In September, BlackRock's digital assets research team published an eleven‑page paper titled “The Machine-Native Economy”. The trade press turned it into a headline: AI agents will drive the next wave of stablecoin demand. The paper itself is more cautious. Its cautious passages are the ones that matter most for companies.
What BlackRock Actually Writes
The thesis rests on a pairing. Artificial intelligence is machine-native intelligence, and digital assets are “machine-native money”, meaning money that machines can process without a detour. Once agents stop merely generating text and start buying data, services and computing power on their own, BlackRock argues, they need different payment rails. These have to run around the clock and settle very small amounts economically. Stablecoins are likely to lead transactional use, because their stable value provides a reliable unit of account. Among the protocols the paper names are x402 from Coinbase, the Agentic Commerce Protocol from Stripe and OpenAI, and Google's Agent Payments Protocol.
The paper itself contains three caveats. Agentic payment activity is still at an early stage. The only empirical evidence it cites, a study by the Bitcoin Policy Institute, is based on simulated model responses rather than observed agent behavior; BlackRock says so explicitly. And the disclaimer at the end makes clear that the material is meant to be neither a forecast nor research. Among the four authors are BlackRock's Head of Digital Assets and its Head of U.S. iShares Product Innovation. That does not make the analysis wrong, but it places it: this is the digital assets unit of an asset manager writing about its own market.
The Arithmetic Behind the Thesis
More robust than any forecast is an argument BlackRock only touches on: existing payment rails charge fees that make very small payments uneconomic. A public price list is enough to check this. In Germany, Stripe charges 1.5% of the amount plus €0.25 for a payment with a standard card issued in the European Economic Area (EEA). At €100, the fee is €1.75, or 1.75% of the amount. At €10 it is €0.40, or 4%, and at €1 already €0.265, or 26.5%. A payment of one cent costs €0.25015 in fees, about 25 times the amount.
At €100, the fixed component barely matters; below a few euros, it sets the price. An interface that wants to charge one cent per request cannot bill through cards. Coinbase's white paper on x402 is built on exactly this point: providers fall back on subscriptions and bundles, which are contracts a human has to sign. x402 is meant to close the gap by letting the agent pay for the individual request, without an account and without a contract. Card payments are therefore ruled out for micropayments not by principle, but by arithmetic.
The blockchain route is not free either. Network fees arise, and someone has to bear them; how x402 handles this is described in the knowledge hub article AI Agents & Programmable Money. The paper also argues from a U.S. perspective, comparing against cards and the Automated Clearing House (ACH). In Europe, the closer comparison would be instant credit transfers. They solve speed, but not the question of who initiates and books a transfer for a single request worth one cent. How many such payments actually flow between companies today, the paper does not say. A primary source with reliable figures on this is not yet available.
What This Means for a Machine Builder
An example makes the consequences tangible. A machine tool manufacturer offers its customers a forecasting service: from a machine's operating data, it calculates when a spindle will need maintenance. Today it sells the service as an annual license. If the customer's maintenance agent buys each individual forecast in the future, the manufacturer can bill per request, even for a few cents. Conversely, the design agent in its own engineering department could buy computing time for simulations per job. BlackRock describes exactly this scenario for compute.
For the finance department, what changes is less the paying than the booking. An annual invoice turns into thousands of payments a month, usually in a dollar stablecoin. Each one needs to be assigned to a customer or supplier, valued in euros at the time of payment, and backed by a supporting document. Whether a daily summary posting is permissible and how VAT applies should be settled before the first pilot, not after. The basics are covered in the knowledge hub under Posting, Valuation & Tax.
Where the Risks Lie
The first risk is currency. By BlackRock's own account, stablecoins are most commonly denominated in U.S. dollars. A company that reports in euros and receives thousands of small dollar amounts builds up a currency exposure that nobody deliberately took on. The second is authority. x402 does not regulate budgets, so the operator of the agent has to set the limit itself. The third is expectation: a thesis from the digital assets unit of an asset manager about demand for digital assets is a pointer, not a market finding.
The Real Question
BlackRock's paper explains why machines might need a different kind of money. Whether the companies selling to them can actually process these payments is a question it does not address. For mid‑sized companies, the order is therefore reversed: booking first, then the means of payment. The guiding question is which of its own services a company would sell per request if billing a single cent cost nothing. Whoever has an answer also knows the first process that would have to be rebuilt for it.