Total agentic commerce transaction value is projected to grow from $8 billion in 2026 to $3.5 trillion by 2031, according to Juniper Research. Cards are carrying almost all of that volume right now, and the same research also points to strong growth in account-to-account payments, which Juniper expects to surge 113% between 2025 and 2030, growing from $91.5 trillion to $195 trillion globally. It’s tempting to read that as A2A overtaking cards. A closer look at how agentic transactions actually have to work argues against picking a single winner at all: agentic commerce is shaping up to be a multi-rail story, with different rails suited to different jobs.
Visa and Mastercard have invested heavily in agentic commerce specifically to secure their position before the market matures, reporting from American Banker notes, building tokenized credentials and purchase-protection constructs specifically for agent-initiated transactions. That is not just a defensive land-grab: card networks are making genuine progress on the identity, delegation, and authorization primitives agentic commerce needs, in some respects faster than account-to-account schemes on traditional rails have managed. Nick Maynard, VP of research at Juniper, still expects the field to open up to other payment methods as agentic transaction volume grows, on the basis that consumers pay the way they want to pay, not the way any single merchant or bank prefers.
Juniper’s own research carries a warning for institutions treating cards as a settled default: failing to support local payment methods, specifically digital wallets and account-to-account payments, will limit how far the agentic commerce market can grow. That is a real signal to diversify rails, not a signal that account-to-account payments are destined to become the default for agent-initiated transactions.
Card rails were built for a human presenting a card at a point of sale, with two-factor checks designed around that moment; interchange and per-transaction fees also make sub-cent and low-cent payments uneconomic no matter how good the authentication gets. That is a genuine structural gap for agentic commerce’s highest-frequency use cases. But account-to-account payments on traditional rails are not a clean substitute. The account validation, compliance checks, and reconciliation processes built into existing A2A infrastructure run on timelines designed for human-initiated, relatively infrequent transfers, not the velocity of machine-to-machine micro-transactions. Cost is not the only constraint A2A has to solve; speed of verification is a separate one.
This is also where agentic commerce stops being a single-rail story. Whichever payment method an agent uses, the questions that matter are the same ones open finance infrastructure has had to answer for years: is this participant accredited to move money on someone else’s behalf, is that authority being monitored in real time, and if something goes wrong, where does the liability land.
Payments guru Jeremy Light argues the more interesting shift agentic payments enable isn't automating today's checkout, but a wave of continuous, high-frequency micro-transactions no human could realistically manage: a smart meter shopping across electricity suppliers every half hour and paying for usage as it goes, an eSIM picking the cheapest data tariff daily, a news app spending a few cents at a time across multiple publishers against a small daily budget, or royalties settling automatically each time a song, film, or image is used. He calls this "payment atomisation," an ever-growing volume of ever-smaller transactions that agentic commerce will only accelerate.
None of that works well on card rails, for the reasons above. But traditional A2A rails are not built for it either: instant, always-on micro-settlement at machine speed is a different problem from a monthly salary transfer or a one-off bill payment, and the account validation and reconciliation steps built into those rails were not designed to run at that velocity. Stablecoins have emerged as the payment method purpose-built for this layer: near-instant, programmable, API-native settlement with negligible marginal cost per transaction. That makes them the more natural candidate to carry agentic commerce’s atomised, high-frequency micro-payment volume, rather than account-to-account payments on existing rails.
A white paper from the Coalition for Financial Ecosystem Standards (CFES) found that approaches to agentic payments remain fragmented and regulatory guidance is still unclear, and argues that industry standards need to focus on verification, authority, and responsibility to close that gap. Sima Gandhi, senior advisor at FS Vector and CFES co-founder, put the current state bluntly: "We don't know today who is responsible for what." Her example is worth sitting with: today, if someone hands their credit card to a nanny for groceries and the nanny uses it in the Bahamas, liability is clear. If an agent is authorized to buy paper towels and instead books a trip, or simply overspends because it's chasing a deal, nobody has settled who's on the hook.
That is a verification, authority, and liability problem that sits above the rail question, not underneath it. It shows up on A2A payments, which move directly between accounts with none of the dispute and chargeback infrastructure built up around cards over decades. It shows up just as much on stablecoin rails, where settlement finality is often irreversible and reversal mechanisms barely exist. And it shows up on cards, where agent-initiated authority does not map cleanly onto liability frameworks built for a human swiping a card. Whichever rail, or combination of rails, ends up carrying agentic commerce’s volume, the institutions moving that money will need the same three things the CFES paper calls for: a way to verify who’s actually authorized to act, continuous visibility into what that authority is being used for, and clarity on where responsibility sits when it’s used badly.
The shift underway is not cards losing to A2A, or A2A losing to stablecoins. It is agentic commerce becoming multi-rail, with different payment methods carrying different jobs, and every one of those rails converging on the same missing infrastructure: standardized accreditation of who is allowed to move money on someone else’s behalf, risk intelligence that updates continuously rather than after the fact, and liability rules that resolve deterministically instead of case by case. Institutions building or buying into agentic payment infrastructure, on any rail, have a chance to get that accreditation, authority scoping, and liability allocation built in from the start, rather than retrofitting it later the way the card industry is still doing for standard two-factor authentication.
This is the same rail-agnostic trust model Invela was built to provide. Our open finance infrastructure started solving for accreditation, real-time risk monitoring, and liability allocation before agentic commerce made that same gap visible on cards and stablecoins, which is why Invela's infrastructure is extending across payment rails.
Invela is the infrastructure layer that makes open finance trustworthy - accrediting who's in the network, monitoring risk in real time, and ensuring liability lands in the right place. Open finance, covered.
-p-2000.jpg)
