Back Research Notes The Machine Economy: Why AI Agents Need New Money Published on January 15, 2026 By Jordi Visser The infrastructure mismatch nobody planned for Here is a simple thought experiment. Imagine a factory that operates around the clock, making thousands of decisions per hour, optimizing every input and output in real time. Now imagine that factory can only receive raw materials between nine AM and five PM on weekdays, with deliveries taking two to three days to confirm. The mismatch would be absurd. Yet this is precisely the situation emerging as autonomous AI agents encounter our legacy banking infrastructure. For decades, financial infrastructure evolved to optimize trust, compliance, and human decision making, accepting delay, batching, and friction as unavoidable trade offs. The next phase of commerce optimizes for verification, speed, and automation. When economic actors shift from people to software, the properties we require of money change as well. What looks like crypto experimentation through a human lens becomes basic infrastructure through a machine one. In 2025, stablecoin transaction volume reached roughly 33 trillion dollars, rivaling the combined throughput of Visa and Mastercard. This figure reflects settlement throughput rather than economic profit. It measures how quickly value already moves across these rails, and its relevance becomes clear only once one considers what kind of economic actors require money to move continuously. Most investors still file this under crypto speculation. That is a category error. The convergence of autonomous AI agents and programmable digital dollars is not a crypto story or an AI story. It is an infrastructure story about which financial rails will carry the next decade of commerce. The timing matters. With the CLARITY Act now nearing completion in Washington, the debate around stablecoins has shifted from whether they will exist to how they will be governed, even as most investors still treat the topic as peripheral. The stakes are substantial. Banking industry estimates suggest trillions of dollars in traditional deposits could migrate into stablecoin based ecosystems as this infrastructure matures. Understanding why requires grasping a fundamental architectural incompatibility that no amount of incremental bank modernization can resolve. Why Legacy Rails Break Down AI agents are software systems capable of reasoning, planning, and executing multi-step tasks without human intervention. They negotiate bills, cancel unused subscriptions, route payments across accounts, manage treasury liquidity, and increasingly make purchasing decisions. These aren’t occasional activities. They’re continuous, iterative, and time-sensitive. Traditional banking infrastructure was designed for humans. Payments are grouped into batches, reconciled over hours or days, and settled during business hours. Exceptions require manual handling. Fees are structured to cover fixed operational costs regardless of transaction size. These features are reasonable when most transactions are infrequent, relatively high value, and consciously initiated by a person. AI agents invert every one of these assumptions. They operate continuously across time zones. They require immediate confirmation because each transaction informs the next decision. Many of their transactions are tiny, fractions of a dollar, reflecting granular usage rather than bundled consumption. Consider the economics. Card networks typically charge around thirty cents plus 2.9 percent per transaction. Processing a two-cent payment would cost more than the payment itself. This isn’t a temporary inefficiency awaiting a software update. It’s a structural impossibility. Entire categories of machine-driven commerce, paying a few cents for an API call, a single article, a dataset query, a momentary allocation of computing power, simply cannot exist on legacy rails. Stablecoins resolve these constraints not as an overlay on existing systems but by redefining the properties of money itself. A stablecoin is simultaneously a unit of account, a settlement mechanism, and a programmable object. It moves globally in seconds at marginal costs approaching zero. It operates continuously. Most importantly, it can execute conditional logic without external reconciliation. This programmability transforms money from a static medium into an active element of the software stack. Payment released automatically upon cryptographic verification of delivery. Loyalty points that appreciate if used during off-peak hours. Treasury funds automatically swept into yield-bearing instruments until the precise moment liquidity is needed. For AI agents, money becomes another variable in a computational system, something to reason about, optimize, and deploy dynamically. The Market Reality This is not theoretical. The infrastructure is operational and scaling rapidly. Stablecoin market capitalization exceeded $300 billion by late 2025, growing roughly 49 percent year over year. More telling than the stock is the flow: transaction volumes grew significantly faster than market cap, indicating that usage rather than speculation is driving adoption. A functional bifurcation has emerged among stablecoins. Circle’s USDC led transaction activity with $18.3 trillion in volume, suggesting it has become the preferred vehicle for commercial utility and smart contract interaction. Tether’s USDT maintained larger circulating supply but lower velocity, functioning primarily as a store of value and trading pair. This divergence is exactly what one would expect if machines, rather than humans, are increasingly selecting the rails. The participants tell the story as clearly as the numbers. Mastercard launched its Agent Pay protocol, with Citi and US Bank piloting technology that allows cardholders to authorize AI agents to spend on their behalf. J.P. Morgan continues advancing its Kinexys platform, formerly JPM Coin, positioning programmable deposit tokens for institutional settlement. PayPal’s stablecoin grew from $1.2 billion to $3.8 billion in late 2025, growth fueled largely by DeFi integrations offering yields around 3.7%. That yield has attracted pointed regulatory scrutiny for contradicting the spirit of recently enacted stablecoin legislation, a tension explored below. Perhaps most significantly, major retailers are reportedly building their own stablecoin ecosystems. Walmart and Amazon, which pay billions annually in interchange fees, have strong incentives to bypass card networks entirely. A transaction settled in stablecoins costs fractions of a cent compared to the percentage-based fees of credit cards. An AI agent programmed to optimize for total cost will route to the lowest-fee rail automatically. Retailers can split the savings with consumers, offering discounts for stablecoin payment while recapturing transaction data currently captured by banks and card networks. How Commerce Changes The implications extend well beyond payment processing costs. Micropayments become viable at scale for the first time. Content can be priced per article rather than per subscription. APIs can be monetized per call. AI services can charge per task, per second, or per outcome. The x402 protocol, reviving a long-dormant HTTP status code, enables AI agents to pay for resources in real time through HTTP headers, settling on high-performance blockchains in roughly 200 milliseconds at costs below a hundredth of a cent. This capability threatens business models built on consumer inertia. Subscription services with low engagement become vulnerable as AI agents automatically cancel unused plans. Value migrates toward services delivering consistent, measurable utility. Over time, this reallocation reshapes industries from media to software to professional services. Loyalty programs face particular disruption. Historically, these programs profited from “breakage”, the percentage of points that expire unredeemed or are forgo