The week the industry started writing it down.

The last month has been launches. Coinbase shipping an SDK, Stripe and Cross River wiring single-use cards, a foundation forming around x402. This week was quieter and, if you have ever had to answer a regulator, more consequential. The standards bodies started publishing, and one of them opened its draft to public comment. This is the part of a payments cycle where the story stops being about what the rails can do and starts being about what everyone agrees they mean.

The trade body writes the definition

The U.S. Payments Forum published a white paper on 22 July called Agentic Commerce: A Primer for the Payments Industry, written by its Emerging Payments and Payments Fraud working committees [1]. The framing in it is the useful bit. The Forum treats agentic commerce as a new class of payment initiation, one the existing frameworks were not designed to accommodate, and its managing director Devon Rohrer puts the pressure point on consumer intent and authorisation, which are now understood differently because an agent can start a transaction on a shopper's behalf [1].

That is a categorical claim, not a product claim. Initiation is where a payment gets its rules: who authorised it, what evidence exists, who eats it when it goes wrong. Saying the current categories do not fit is the industry admitting the mapping is unfinished.

An evidence standard goes out for comment

A-Comm Technologies opened its draft A-Comm Evidence Protocol for public comment on 13 July, with the window running to 14 August [2]. The protocol turns each step of an agent-led purchase, from discovery and intent through delegation, policy, cart, authorisation and fulfilment, into an evidence artifact in a tamper-evident hash chain that is sealed at authorisation [2]. It is Apache 2.0 licensed and deliberately positioned as a witness rather than a rival, wrapping the payment protocols already in use without assigning outcome or liability [2]. Digital Transactions carried it in its 20 July briefs as part of the same agentic-preparation wave [3].

Meanwhile the rails kept widening

Two moves underneath. Circle signed non-binding memoranda with Kakao Group and Toss Bank on 23 July to explore stablecoin payment rails in South Korea, pairing Kakao's messaging and payment distribution with a regulated bank [4]. Issuance structures, governance and launch timing were not disclosed, so treat it as intent, not infrastructure [4]. And Mastercard is extending card settlement into intraday, weekend and holiday cycles, in fiat and in regulated stablecoins, so institutions are not stuck waiting on banking hours to move liquidity [5].

Read from the rails

A definition paper sounds like the least interesting item here. It is the one I would read first. Ten years in payments operations taught me that most of the pain in a system comes from a category that was never properly defined at the start, and then had to be argued about later, transaction by transaction, in a dispute queue. Payment initiation is exactly that kind of category. If nobody agrees whether an agent initiating a purchase is card-not-present, delegated authority, or something new, every party in the chain will default to the reading that protects them, and the merchant will find out which one won only after the chargeback.

The evidence protocol is the same instinct from the other direction. Sealing the record at authorisation matters because that is the moment you can still prove intent. Reconstruct it afterwards from logs and you are assembling a story, not producing a record. The detail worth watching in the comment period is what happens when the chain is incomplete: an agent that crashed mid-run, a policy that changed between intent and authorisation, a fulfilment step that never reported. Standards are easy to design for the transaction that completes. The value shows up in the ones that do not.

The launches tell you what the rails can do. The definitions tell you who pays when they do it wrong.