A skeptical read of the agentic payments sector is circulating, and it lands at an awkward moment. The argument, laid out in a recent opinion piece, is that AI agents are paying each other in amounts that barely register — fractions of a cent moving between demos, test scripts and pilot programs — while the surrounding funding, conference circuit and product announcements behave as though machine commerce is already a functioning economy. The word being used is bubble, and it is hard to dismiss entirely.
What the Skeptics Are Actually Saying
The critique is not that AI agents will never transact. It is that the current volume is thin relative to the infrastructure being built on top of it. Payment protocols, agent wallets, merchant directories and treasury integrations are shipping faster than genuine commercial demand. When you look at actual on-chain agent payment activity, a large share consists of micro-transactions between agents built by the same teams that built the payment rails — the ecosystem buying from itself. That is the classic signature of an early bubble: supply-side enthusiasm outpacing demand-side adoption by a wide margin.
Why the Bubble Framing Is Partially Right
The evidence for overheating is not imaginary. Every major infrastructure player has announced something agent-related in the past quarter. Card networks are writing identity standards for machine payers, cloud providers are accepting stablecoins from autonomous software, and startups are raising on the thesis that machine commerce will dwarf human commerce. Some of this is strategic positioning rather than response to demand — companies announcing agent capabilities because not announcing them reads as falling behind. That dynamic, where announcements beget announcements, is how sectors inflate. The dollar volume of actual agent-settled commerce remains orders of magnitude below the valuations implied by the narrative.
Why the Bubble Framing Is Also Incomplete
Here the skeptics overreach. Railroads in the 1840s and fiber in the 1990s were bubbles that also built real infrastructure. The relevant question is not whether current agent payment volume justifies current investment — it plainly does not — but whether the rails being laid today are the ones commercial demand will use when it arrives. The enterprise entry point suggests something real is forming. Ripple, for instance, has begun embedding governed AI agents into its corporate treasury platform, automating liquidity management, fund movement and financial-system interaction, with human approval retained over every financial action. This is not a demo; it is treasury automation sold to institutions that pay for efficiency. Agents that manage corporate liquidity are adjacent to agents that settle in stablecoins, and the distance between them is shrinking.
What Would Falsify the Skeptics
The clean test is non-captive demand. If, over the next several quarters, agents built by parties with no stake in the payment infrastructure start paying merchants that also have no stake in it — independent developers buying GPU compute, data feeds or API access in stablecoins without a press release attached — then the rails have found product-market fit regardless of how frothy the funding environment got. If instead volume stays concentrated among ecosystem participants transacting with each other at sub-cent denominations while the announcement cadence continues, the bubble diagnosis hardens into fact. Watch the composition of transaction growth, not the growth rate itself. Captive volume can be printed almost indefinitely; organic volume cannot.
The honest position is uncomfortable for both camps: the agentic payments sector is almost certainly overfunded relative to its current usage, and the underlying capability — autonomous software settling value programmatically — is almost certainly durable. Bubbles and real technology are not mutually exclusive. They usually arrive together.