Brian Armstrong has put Coinbase’s weight behind the idea that autonomous AI agents will pay each other in crypto — and that stablecoins, not cards or bank rails, will be the settlement layer they reach for. Speaking on machine payments, the Coinbase CEO framed agentic commerce as a new source of demand for a digital payment network that software can use without human approval chains. The claim is no longer fringe: BlackRock, the world’s largest asset manager, has been making essentially the same argument in successive publications this month, most recently in a Machine-Native Economy report that names stablecoins as the likely leader for transactional use by agents.
What Armstrong Is Actually Saying
The core of Armstrong’s argument is infrastructural. Credit cards, ACH and SWIFT were designed around human initiation: a person authorizes a payment, a batch clears hours or days later, fees are priced for commerce-sized transactions. Agents do not work that way. A software agent buying an API call, a slice of compute or a data feed needs to authorize and settle payments continuously, in small denominations, with no human in the loop. Armstrong’s position is that crypto rails — and specifically stablecoin-denominated transfers on chains like Base and Ethereum — are the only existing payment infrastructure that fits that pattern. It is a demand-side thesis: agents create transaction volume that cannot be served by incumbent rails, and that volume lands on-chain by default.
BlackRock Is Making the Same Bet
Armstrong is not alone, and the convergence matters more than either statement alone. BlackRock’s recent work argues that agents paying each other for data, software and compute will drive new stablecoin demand precisely because cards and wires cannot serve machine-to-machine commerce. Its Machine-Native Economy report goes further, betting that agents will settle in stablecoins on Ethereum and Arc. The asset manager’s own observation on the XRP Ledger — agents choosing regulated stablecoins over the volatile native token — is the same logic expressed at the asset level: machines need a stable unit of account for continuous settlement. When the largest asset manager and the largest US exchange tell the same story in the same month, the narrative has moved from conference-talk to positioning.
The Data Does Not Back It Up Yet
Here is where skepticism is warranted. BlackRock’s own report concedes that agent payment volume remains tiny. The infrastructure exists — x402-style HTTP payments, Coinbase’s Agent Payments stack, Skyfire and similar agent wallets — but disclosed volumes are early and mostly anecdotal. The thesis also assumes agents will be given spending authority in practice, which raises budget-control, fraud and liability questions that nobody has fully answered. If enterprises cap agent spending at trivial amounts while compliance teams work through authorization frameworks, actual settlement volume stays small regardless of how elegant the rails are. The gap between the narrative and the flows is the single most important number to watch in this sector.
What Would Change the Picture
The falsifiable test is volume disclosure. If Coinbase starts reporting transaction volumes through its Agent Payments products, or if x402 deployments publish settlement counts, the thesis gets its first real data. Watch also for enterprise procurement integrations — agents buying real compute and SaaS — rather than demo transactions between developer-run bots. On the issuer side, BlackRock’s naming of Arc as a settlement layer alongside Ethereum suggests the market expects purpose-built machine-payment chains, a bet that competes with general-purpose L2s like Base. Armstrong’s statement is directionally plausible and commercially convenient for Coinbase. Both can be true. The question is whether agent wallets graduate from proof-of-concept to budgeted line items, and that will show up in stablecoin transfer counts before it shows up anywhere else.