Agentic commerce, in which autonomous AI agents research, compare and purchase on a shopper’s behalf, ceased to be a demonstration in July 2026. At its Payments Forum in Paris, Visa announced that AI agents are now completing live purchases at independent merchants across Europe, backed by more than 30 issuing banks. These were not transactions staged inside a controlled Visa storefront but genuine agents operating on live merchant sites, selecting items and settling payment within limits their owners had defined. It is a significant payments milestone. The commercial development beneath it is a pricing one, and it is the part most coverage has passed over.
Key takeaways
- Visa’s Trusted Agent Protocol resolves agent identity: it establishes which AI agents a merchant should trust and admit.
- It does not govern what the agent buys once admitted. That is decided by price and machine-readable product data.
- A verified agent disregards branding, design and persuasion, ranking the products that qualify on value.
- Agentic shopping adoption is projected to rise from 19% to 46% of consumers by the end of 2026, which makes this a near-term pricing question rather than a future one.
What changed in July 2026, and why it matters
The detail that warrants more attention than it has received lies in the infrastructure. To admit those agents, Cloudflare and Akamai, the same edge networks that spent a decade learning to block bots, are now helping merchants whitelist verified buying agents instead. For years, defending a storefront meant keeping automated traffic out; agentic commerce quietly reverses the premise. The first genuine obstacle was never teaching agents how to buy. It was giving merchants a reliable means to distinguish the bots worth admitting from those worth blocking.
This is precisely what Visa’s Trusted Agent Protocol is built to provide. It gives a merchant a consistent signal of agent identity, a means of separating verified AI shoppers from unverified traffic while retaining control over how those agents reach its site and its products. Layered on top is passkey authentication tied to a specific cardholder, with explicit approval required before any payment completes, keeping a person in the loop at the precise moment money moves. Visa frames the effort as the approach it used to scale contactless: establish the standards, the infrastructure and the partners first, with trust designed in from the outset. If the analogy holds, adoption will appear slow until, quite suddenly, it does not.
The stage this has reached should not be overstated. The launch spans roughly 30 issuers, a pilot cohort rather than genuine coverage, and the passkey enrolment it relies on must reach ordinary cardholders before agent checkout becomes routine. Visa has also indicated that the same model will extend beyond consumer retail into commercial and B2B payments, where an agent handling routine procurement is arguably a more natural fit than one making discretionary consumer purchases. Early or not, the direction is set, and the preparation it demands does not wait for the curve to steepen.
Trust was the hard part. It was not the whole part.
Consider what identity, authentication and trust have in common. Each answers the same question: who is this agent, and should a merchant admit it. That was the genuinely difficult question, and resolving it is what turned agentic commerce from a demonstration into a live channel. But it stops at the threshold. The instant the agent is through the door, a different question takes over, one none of that infrastructure addresses: what does it actually buy? Answering that is where a payments story becomes a pricing one.
How do AI shopping agents choose what to buy?
Begin with the gap between how a person and an agent read the same storefront. A human experiences a page as it was designed to be experienced, led by imagery, layout and brand. In an interview with Fast Company Middle East, Visa’s Oliver Jenkyn argued that the web was built for human perception, not machine parsing, and that the visual craft of a product page, everything designed to move a shopper, is effectively invisible to an agent. The persuasion layer, the headline, the hero image, the well-placed review, now addresses an audience that cannot perceive it and would be unmoved if it could. What remains is a compact set of machine-readable facts: availability, specification, price.
The decision then proceeds in two stages, and it is worth keeping them distinct. The first is discovery. Agent selection is governed by data quality, by the machine-readable product feeds and structured attributes that determine whether a product appears in an agentic search at all. Thin or inconsistent data renders a retailer invisible before the contest begins, which is why, in the same interview, Jenkyn anticipates search engine optimization giving way to something closer to agent engine optimization, directed at agents rather than people. The second stage is selection. Clearing the discovery threshold merely places a product in contention. When several comparable options qualify, the agent must still choose, and it chooses on value, weighing price against signals such as reliability and fulfilment and avoiding merchants with a record of delays, cancellations or backorders. Price is not the only term in that equation, but it is the one a retailer can adjust in real time, and typically the one that separates two otherwise similar options.
The old playbook does not transfer
Much of modern retail marketing exists to create preference that survives a price gap. A trusted brand, a loyalty scheme, a familiar interface: each is designed to make a shopper choose one retailer over a marginally cheaper rival. An agent dismantles most of that apparatus. It holds no brand affinity, responds to no well-designed page, and is not the member a loyalty programme was built to retain. Jenkyn characterises loyalty as facing a reckoning in this environment, a shift he calls exciting and daunting in equal measure. The levers that once allowed a retailer to avoid competing on price are precisely those to which an agent is least responsive.
This does not mean the human disappears. For now, most people still want a say: only around one in ten consumers report being willing to let an agent act with full independence, so oversight and approval remain in the loop. But the trajectory is unambiguous, and it points toward a market in which more of the buying is delegated to something that answers to price and data rather than persuasion. Analysts already place agent-influenced spend at 10 to 20% of US e-commerce by 2030. The retailers that prepare their pricing for that buyer, rather than assume their brand will carry them, are the ones that will keep the sale.
Why agentic commerce lands on pricing
When the buyer is a machine, three qualities of your pricing move from the back office to the front line.
- Speed. An agent evaluates and purchases in the time a person spends reading a headline, so a price that is correct once a day is wrong for most of it. The agent acts on that discrepancy immediately, long before a weekly pricing review would register it, and does so across the entire catalogue at once.
- Structure. Because the agent reads data rather than design, prices and product attributes that are inconsistent or not machine-readable fall out of consideration, however strong the underlying offer. To an agent, being unreadable is indistinguishable from being unavailable.
- Defensibility. An agent has no loyalty to reward and no narrative to be swayed by, so a price it cannot rationalise gives it a ready reason to look elsewhere. A price that follows a clear and consistent logic is one it can justify selecting, and continue to select, purchase after purchase.
Where Quicklizard fits
Visa has built the layer that carries a verified agent through the door. What happens on the other side, whether that agent selects you or the retailer beside you, is a pricing decision made in milliseconds against machine-readable signals. That is the layer Quicklizard operates on: pricing that updates quickly enough to be correct at the moment the agent looks, structured cleanly enough to be read, and explainable enough to be defended. Quicklizard calls this Glass Box AI, pricing whose reasoning a retailer, and increasingly a machine, can actually follow.
Visa has done the demanding work of establishing which agents to trust. The harder commercial question is what those agents do once inside, and that is settled on price. Trust gets the agent in. Price closes the sale.
Frequently Asked Questions
What is the Trusted Agent Protocol?
It is Visa’s standard for agent identity. It gives merchants a consistent way to recognise verified AI shopping agents, separate them from unverified traffic, and control how they reach a site. It settles who the agent is, not what the agent buys.
How do AI shopping agents decide what to buy?
In two stages. First discovery, where machine-readable product feeds and structured attributes determine whether a product appears in an agentic search at all. Then selection, where the agent ranks the qualifying products on value, weighing price against factors like reliability and fulfilment. Branding and design carry little weight, because the agent does not read them.
What does agentic commerce mean for retail pricing?
Price becomes a live input to a machine’s decision rather than a signal aimed at a human. It has to update fast enough to be accurate when the agent checks, be clean enough to be read, and follow a logic consistent enough to give the agent a reason to choose it. With agentic shopping adoption projected to reach 46% of consumers by the end of 2026, it is a concern for pricing teams now, not later.
