Hook
Only 14% of consumers trust an AI agent to complete a purchase without verification. 42% reject any transaction above $25. These are not hypotheticals. These are the cold, hard numbers from Product.ai. And Visa has placed a massive bet that by the 2026 holiday season, millions will use AI agents to shop. The Agentic Ready program is their weapon. But the battlefield is not where you think.
Context
Visa's Agentic Ready program is a certification standard for card issuers. It ensures that the issuer's systems can handle AI-agent-initiated transactions. The program covers card registration, tokenization, and authentication through Visa Payment Passkeys. 99% of issuing systems are technically capable, according to Visa. The program is rolling out across Europe, Asia-Pacific, Latin America, Canada, and CEMEA. Over 85 partners have signed up in APAC and LatAm, 30+ in CEMEA, and all five major Canadian banks. Visa's goal: lock in the nascent agentic commerce flow before alternative rails emerge.
But here is the truth that the press release will not tell you. The certification is a defensive move. It is not about new revenue. It is about preserving the existing network toll. If AI agents bypass the card network via open banking or direct A2A payments, Visa loses its intermediary role. Agentic Ready is the moat being dug before the invasion.
Core: The Real Architecture of Trust Has a Blind Spot
Heat maps focus on the issuer side. They show how the German PoC successfully routed a transaction through standard authorization after passkey authentication. They celebrate the 99% readiness figure. They highlight the passkey and tokenization infrastructure.
Let me apply my own cryptographic audit lens. I have been testing systems for integrity since 2017, when I found an integer overflow in a vesting contract that would have drained millions. The critical insight here is not the 99% figure. It is what the 1% represents and what the 99% actually means.
99% of systems can technically route a packet. That is a layer 2 statement. The real question is: can the issuing system distinguish between a human-initiated transaction and an agent-initiated transaction? That distinction is not a simple flag. It requires new metadata fields in the authorization message, new logic in the fraud detection engine, and new rules in the core banking system. The certification is verifying that issuers can add that metadata and handle the agent context. But the metadata only works if the system can trust the agent identity.
And here is the structural blind spot. The Agentic Ready program certifies the issuer. It does not certify the agent developer. The agent is a black box built by a third party. The agent could be a legitimate shopping assistant, or it could be a compromised shell with a prompt injection vulnerability. The agent could execute a transaction that the consumer never intended, then claim it was authorized because the passkey was used. The issuer sees the passkey and assumes the consumer approved. But the consumer may have only approved the agent, not the specific transaction.
This is not a theoretical risk. In my 2020 DeFi yield optimization work, I saw that automated strategies can execute dozens of trades within minutes. The human operator often does not review each one. If one of those trades was maliciously injected, the operator would have no way to prove it was unauthorized. The same logic applies here. The agent developer is the weakest link in the supply chain, and no certification covers them.
Contrarian: The 14% Trust Data Is the Real Signal
Everyone is looking at the 85+ partners and the 99% readiness. They are reading the Visa press release as a victory lap. I see a different picture.
The 14% trust figure is not a barrier. It is a launchpad. Early adopters in any technology plateau at 10-15%. The fact that 14% already trust AI agents without verification means the early majority is within reach. If the 2026 holiday season delivers a smooth experience, that number could jump to 25-30%. A single security incident, however, could push it back to single digits.
But the contrarian take is not about the trust number. It is about the assumption that Visa's certification is the key enabler. The real bottleneck is not the issuer's ability to process agent transactions. It is the consumer's ability to trust the agent itself. The passkey solves the identity of the consumer. It does not solve the identity of the agent. The agent is a piece of software running on a server or a device. Who audits that code? Who guarantees that the agent is not being manipulated by a malicious prompt?
Visa's certification is a necessary condition, but not a sufficient one. The market will need a KYA (Know Your Agent) standard. Someone must verify that the agent's decision logic is transparent, that it does not have hidden instructions, and that it respects the consumer's spending limits. Without that, the issuer is taking on the liability of the agent's actions without the ability to control them.
This is the same mistake I saw in the 2022 LUNA collapse. Everyone assumed the system was sound because the code was audited. But the code was not the problem. The liquidity crisis was a failure of trust in the mechanism. Here, the trust mechanism is incomplete. The certification covers the issuer, but the real risk is the agent. The smart contracts execute, they do not empathize. The agent executes, but not necessarily the consumer's intent.
Takeaway
The race for agentic payment is not about who certifies the issuer first. It is about who certifies the agent. Visa has built a solid foundation, but the house is missing a wall. The question is not whether millions will use AI agents to shop in 2026. The question is whether the first major agent fraud incident will happen before the industry builds a KYA standard. Audit the code, then audit the team, then sleep. But right now, we are only auditing the issuer. The agent remains unaccountable. Ledger lines don't lie, but the ledger only records what the agent tells it to record.