A-Comm Technologies founder and CEO Krystal Gracier opened the draft A-Comm Evidence Protocol for public comment on July 13th, according to a July 20th announcement. The San Francisco payments founder is asking merchants, banks, payment networks and agent developers to review a proposed record for showing what an AI agent was instructed to buy, what it did and what ultimately happened to the order.
The comment period runs through August 14th. AEP remains a draft, version v1.0.3-rc.2, with no disclosed merchant, payment network or agent platform adoption.
Gracier came to the problem through the institutions that will eventually have to investigate these transactions. FINRA records confirm earlier roles at Morgan Stanley, Citi and Wells Fargo. A-Comm says she later worked at Visa and Kraken, directed Visa's relationship with Apple Pay and advised Google during early discussions around its Agent Payments Protocol. Those Visa and Google details come from A-Comm's biography.
Her thesis is rooted in a basic payments problem. When a person clicks a checkout button, the merchant can retain the cart, payment authorization and account signals. An agent introduces another decision-maker between the shopper and the transaction. A dispute can turn on whether the agent followed a budget, substituted an item, acted under an expired instruction or completed a purchase the consumer never approved.
"For agentic commerce to scale, it needs to be observable, traceable, and auditable," Gracier said in the announcement.
A record that follows the transaction
The AEP specification defines evidence artifacts for discovery, referral, intent, delegation, merchant policy, cart contents, authorization and fulfillment or refunds. A-Comm says those artifacts are linked in a tamper-evident hash chain, sealed at authorization and exportable so another party can re-verify the record.
In practical terms, the record is meant to answer a sequence of questions that existing payment logs often split among different parties: What did the shopper request? Which agent acted? What limits did the shopper set? What entered the cart? What amount was authorized? What arrived?
A-Comm says AEP does not decide which party is responsible or whether a chargeback should succeed. It packages evidence for banks, networks, processors and merchants to evaluate under their own rules.
That limitation is central to Gracier's positioning. Visa's Trusted Agent Protocol helps merchants recognize approved agents and verify signed information attached to their requests. Mastercard's Verifiable Intent focuses on cryptographic proof of what a consumer authorized. AEP is designed to wrap records from existing agent, identity, checkout and payment protocols, including Visa TAP and Mastercard Verifiable Intent, rather than replace them.
A-Comm argues that the missing piece is a portable outcome history that survives beyond authorization and follows the order through delivery, cancellation or refund. That is a useful gap to target, though the payment networks already control the rules, integrations and commercial relationships that determine which evidence issuers accept.
Gracier is racing the holiday calendar
The timing reflects an approaching retail deadline. Retailers are preparing for the 2026 holiday season, when agent-initiated purchases could move beyond product recommendations and into payment authorization. Evidence for a disputed holiday purchase has to be captured when the transaction occurs; it cannot be reconstructed reliably after a cardholder files a claim months later.
A-Comm cites Salesforce's 2025 holiday data, which attributed 20% of global online sales to the influence of AI and agents. That measurement included personalized recommendations and conversational customer service. It shows broad AI involvement in shopping, though it does not establish that agents initiated or authorized one-fifth of purchases.
The opening for A-Comm is narrow. A small vendor cannot impose a format on banks and processors. Publishing AEP under Apache 2.0 removes licensing fees and gives outside implementers access to the specification, test vectors and reference materials.
The open standard is also A-Comm's commercial entry point
AEP is free, but it sits inside a broader A-Comm product plan. A-Comm is onboarding a limited number of merchants for a voice and text storefront agent that recommends products, builds carts and completes purchases through a merchant's existing checkout. A-Comm also advertises a Context MCP, a service through which shoppers can approve scoped access to derived signals from services such as Gmail, Pinterest, Instagram, TikTok and Notion.
The evidence protocol could make those products easier to sell. A merchant using A-Comm's shopping agent would also have access to an evidence-capture system designed around the agent's activity. A-Comm's governance document says A-Comm operates a commercial reference implementation and evidence custodian service, while the specification defines those roles generically and does not require A-Comm software.
That separation is the heart of Gracier's strategy: make the record format open, then compete to provide the agent, capture service and evidence storage around it. The approach can lower resistance among merchants and payment providers that would reject a proprietary standard controlled entirely behind a vendor contract.
Governance remains an unresolved test. A-Comm currently stewards the maintainers during what it calls a bootstrap phase. A-Comm plans to add outside maintainers, publish conflict-of-interest rules and eventually engage a neutral standards organization. Those steps have not yet occurred.
The comment process requires A-Comm to publish a disposition for each submission, including whether the proposal was accepted, modified, deferred or declined. Release gates for the final v1.0.3 specification include conformance test vectors and a completed disposition log.
A-Comm has not disclosed funding, revenue, transaction volume or a named early-access merchant. A-Comm says its strategic angel investors include people affiliated with Visa, Mastercard, American Express, SoFi, Citi, Wells Fargo, Qualcomm, Fintech Brainfood and AIUC. The wording refers to individual backers from those organizations and does not establish corporate investments, integrations or partnerships.
Gracier has put a technically specific proposal into the open before agent purchases generate a meaningful dispute history. AEP's value will depend on external implementation: merchants capturing the artifacts, agent platforms signing them and payment providers accepting the resulting bundles. The public review gives Gracier a chance to turn a founder-written specification into shared infrastructure before the largest networks settle the question themselves.