AI agents are beginning to make payments through crypto protocols, but high transaction counts do not necessarily mean significant money is changing hands. Coinbase’s x402 protocol has settled about 200 million transactions since launch. In a 35-day sample, agents made 6.4 million transactions worth roughly $120,000, with 90.8% of individual payments below one cent. The figures offer a measure of how far machine-managed finance has progressed: the activity observed so far is dominated by tiny payments, not agents controlling large sums.
x402 sample: 6.4 million transactions, about $120,000
x402 is a protocol designed for payments by AI agents. Supporters often point to its transaction volume as evidence that machines can handle money without human intervention. German scientist and AI researcher Joscha Bach has argued that money can be understood as a pattern of causal relationships rather than something that must depend on paper or another physical form. If machines can process money, software could in principle move financial activity beyond direct human operation.
But transaction counts alone reveal little about the value transferred or how much autonomy agents have. In the 35-day sample, 6.4 million transactions added up to about $120,000, and the overwhelming majority were worth less than one cent each. The protocol therefore shows frequent activity, but generally at very low values per payment. Assessing the practical progress of agent-led payments requires looking not only at volume, but also at total funds moved, the tasks agents can perform and the scope of their decision-making authority.
Processing payments is not the same as setting goals
Bach’s comments address a broader question about the limits of automation: machines can handle funds and participate in market activity. The payment data, however, do not by themselves show that agents can make independent decisions. The products examined did not allow an AI agent to set its own goals. Executing a payment is distinct from deciding why to pay or what outcome to pursue.
The roughly 200 million cumulative transactions and the 6.4 million in the 35-day sample show substantial activity on x402, but they do not establish that AI agents can independently manage money. At the same time, the finding that 90.8% of sampled payments were below one cent indicates that current flows remain concentrated in very small transactions. For market participants tracking crypto payment infrastructure, transaction counts, dollar value and agent permissions are separate measures; volume alone is not evidence that machine-run finance is mature.