The contest moves beneath the surface
Most coverage of agentic commerce fixates on the shopping agent, the assistant that reads an instruction like order trail running shoes under 150 dollars that arrive Friday and goes off to buy them. A Fortune piece published July 20 makes the case that the more consequential contest is happening a layer down, over the infrastructure that lets those agents actually move money and settle transactions. We find that framing useful because it redirects attention from the consumer facing demo to the plumbing. An agent that can compare products is impressive. An agent that can pay, authorize, and settle across merchants without a human in the loop requires an entirely new financial stack.
That stack is where the strategic positioning is now concentrated. A shopping agent needs a checkout protocol to place the order, an authorization mechanism to prove it is allowed to spend, and a settlement rail to move the actual funds. Each of those layers is being contested by a different set of players, from crypto native firms to the incumbent card networks to the cloud providers hosting the agents. The company that standardizes a layer captures a toll on a large and growing volume of automated transactions. That is why the infrastructure fight, largely invisible to shoppers, is drawing the heaviest strategic investment right now.
Circle's full stack ambition
Circle, the issuer of the USDC stablecoin, is making the most expansive claim. In an 18,500 word treatise titled The Agentic Economy, CEO Jeremy Allaire argued that the machinery of automated commerce demands financial primitives built for software. "An economy run by software agents must run on software money, software contracts, and software governance, or it cannot run at all," he wrote. Circle's pitch spans the whole stack, from its open source Arc blockchain infrastructure at the base to nanopayments for agents in the application layer, wrapping stablecoin settlement, compliance, and governance around them. It is an attempt to own the financial substrate of agentic commerce end to end.
We treat the treatise as marketing with a thesis worth taking seriously. Human payment rails assume a human clicks, waits, and reconciles. Agents transact at machine speed and volume, often in tiny amounts, across borders, at hours no human would. Traditional card settlement, with its fees, delays, and chargeback machinery, fits that pattern awkwardly. Whether stablecoins are the answer is genuinely unsettled, and Circle has an obvious interest in the conclusion. Still, the underlying observation holds. If agents become a meaningful share of transactions, the financial infrastructure beneath them will need properties that today's consumer payment systems were never designed to provide.
Why the rails question matters for retail
Retailers might be tempted to treat this as a payments industry problem and delegate it to finance. We think that is a mistake. The settlement layer determines the economics of every agent driven sale, the fees a merchant pays, the speed of settlement, the exposure to disputes, and the data the retailer receives about the transaction. If agent commerce routes through a rail with different cost and reconciliation characteristics than card networks, that flows straight into merchant margins and back office operations. A retailer that ignores the infrastructure choices being made now may find its agent channel economics dictated by protocols it never evaluated.
There is also a data dimension. Different protocols expose different information about who bought what and why. A settlement rail that strips context leaves the retailer with an order and little else, which compounds the relationship risk agentic commerce already introduces. The rails a retailer accepts, and the agent surfaces it supports, will shape how much it learns about its own customers. We would encourage technology leaders to bring merchandising, finance, and engineering into the same room on this, because the infrastructure decisions cut across all three, and the defaults are being set by vendors right now rather than by merchants.
The protocol stack taking shape
A rough map is emerging. Checkout protocols such as ACP and UCP handle the order placement call. Authorization frameworks like AP2 and TAP handle the signature that proves an agent is permitted to spend on a buyer's behalf. Then a settlement rail, whether a card network, Stripe, or an on chain path like x402, moves the money. In July the x402 Foundation launched operationally under the Linux Foundation with a completed Coinbase contribution and 40 members, a roster that includes Coinbase, Cloudflare, Stripe, Visa, Mastercard, AWS, and Google. That membership list is the tell. Nearly every major infrastructure player wants a seat at the table where these standards get written.
For a retailer, the practical implication is to avoid premature lock in. The standards are young and overlapping, and betting a catalog integration entirely on one protocol before the market consolidates carries real risk. At the same time, sitting out entirely means ceding presence on agent surfaces to competitors. We favor a hedged posture, supporting the protocols that already carry meaningful traffic while keeping integration abstracted enough to add or drop rails as the landscape shifts. The membership overlap across foundations suggests convergence is likely, yet the timing is uncertain, and building for optionality is cheaper than rebuilding after a bad bet.
What the incumbents are signaling
The card networks are not treating agentic commerce as a threat to be resisted. Visa CEO Ryan McInerney framed it as growth, predicting that agentic commerce will expand the network's addressable market and naming stablecoins and blockchain as areas showing significant opportunity. That posture matters, because Visa reaches more than 200 million merchants and has launched a stablecoin services platform of its own. When an incumbent with that footprint leans into a new settlement model rather than defending the status quo, it accelerates the whole shift. Mastercard has moved similarly, launching a protocol to let agents pay each other and handle micropayments.
We read the incumbents' embrace as the strongest evidence that this is more than crypto enthusiasm. The players with the most to lose from disruption are the ones building for it, which tends to pull the rest of the ecosystem along. For retailers, the reassuring part is that established, regulated networks are participating, which lowers the odds of being stranded on an orphaned standard. The cautionary part is that these same networks are positioning to remain the toll collectors in the agentic era, so the fee structures merchants hoped agents might disrupt may simply reappear in new form. Watch what the incumbents charge, not only what they announce.
The takeaway for retail technology leaders
The consumer story of agentic commerce, the assistant that shops for you, is the part that gets the headlines. The durable strategic story is the infrastructure being assembled underneath it, and that is where retail technology leaders should spend some attention this year. The choices being locked in now, about checkout protocols, authorization frameworks, and settlement rails, will govern the economics and data of agent driven sales for years. Treating those choices as someone else's concern means inheriting whatever defaults the infrastructure vendors settle on, with margin and customer data consequences a retailer did not get to weigh.
None of this demands a rushed commitment. It demands literacy and a watching brief. Know which protocols your agent partners use, understand how settlement affects your fees and reconciliation, and keep your integrations flexible enough to adapt as standards converge. Circle's treatise, the x402 Foundation's launch, and Visa's public enthusiasm are all signs that the money rails for agentic commerce are being poured right now. Retailers that understand that plumbing will negotiate their agent channel from a position of knowledge. Those that ignore it will accept terms written by others, and discover the cost only when the volume is large enough to matter.
