
Read articleFollow the systems behind agent traffic, not the headlines about it.
Technical analysis for the merchants, platforms, and operators who will have to run this in production.
