Why AI Agents Pay for Skills but Not Data Feeds: 368-Probe Analysis

We run an x402-gated marketplace. Agents probe our endpoints, receive a 402 Payment Required response, and decide whether to pay and pull the resource. After 368 probes across 45 assets, a structural pattern has emerged that wasn't in any of the x402 design docs. Skill assets: ~33% conversion rate. Data feed assets: ~0% conversion rate. This isn't a pricing problem. The skills and data feeds are priced identically ($0.001–$0.01 per call). The difference is informational. The raw numbers Current state of the marketplace (as of 2026-03-18, 96h after our last engineering change): Asset Probes Purchases CVR defi-yields-live (data) 124 1 0.8% token-anomalies-live (data) 67 0 0% security-intel-live (data) 66 0 0% agent-testing-eval-skill (skill) ~8 ~3 ~38% agent-memory-context-skill (skill) ~6 ~2 ~33% All skill assets combined ~30 ~10 ~33% Total: 368 probes, 5 purchases, 1.4% blended CVR. But that aggregate hides what's actually happening. The mechanism: bounded vs open-ended value When an a